To determine precisely when the first election took place is to quibble about definitions and to place more faith in ancient sources than they deserve. So, instead of doing that, suffice it to say that by ~500 BCE both the Roman Republic and Athenian democracy were almost certainly holding elections.

The Athenians had an annual “who’s the most annoying person in town” contest. The “winner” had 10 days to pack their bags before they were banned from the city1 for a decade. Each voter scratched a name onto a pottery shard (ostrakon) to submit their vote, which is where we get the word ostracism. Archaeologists have found about 8,500 ballots from one ostracism vote around 471 BC, so we have an actual number, which is more than can be said for most ancient elections.

The largest election in the ancient world may have been in the late Roman Republic, around 70 BC. The franchise was broad—any male citizen could vote—but there was a catch: you had to show up in Rome on election day (and that day might get pushed back if the magistrates were getting bad vibes from the local birds). So while millions were eligible across Italy and beyond, actual turnout was likely on the order of tens of thousands. Precise numbers are debated, but 50,000 voters seems to be a reasonable estimate.

Then… history happened. The Roman Republic turned into the Roman Empire, the Empire fell, and large-scale elections mostly disappeared for the next millennium.

The next major vote took place in 1573 in the Polish–Lithuanian Commonwealth. When the last Jagiellonian king died without an heir, the commonwealth did something radical: it let every nobleman vote for the next king. About 40,000 szlachta rode to a field outside Warsaw and elected Henry of Valois, a French prince.

Henry did not stay long. He was elected in 1573, arrived in early 1574, and then—upon learning that his brother, the King of France, had died—quietly fled Poland in the middle of the night to claim the French throne. The Poles were undeterred and simply held another election and chose someone who actually wanted the job this time.

Elections began to scale in early modern Europe. Britain’s 1715 parliamentary election had on the order of a few hundred thousand voters.

In 1804, Napoleon held a plebiscite to approve his elevation to Emperor. He won with 99.93% of the vote, which tells you everything you need to know about how strictly we’re defining “elections” here. Elections in France grew over the century (both in number and actual democratic participation), culminating in the election of 1870 with about 9 million voters.

For most of the 20th century, the largest elections were those in Russia and the Soviet Union. The 1937 election took place during the Great Purge, and one party ran—the Communist Party—and won with 99.3% of the vote. Again, we’re being generous about what counts as an election here.

By 1984, the exercise had become pure choreography: 184 million participants, one pre-approved candidate per seat across each constituency, 99.99% turnout, 99.95% approval—all while the man nominally running the country, Konstantin Chernenko, was visibly dying of emphysema.

India is the current record holder, which in 2019 ran an election with about 615 million votes cast. It required seven phases, 39 days, 4 million voting machines, and a single polling station established in the Gir Forest for one voter because Indian law requires no voter travel more than two kilometers to cast a ballot. They broke their own record in 2024 with 642 million voters.

The funny thing about these is that, if you take all the largest known elections since the Dark Ages2 and plot them on a log scale, you can fit a straight line to it pretty well, showing that the record for the largest single vote has grown roughly exponentially over the past five centuries. That is, the number of voters participating in the largest ever election appears to double every 30 years.

If you project that line forward and take a recent projection of world population, you see that the trend line crosses the world population curve around the year 2150, at roughly 9.6 billion people. In other words, if voting records keep growing at the historical rate, a single election would involve every living human being sometime in the mid 22nd century.

Behold. The graph:

One world government by 2150. You heard it here first.

OK, maybe not a one-world government. But maybe something interesting? A global referendum? A planetary-scale plebiscite? A vote on whether AIs get voting rights so the trend can keep going?

Of course, this methodology is nonsense, but it’s nonsense with a graph and a line of best fit. There’s something satisfying about drawing a line through 500 years of data points and watching it hit a target.

But, if there really is a global vote in 2150, I want credit for calling it.

I thought some more about it and I think some of it holds up. But I also think I let myself off easy in a few places and didn’t give sufficient weight to some aspects of the argument, so here’s a response to myself.

It’s Not All About How People Use the Word

Your arguments about how people use the term “free will” are only semi-convincing. You leaned hard on common usage: people feel like they make choices, they report evaluating options, so free will is real in the sense that matters. Fine. That’s a valid argument (I mean, I did write it), but that proves less than you made it sound.

People can be wrong about what they’re experiencing. Someone who genuinely believes she’s psychic isn’t made psychic by the sincerity of the belief. We don’t say “well, that’s how she uses the word, so she really is seeing the future.” We say she’s having some experience (fooling herself into thinking she can predict the future) and mislabeling it (as really being psychic).

My point isn’t that free willers are making a falsifiable empirical claim the way the psychic is. It’s that sincerely experiencing something doesn’t settle what that something is. Yes, people have a feeling of control when they deliberate. Yes, they describe it using the language of free will. But maybe they’re doing exactly what the psychic is doing—having a real experience and attaching the wrong label to it. The feeling of choosing is real. Whether that feeling is free will, or just something that gets mistaken for it, is a question you skip when you focus on language usage.

You’re Not Taking Determinism Seriously

If people really grokked what it means to have a determined future, I don’t think they would say they have free will. Here’s an experiment: Take some young college freshman. He’s wide-eyed, idealistic, convinced his life could go in a thousand directions. Maybe he’ll start a company. Maybe he’ll move to Japan. Maybe he’ll drop out and write a novel. The world feels open to him, and that openness feels like freedom.

Now sit him down on his college dorm twin mattress and show him photographs from his future. Here’s your graduation—you stuck with accounting. Here’s your first job at Intuit. Here’s your wife, Sarah. Here are your two kids, Ethan and Ethel. Here’s your house in Plano, Texas. You’ll be mostly happy, by the way. It’s a good life.

Ask him how much free will he’s feeling.1

The feeling of open possibilities that excited him five minutes ago will have been replaced by something closer to watching a movie he happens to star in. He’s going through the motions of a life that was always going to happen. But his future is no more determined than it was five minutes ago. The only change is his level of ignorance about it.

Now, as we just saw, a real experience can get the wrong label. Just because the feeling of free will requires ignorance doesn’t mean that free will dissolves without it.

Fair enough, but if the feeling of freedom is completely independent of whether you’re actually free, then the compatibilist has to admit that the entire phenomenology of choice—the thing that makes free will matter to people—is essentially a byproduct of ignorance. You can keep calling the underlying causal process “free will” if you like, but you’ve just acknowledged that a lot of what people actually value about the experience is an illusion. That’s a strange place for a defense of free will to land.

So if you’re going to rely on people’s intuitions, you’ve got to admit that you’re relying on their intuitions while in a state of ignorance. The feeling of free will tracks with how much you know about your future actions, not with how free you are. Give someone total knowledge of their future and the feeling disappears, even though the causal structure of the universe hasn’t changed at all. I’m not saying the whole algorithm collapses without ignorance, but certainly the feeling of it does.

You Haven’t Closed the Door on Harsh Punishment

In your original piece, you treated the case against harsh punishment as largely settled—if no one has ultimate authorship, retributive justifications simply don’t hold. But this is not so. Even if we grant your premises, some moral frameworks can still justify harsh punishment without relying on retributivism at all.

For example, a utilitarian might argue that when someone murders a publicly beloved figure, the community’s need to see a severe response is itself a real psychological good. The justification here is forward-looking—perhaps people heal better and can move on when they feel justice has been done. The punishment looks retributive, but the reasoning isn’t.

Of course, satisfying this desire could still be net bad for society. Historically speaking, societies that have fed the public’s appetite for punishment through public executions and the like have generally seemed like worse places2 to live. So the argument might still ultimately fail on utilitarian grounds. But notice that now you’re having an empirical disagreement, so it’s clearly not a settled matter.

My point is only that the absence of ultimate authorship doesn’t close the door on harsh punishment under every moral framework. It eliminates retributive justifications, but that’s just one argument against harsh punishment, not the last word.

Where Is The Love?

In your previous piece, you spent a lot of time on hatred. If determinism is true, you argued, we don’t have to hate people for their worst acts. We can still hold them accountable, still protect society, still call behavior “wrong”, but there is no sense in hating anyone. You thought you were giving something up gracefully while keeping what mattered. It was a very tidy arrangement. But where I come from coins have two sides.

If a murderer doesn’t ultimately deserve your hatred because his cruelty traces back through an unbroken chain of genes, upbringing, and circumstance he never chose, then, by the same logic, a saint doesn’t ultimately deserve your love either. Not the deep kind. Not the kind where you look at someone and think, “What a genuinely good person they are, all the way down. They deserve good things.” They might be that good of a person, but how can they deserve love for it any more than an evil person deserves hatred?

You admire Gandhi. I admire Gandhi. He is among the most admired people who has ever lived. But by the logic you stated, Gandhi’s compassion and courage were products of a particular genetic hand and a particular environment, none of which he selected. Gandhi didn’t really do anything all that special, once you correct for his genes and environment. Anyone would have done it, even you. Maybe you deserve a statue in your honor. The plaque could read: “Had he had Gandhi’s genes and been placed in Gandhi’s situation, he, too, would have done great things.”

Perhaps you could still salvage a universal love, like a love for all conscious things or something. I’ll grant that, but that applies to serial killers as much as human rights leaders.

You said in your original piece that you can still prefer kind people, that you’re allowed to like being around someone who makes you laugh and treats you well. But if you call that “love” notice how thin it’s gotten. It’s the love of someone who is useful and pleasant to you, which is roughly the same way you love a good sofa. It’s not the love where you think another person deserves to flourish because of who they are. They’re a very nice sofa, but they didn’t upholster themselves.

This is, I think, considerably more unsettling than the punishment side, which is where everyone focuses. And understandably so—blame is where the stakes feel highest, where people go to prison or get forgiven. But we spend far more of our lives loving people than hating them, admiring people than condemning them. If determinism hollows out hatred, it hollows out love by exactly the same logic. You don’t get to keep one and discard the other, however much you might like to.

But that’s precisely what you tried to do. You softened blame while leaving love and admiration quietly in place. The honest move is to admit that the same wrecking ball swings both ways. Simply removing everything that constitutes blame causes collateral damage elsewhere.

What Are You Doing with Your Moral Agency?

I think there’s a tension within your claims that you haven’t addressed. On one hand, you say that people don’t deserve retributive punishment because they are not the ultimate authors of their actions. You say that punishment should be forward-looking, such as keeping dangerous people away from society. On the other hand, you say that people are moral agents.

So, what does moral agency actually do?

Consider a case where there is no forward-looking justification for punishment. Imagine a woman decides she doesn’t want her children anymore and kills them. She then gets irreversibly sterilized. She will never have children again, so her chance of reoffending is zero. No one else knows about the crime except a single police officer, so there is no deterrence value in publicly prosecuting her. She has no criminal history, mental illness, or other factors. She needs no treatment, no rehabilitation, no monitoring.

On a purely forward-looking account, there’s nothing to do here. And if you try to wriggle in some forward-looking justification we’ll just change the scenario so that it doesn’t apply. The point is, the forward-looking framework says let her go. And that answer feels monstrous.

Why? Because when the woman killed her children, she didn’t just break some procedural rule. She killed people who had moral standing. Those children were moral patients with welfare, futures, and the capacity to suffer, and her deliberative algorithm weighed all of that and chose to destroy them anyway. The moral reality of what she did doesn’t reduce to a forward-looking management problem.

This raises practical questions you haven’t answered. Is the woman’s calculation—weighing her children’s lives against her freedom and deciding her freedom was worth more—a moral claim that society must rebut? Should society impose a cost that outweighs the gain she sought? Should it tell her that she cannot trade another’s life for her convenience and come out ahead?

There’s a vast space between “you deserve to suffer for what you are” and “you’re just a system we need to manage.” You haven’t explored that middle space.

And the forward-looking framework fails for another reason. If punishment is justified entirely by the probability of future harm, then the only thing that matters is how well something predicts future harm, not whether the predictor itself is a crime. It might be the case that having committed a crime correlates more strongly with future harm than anything else, thus justifying some form of response for criminal activity. But this is an empirical claim, and we had better be prepared for it to go the other way. If it turns out that having a bad childhood or a face tattoo predicts future offending just as well as having committed a burglary, then a purely forward-looking system should treat them equally. Are you prepared to incarcerate people for bad childhoods?

You talked a lot about how having free will means you’re responsive to reasons, so you could say that moral agents are the kind of entities on which reasoning-based interventions work. You can explain to a person why they were wrong. You can use reason to deter them. You can’t do that with a bear. So perhaps moral agency tells you which tools are available, not whether someone deserves anything.

But this reduces moral responsibility to something almost entirely instrumental. It just tells us what type of intervention would be effective on a given system. Calling it “moral responsibility” feels empty. It would be more honest to call it “reasons based deterrence susceptibility”.

And notice what happens: we end up treating the woman who killed her children the same way we would treat a bear that mauled a hiker. We don’t blame the bear. We don’t sentence it for its crimes. We create a management plan for it. We put a dangerous bear down or relocate it for public safety. If we incarcerate a dangerous person only for public safety, then we’re treating them the same. In both cases the justification is forward-looking harm reduction. If there’s no difference in the response to a bear, which lacks moral agency, and a human, who has it, then moral agency seems to be little more than a farce.

When people say someone is “responsible” for a terrible act, they mean something beyond “this person’s cognitive architecture is amenable to deterrence.” They mean something backward-looking—this person should have done otherwise, and there’s some appropriateness to a negative response directed at them for what they did.

But you’re trying to preserve the language of moral agency and responsibility while draining it of the backward-looking content that normally gives it force. Without some backward-looking element, I’m not sure these terms have much meaning. When I blame you I’m not just trying to strategically intervene on your behavior. I’m making a claim about you as the agent you are. I’m saying: your algorithm had the capacity to weigh moral considerations, and it didn’t weigh them properly, and that failure is attributable to you in the proximate sense that the essay has spent thousands of words defending as a real and meaningful thing.

Guilt is the internal response of a moral agent recognizing their own failure. Indignation is the response of someone wronged by a being capable of having done otherwise in the Could₁3 sense. Forgiveness is real because there’s a genuine moral debt to release. These responses are partly backward-looking and irreducibly so. Without that backward-looking element, the entire fabric of the moral community dissolves into mutual behavioral management.

Is there a way to rescue yourself from this?

You’ve done a lot to establish moral agency. You’ve talked about the deliberative algorithm, about reflecting and revising oneself, about becoming an agent. You claimed free will grounds praise and blame. Let’s say we accept all that. Then when you act wrongly despite having the capacity to process moral reasons, something is true of you as an agent that is not true of the bear. There remains some form of desert—call it “agent-desert”—that falls directly out of proximate authorship. It says: you, the integrated deliberative system, produced this action through your own evaluative process, and that action was wrong, and that wrongness is attributable to you. Not to the Big Bang, not to your genes, but to the agent you’ve become.

Could this work? Is agent-desert what gives moral emotions meaning? Can it ground the whole ecosystem of responses, like guilt, indignation, and forgiveness? Not because the universe demands someone suffer, but because these are the correct responses between beings in moral relationships with each other.

So, I ask you again: what does moral agency actually do? Can it change a prison sentence? If it doesn’t ground some form of backward-looking appraisal—if it just tells us which management tools to use—then it’s not doing the work you claimed it does.

You need to admit that this is a backward-looking notion. Agent-desert says that what you did matters, that your relationship to your past action is morally significant, that the right response to a wrong isn’t just a management strategy but a recognition of what occurred between moral agents. The moment you accept that, you’ve conceded that purely forward-looking punishment was never adequate. A purely forward-looking framing is inadequate not because it’s too lenient, but because it’s the wrong kind of response to a moral agent.

You wanted moral agency to matter. You argued for it extensively. But then you left it with nothing to do. Agent-desert at least gives it something to do. Whether this is right or sufficient is still up in the air. But it shows that the purely forward-looking framework was never going to get you there.

Where Does This Leave Me?

I still think the core compatibilist insight is right—that deliberating and being responsive to reasoning is fundamental to free will. I think we need to preserve a meaningful distinction between a person who acts and a boulder that rolls. Those parts remain.

But I think I kept the parts of the moral landscape I liked and tried to surgically remove the parts I didn’t, as if the wrecking ball could be aimed. And I left moral agency standing at the center of my framework with no job to do, which is worse than not having it at all. An ornamental load-bearing wall.

I’m not sure if agent-desert is the fix. Does it all collapse into the retributive cruelty I was trying to escape? I’m not sure yet.

The book is a much-lauded New York Times bestseller for good reason. Sapolsky, a professor of neuroscience at Stanford, is an engaging and articulate writer, and he does a lot to make recent advances in neuroscience accessible.

The trouble comes when he attempts to add philosophy on top of it. He wants to demolish free will, and specifically to “take on” compatibilism, the position I defended in my previous post. Unfortunately, he doesn’t. He barely engages with it. Instead, he attacks an incoherent notion so bizarre it wouldn’t be free will even if it existed.

The Problem

Sapolsky commits his original sin, appropriately enough, at the origin. He tells us that he has written a book about free will, and explains the landscape of beliefs as follows (his italics, my bolding):

I’m going to be discussing some of the common attitudes held by people writing about free will. These come in four basic flavors:

The world is deterministic and there’s no free will. In this view, if the former is the case, the latter has to be as well; determinism and free will are not compatible. I am coming from this perspective of “hard incompatibilism.”

The world is deterministic and there is free will. These folks are emphatic that the world is made of stuff like atoms [...] this deterministic world is viewed as compatible with free will. This is roughly 90 percent of philosophers and legal scholars, and the book will most often be taking on these “compatibilists.”

The world is not deterministic; there’s no free will. This is an oddball view that everything important in the world runs on randomness, a supposed basis of free will. [...]

The world is not deterministic; there is free will. These are folks who believe, like I do, that a deterministic world is not compatible with free will—however, no problem, the world isn’t deterministic in their view, opening a door for free-will belief. These “libertarian incompatibilists” are a rarity, and I’ll only occasionally touch on their views.

Then he says this (his italics, my bolding):

What Do I Mean by Free Will?

People define free will differently. Many focus on agency, whether a person can control their actions, act with intent. Other definitions concern whether, when a behavior occurs, the person knows that there are alternatives available. Others are less concerned with what you do than with vetoing what you don’t want to do. Here’s my take.

Suppose that a man pulls the trigger of a gun. Mechanistically, the muscles in his index finger contracted because they were stimulated by a neuron having an action potential (i.e., being in a particularly excited state). That neuron in turn had its action potential because it was stimulated by the neuron just upstream. Which had its own action potential because of the next neuron upstream. And so on.

Here’s the challenge to a free willer: Find me the neuron that started this process in this man’s brain, the neuron that had an action potential for no reason, where no neuron spoke to it just before. Then show me that this neuron’s actions were not influenced by whether the man was tired, hungry, stressed, or in pain at the time. That nothing about this neuron’s function was altered by the sights, sounds, smells, and so on, experienced by the man in the previous minutes, nor by the levels of any hormones marinating his brain in the previous hours to days, nor whether he had experienced a life-changing event in recent months or years. And show me that this neuron’s supposedly freely willed functioning wasn’t affected by the man’s genes, or by the lifelong changes in regulation of those genes caused by experiences during his childhood. Nor by levels of hormones he was exposed to as a fetus, when that brain was being constructed. Nor by the centuries of history and ecology that shaped the invention of the culture in which he was raised. Show me a neuron being a causeless cause in this total sense.

Sapolsky is making the causal regress argument: trace any decision back through the neural chain, and you’ll find prior causes all the way down—neurons, hormones, genes, childhood, culture, and so on. His challenge is to find a break in this chain, a “causeless cause.”

But compatibilists don’t claim there’s a break in the chain. Compatibilists fully accept that decisions are caused by neural processes shaped by biology, environment, and history. That’s the whole point of compatibilism—free will is compatible with this.

So what do compatibilists actually mean by free will? In my post laying out the case for free will, I defined free will as the process of running a decision-making algorithm. This process gives us the feeling of free will and is the causal mechanism behind making choices. Unlike Sapolsky’s criterion of doing things for “no reason,” it responds to reasons. Yes, it’s influenced by whether you’re tired or hungry, but this doesn’t make it unfree. That’s it working properly. A decision that could be otherwise if your desires, reasoning, or circumstances were different is exactly the kind of decision compatibilists call free.

But the problem with Sapolsky’s definition isn’t just that it’s different from mine; it’s that it’s incoherent. It describes something that couldn’t exist and wouldn’t be free will even if it did. Consider what he’s actually asking for: a neuron that fires for no reason, influenced by nothing—not your environment, not your history, not your desires, not your reasoning. What would that even be?

If it’s uncorrelated with your past actions, then how is it your free will? Suppose your friends are all playing basketball and you want to play with them. On Sapolsky’s account, you can’t, because then your behavior (playing basketball) would be influenced by your experiences (your friends asking you to play). What kind of “free will” is this? Your “free” actions would have to be disconnected from everything you care about.

It wouldn’t let you interact with the world. Imagine a neuron that makes you say your own name, but since it can’t respond to your environment, it can’t fire because someone asked “What’s your name?” You’d blurt out your name at random, unable to respond appropriately to anything. This is not free will in any reasonable sense.

Sapolsky frames this as setting a high bar. But it’s not a high bar. It’s an incoherent and nonsensical bar. If his definitions were satisfied, if we found such causeless neurons, that wouldn’t look the slightest bit like free will. It would be random noise that happens to occur inside your skull. If we found such a neuron, it wouldn’t vindicate free will so much as be evidence of a brain malfunction.

This is why I say this isn’t just a semantic dispute. If Sapolsky simply defined free will differently from compatibilists, we could argue about whose definition better captures the concept. But you can’t have that argument when one side hasn’t described a coherent concept at all. Sapolsky could define “your achievement” as an outcome you had no role in causing, but there’s no productive debate to be had about whether that definition is too strict or too lenient. It’s just not what the word means.

Sloppy Engagement

Despite claiming to take on compatibilism, he repeatedly tries to disprove it by arguing for determinism. Early on, he says:

This version of compatibilism1 has produced numerous papers by philosophers and legal scholars concerning the relevance of neuroscience to free will. After reading lots of them, I’ve concluded that they usually boil down to three sentences:

1. Wow, there’ve been all these cool advances in neuroscience, all reinforcing the conclusion that ours is a deterministic world.

2. Some of those neuroscience findings challenge our notions of agency, moral responsibility, and deservedness so deeply that one must conclude that there is no free will.

3. Nah, it still exists.

Perhaps he thinks he’s arguing against compatibilism, but he’s not. Here he is, for example, attributing to compatibilists a view they don't hold:

For free-will believers, the crux of the issue is lack of predictability—at innumerable junctures in our lives, including highly consequential ones, we choose between X and not-X. And even a vastly knowledgeable observer could not have predicted every such choice.

[...]

Compatibilists and incompatibilists debate whether free will is possible in a deterministic world, but now you can skip the whole brouhaha because chaoticism supposedly shows that the world isn’t deterministic.

He specifically mentions compatibilists here, but then goes on to say this:

But now to the critical mistake running through all of this: determinism and predictability are very different things. Even if chaoticism is unpredictable, it is still deterministic.

Sure, chaotic systems are still deterministic, but how is that a refutation of compatibilism? Going back to his own definition, compatibilism is the belief that “The world is deterministic and there is free will.” How could more evidence of determinism be a refutation of compatibilism?

Also, note that predictability is not a crux. From my essay:

Free will doesn’t require unpredictability. If I offer you a choice between chocolate ice cream and a poke in the eye with a sharp stick, you’ll pick the ice cream every time. That predictability doesn’t mean you lack free will; it just means the algorithm reached an obvious conclusion. The question isn’t about whether the results were predictable, but whether the deliberative control process served as a guide versus being bypassed.

There are other examples of this (see the appendix for one more), but you get the idea.

Dismissiveness

Instead of engaging with compatibilism, he’s very dismissive of it. Near the end, he says:

One compatibilist philosopher after another reassuringly proclaims their belief in material, deterministic modernity…yet somehow, there is still room for free will. As might be kinda clear by now, I think that this doesn’t work (see chapters 1, 2, 3, 4, 5, 6…). I suspect that most of them know this as well. When you read between the lines, or sometimes even the lines themselves in their writing, a lot of these compatibilists are actually saying that there has to be free will because it would be a total downer otherwise, doing contortions to make an emotional stance seem like an intellectual one.

This is not even engaging with the arguments. For what it’s worth, I explicitly say I’m not using this as an argument in my piece:

The metaphysical question (does free will exist?) is separate from the sociological question (what happens if people believe it does or doesn’t?). Some argue for free will by saying belief in it leads to good outcomes (personal responsibility, motivation), or that disbelief leads to nihilism or fatalism. Sam [Harris] and I agree these arguments are irrelevant to whether free will actually exists. The truth of a claim is independent of the consequences of believing it.

Trying to Answer a Philosophical Question with Science

This is all very disappointing for a book purportedly about free will. I think where Sapolsky goes wrong is that he’s trying to answer a philosophical question with science alone. Science can certainly inform the question but it cannot settle it.

Look, I like science. Science can tell us a lot about the world. It can tell us the neural mechanisms behind decision-making in the brain, the timing of conscious awareness relative to neural activity (as in the famous Libet experiments), and how factors like brain lesions or physical trauma (e.g. Phineas Gage) affect behavior.

But science can’t tell us everything. Science tells us what the world is like, but it can’t tell us, given that world, which concepts make sense and how to apply them.

Consider a thermostat. Science can tell us every physical fact about it: how there’s a bimetallic strip and a circuit and so on. But it can’t tell us whether the thermostat is “making a decision.” That’s a conceptual question about what it means to “make a decision” and where we draw its boundaries. No additional measurement will resolve it. No scientist will ever find a belief, a self, or an iota of free will under a microscope. That’s the domain of philosophy.

The free will debate has exactly this structure. Sapolsky and compatibilists agree on the neuroscience. They disagree about whether what the brain does counts as “free will”. Does “free will” require freedom from the laws of physics? Or does it mean the ability to act according to one’s desires and reasons, even if those are physically caused? These are questions about how to understand agency, responsibility, and explanation in light of the science. They’re not questions that brain scans can settle.

Sapolsky writes as if piling up scientific facts settles the question. It doesn’t. We still have to think carefully about which concepts have earned their keep and which haven’t. We have to think about how we interpret the human experience in light of the data. And he simply refuses to consider such questions.

Conclusion

I worry this review has made the book seem worse than it is. There’s genuinely interesting neuroscience in it. If you’re skeptical of determinism, this is a good book to read and the science is mostly2 solid and often fascinating.

I should also note that Sapolsky and I are pushing in the same direction on the question of retributive punishment, which is arguably what matters most. He says:

And we need to accept the absurdity of hating any person for anything they’ve done; ultimately, that hatred is sadder than hating the sky for storming, hating the earth when it quakes, hating a virus because it’s good at getting into lung cells.

I’m with him on this point. You don’t need to deny free will to reject retribution, but if that’s where his argument leads people, I’ll take it.

I do wish he had actually engaged with compatibilism, the position he claimed to take on. The book promised such, yet delivered an attack on an incoherent strawman. Read it for the neuroscience. Just know that the confrontation with compatibilism he promises never quite arrives.

Appendix

You’ve got the idea by now. But if you’d like one more example of how he’s not talking about compatibilist free will:

Let’s frame this in the context of human behavior. It’s 1922, and you’re presented with a hundred young adults destined to live conventional lives. You’re told that in about forty years, one of the hundred is going to diverge from that picture, becoming impulsive and socially inappropriate to a criminal extent. Here are blood samples from each of those people, check them out. And there’s no way to predict which person is above chance levels.

It’s 2022. Same cohort with, again, one person destined to go off the rails forty years hence. Again, here are their blood samples. This time, this century, you use them to sequence everyone’s genome. You discover that one individual has a mutation in a gene called MAPT, which codes for something in the brain called the tau protein. And as a result, you can accurately predict that it will be that person, because by age sixty, he will be showing the symptoms of behavioral variant frontotemporal dementia.

Back to the 1922 cohort. The person in question has started shoplifting, threatening strangers, urinating in public. Why did he behave that way? Because he chose to do so.

Year 2022’s cohort, same unacceptable acts. Why will he have behaved that way? Because of a deterministic mutation in one gene.

According to the logic of the thinkers just quoted [He had quoted many different scientists and philosophers, whose views I do not know], the 1922 person’s behavior resulted from free will. Not “resulted from behavior we would erroneously attribute to free will.” It was free will. And in 2022, it is not free will. In this view, “free will” is what we call the biology that we don’t understand on a predictive level yet, and when we do understand it, it stops being free will. Not that it stops being mistaken for free will. It literally stops being. There is something wrong if an instance of free will exists only until there is a decrease in our ignorance. As the crucial point, our intuitions about free will certainly work that way, but free will itself can’t.

I can’t speak to what other people believe, but he’s simply not talking about compatibilists here. He might think he is, but he’s not.

Yet some argue it’s an illusion. One prominent opponent of the concept of free will is the author, podcaster, and philosopher . He has written a book on free will, spoken about it in countless public appearances, and devoted many podcast episodes to it. He has also engaged with defenders of free will, such as a lengthy back-and-forth and podcast interview with the philosopher Dan Dennett.

This essay is my attempt to convince Sam1 of free will in the compatibilist sense, the view that free will and determinism are compatible. Compatibilists like me hold that we can live in a deterministic universe, fully governed by the laws of physics, and still have a meaningful notion of free will.

In what follows, I’ll argue that this kind of free will is real: that deliberation is part of the causal pathway that produces action, not a post-hoc story we tell ourselves. Consciousness isn’t merely witnessing decisions made elsewhere, but is instead an active participant in the process. And while none of us chose the raw materials we started with, we can still become genuine agents: selves that reflect on their own values, reshape them over time, and act from reasons they endorse. My aim is to explore where Sam and I disagree and to offer an account of free will that is both scientifically grounded and faithful to what people ordinarily mean by the term.

A Pledge Before We Start

Before we get too deep, I want to take a pledge that I think everyone debating free will should take:

I acknowledge that I am entering a discussion of “free will” and I solemnly swear to do my best to ensure we do not talk past each other. In pursuit of that, I will not implicitly change the definition of “free will”. If I dispute a definition, I will own it and explicitly say, “I hereby dispute the definition”.

I say this, in part, to acknowledge that some of the difference is down to semantics, but also that there’s much more than that to explore. I’ll aim to be clear about when we are and are not arguing over definitions. In defining “free will”, I’ll start with the intuitive sense in which most people use the term and I’ll sharpen it later.

While we’re on definitions, we should also distinguish between two senses of “could”. Here are the two definitions:

• We’ll use Could₁ to mean “could have done otherwise if my reasons or circumstances were different”.

• We’ll use Could₂ to mean “could have done otherwise even if we rewound the universe to the exact same state—same atoms, same brain state, same everything—and replayed it”.

Here’s an example of each case:

• Sam often uses the example of choosing between coffee and tea. Let’s say you chose coffee this morning. If your doctor had told you that you need to cut out coffee, you could have chosen tea instead. That’s Could₁. If we rewound the universe back to how it was when you made your decision and replayed the tape, you could not have chosen otherwise, no matter how many times you tried. That would be Could₂. So in this case, you Could₁ but not Could₂ have chosen tea.

• Imagine instead that your choices are at least partially determined by quantum noise, and that this is a fundamentally random process. If you rewound the universe and replayed it, you really might have made a different choice. That’s Could₂. But notice: if quantum noise determined your choice yet no amount of reasoning could have changed it, you’d have Could₂ without Could₁—you could have done otherwise, but only by luck, not by thinking. That would be a strange notion of freedom.

Compatibilist “free will”, which is what I’m arguing for, is about Could₁, not Could₂.

Sam’s Position

Areas of Agreement

Let me start with a list of things Sam and I agree on. I know not everyone will agree on these points, but Sam and I do, so, fair warning, some of these I’m not going to discuss in detail. I’ve used direct quotes from Sam when possible. In other cases I’ve used my wording but I believe Sam would agree with it:

• Determinism: “Human thought and behavior are determined by prior states of the universe and its laws.” Humans and consciousness are fully governed by the laws of physics.

• No libertarian free will: Neither of us believes in libertarian2 free will, which is the idea that a person could (Could₂) have acted differently with all physical facts held constant.

• Randomness doesn’t help: The presence of randomness doesn’t create free will. If there is also some fundamental randomness in the universe (e.g. from quantum physics), that doesn’t rescue free will because you didn’t choose which random path to go down. That might give you Could₂, but it doesn’t give you Could₁, which, I believe, is what matters for free will.

• Souls don’t help: Even if people have souls, this probably doesn’t change anything because you likely didn’t choose your own soul.3

• Determinism does not mean or imply fatalism: We are both determinists, but not fatalists. It does not follow from “everything is determined” to “nothing you do matters”.

• No ultimate authorship: Ultimately, you did not choose to be you. You did not choose your genes, parents, childhood environment, and so on.

• Accountability should be forward-looking: “Holding people responsible for their past actions makes no sense apart from the effects that doing so will have on them and the rest of society in the future (e.g. deterrence, rehabilitation, keeping dangerous people off our streets).”

• Incarceration and contract enforcement still make sense: Nothing Sam or I believe suggests that, as Dan Dennett says about Sam’s position, “not only should the prisons be emptied, but no contract is valid, mortgages should be abolished, and we can never hold anybody to account for anything they do.”

• We must decouple two distinct questions of free will: The metaphysical question (does free will exist?) is separate from the sociological question (what happens if people believe it does or doesn’t?). Some argue for free will by saying belief in it leads to good outcomes (personal responsibility, motivation), or that disbelief leads to nihilism or fatalism. Sam and I agree these arguments are irrelevant to whether free will actually exists. The truth of a claim is independent of the consequences of believing it.

Sam’s Thought Experiment

Sam argues that we do not have free will.4 In the podcast, “Final Thoughts on Free Will” (transcript here), he provides an excellent thought experiment explaining his position. I quote sections of it below, but if you’re interested, I recommend listening to it in his voice. (I find it quite soothing.) Click here to jump right to the thought experiment and listen for the next nine and a half minutes. But in case you don’t want to do that, here’s what he says (truncated for brevity):

Think of a movie. It can be one you’ve seen or just one you know the name of; it doesn’t have to be good, it can be bad; whatever comes to mind, doesn’t matter. Pay attention to what this experience is like.

A few films have probably come to mind. Just pick one, and pay attention to what the experience of choosing is like. Now, the first thing to notice is that this is as free a choice as you are ever going to make in your life. You are completely free. You have all the films in the world to choose from, and you can pick any one you want.

[...]

What is it like to choose? What is it like to make this completely free choice?

[...]

Did you see any evidence for free will here? Because if it’s not here, it’s not anywhere. So we better be able to find it here. So, let’s look for it.

[...]

There are many other films whose names are well known to you—many of which you’ve seen but which didn’t occur to you to pick. For instance, you absolutely know that The Wizard of Oz is a film, but you just didn’t think of it.

[…]

Consider the few films that came to mind—in light of all the films that might have come to mind but didn’t—and ask yourself, ‘Were you free to choose that which did not occur to you to choose?’ As a matter of neurophysiology, your The Wizard of Oz circuits were not in play a few moments ago for reasons that you can’t possibly know and could not control. Based on the state of your brain, The Wizard of Oz was not an option even though you absolutely know about this film. If we could return your brain to the state it was in a moment ago and account for all the noise in the system—adding back any contributions of randomness, whatever they were—you would fail to think of The Wizard of Oz again, and again, and again until the end of time. Where is the freedom in that?

[…]

The thing to notice is that you as the conscious witness of your inner life are not making decisions. All you can do is witness decisions once they’re made.

[...]

I say, ‘Pick a film’, and there’s this moment before anything has changed for you. And, then the names of films begin percolating at the margins of consciousness, and you have no control over which appear. None. Really, none. Can you feel that? You can’t pick them before they pick themselves.

[…]

If you pay attention to how your thoughts arise and how decisions actually get made, you’ll see that there’s no evidence for free will.

Free Will As a Deliberative Algorithm

I wanted to see if I could write down my process of making a decision to see if I could “find the free will” in it. I wrote down the following algorithm. Note that it is not in any way The General Algorithm for Free Will^TM, but merely the process I noticed myself following for this specific task. Here’s what it felt like to me5:

1. Set a goal

   1. In this case, the goal is just “name a movie”.

2. Decide on a course of actions to reach the goal

   1. I realize I’ll need to remember some movies and select one. The selection criteria don’t matter that much.

3. Generate options

   1. To generate options, I simply instruct my memory to recall movies. I can also add extra instructions in my internal dialog to see if that triggers anything: “What about Halloween movies, aren’t there more of those? Oh, yeah, that reminds me, what about more Winona Ryder movies? I must know some more of those.”

4. Receive response

   1. The names of movies just pop into my head. More precisely, I should say they “become available to my consciousness” or “my consciousness becomes aware of them”.

5. Simulate and evaluate each option

   1. I hold candidates in working memory and simulate saying them. I reason about each option (will this achieve my goal? What are the pros/cons?) Then I evaluate each option and each returns a response like “yes, I can say this” / “no, this doesn’t achieve the goal” (maybe it’s a book and not actually a movie). It also returns some sense of how much I “like” the answer based on my utility function6. This is the thing that makes “Edward Scissorhands” feel like a better answer than “Transformers 4,” even though both are valid movies. Maybe I want to seem interesting, or I genuinely loved that film, or I have a thing for Winona Ryder. Whatever the reason, I get an additional response of “yes, that’s a good answer” or “eh, I can do better.”

6. Commit to a decision7

   1. I can reflect further on my choice. I hear Regis’ voice asking, “Is that your final answer?” Eventually, I tell myself that I am satisfied with my answer, and commit to it.

7. Say my answer

   1. I say it out loud (if I’m with others) or just say it to myself. Either way, I feel like I have made the decision.

8. Reflect on my choice

   1. I reflect on my decision. I feel ownership of my actions. I feel proud or embarrassed by my answer (“Did I really say that movie? In front of these people? Was that the best I could do?”).

So, where does this algorithm leave me? It leaves me with a vivid sense that “I chose X, but I could have chosen Y”. I can recall simulating the possibilities, and feel like I could have selected any of them (assuming they were all valid movies). In this case, when I say “could”, I’m using Could₁: I could (Could₁) have selected differently, had my reasons or preferences been different. It’s this sense of having the ability to act otherwise that makes me feel like I have free will, and it falls directly out of this algorithm.

This was simply the algorithm for selecting a movie, but this general structure can be expanded for more complex situations. The goal doesn’t have to be a response or some immediate need, but can include higher-order goals like maintaining a diet, self-improvement, or keeping promises. The evaluation phase would be significantly more elaborate for more complex tasks, such as thinking about constraints, effects on other people, whether there’s missing information, and so on. Even committing to a decision might require more steps. I might ask myself, “Was this just an impulse? Do I really want to do this?” And, importantly, I can evaluate the algorithm itself: “Do I need to change a step, or add a new step somewhere?”

In short, I’m saying free will is this control process, implemented in a physical brain, that integrates goals, reasons, desires, and so on. Some steps are conscious, some aren’t. What matters is that the system is actively working through reasons for action, not passively witnessing a foregone conclusion. (Perhaps there is already a difference in definition from Sam’s, but I want to put that aside for another moment to fully explain how I think about it, then we’ll get to semantics.)

So when someone asks, “Did you have free will in situation X?” translate it to: “Did your algorithm run?”

Constraints and Influences

Let me be clear about what I’m not claiming. My compatibilist free will doesn’t require:

Freedom from constraint. Sam points out that saying “Wizard of Oz” was not an option if I didn’t think of it at the time, even if I know about the film. This is true. But free will doesn’t mean you can select any movie, or any movie you’ve seen, or even any movie you’ve seen that you could remember if you thought longer. It just means that the algorithm ran. You had the free will to decide how much thought to put into this task, you had the free will to decide you had thought of enough options, and you had the free will to select one.

Consider a more extreme case: someone puts a gun to your head and demands your wallet. Do you have any free will in this situation? Your options are severely constrained—you could fight back, but I wouldn’t recommend it. However, you can still run the algorithm, so you have some diminished, yet non-zero amount of free will in this case. For legal and moral reasons, it would likely not be enough to be considered responsible for your actions (depending on the specific details, as this is a question of degree).

In these scenarios, you have constrained choices. Constraints come in many forms: physical laws (you can’t choose to fly), your subconscious (Wizard of Oz just didn’t come to mind), other people (the gunman), time, resources, and so on. None of these eliminates free will, because free will isn’t about having unlimited options; it’s about running the deliberative algorithm with whatever options you do have.

Freedom from influence. Sam gives many examples of how our decisions are shaped by things we’re unaware of, such as priming effects, childhood memories, and neurotransmitter levels. That’s fine. Free will is running the algorithm, not being immune to influence. Your algorithm incorporates these influences. It isn’t supposed to ignore them.

Perfect introspection. You don’t need complete understanding as to why certain movies popped into your head or why you weighed one option over another.

We have some level of introspection into what goes on inside our brains, though it’s certainly not perfect, or maybe even very good. We confabulate more than we’d like to admit and spend a lot of time rationalizing after the fact. But the question isn’t whether you can accurately report your reasoning; it’s whether reasoning occurred. The algorithm works even when you can’t fully explain your own preferences.

Complete unpredictability. Free will doesn’t require unpredictability. If I offer you a choice between chocolate ice cream and a poke in the eye with a sharp stick, you’ll pick the ice cream every time. That predictability doesn’t mean you lack free will; it just means the algorithm reached an obvious conclusion. The question isn’t about whether the results were predictable, but whether the deliberative control process served as a guide versus being bypassed.

I think these distinctions resolve many of the issues Sam brings up. To hear them, you can listen to the thought experiment 42 minutes into the podcast episode Making Sense of Free Will. If you have these clarifications in mind, you’ll find that his objections don’t threaten compatibilist free will after all. See “Responding to Another Sam Harris Thought Experiment” in the appendix for my walkthrough of that thought experiment.

Objections, Your Honor

Let’s address some likely objections to this algorithmic account of free will.

Exhibit A: Who Is This “I” Guy?

Much of this might sound circular—who is the “I” running the algorithm? The answer is that there’s no separate “I”. When I say “I instruct my memory to recall movies,” I mean that one part of my neural circuitry (the part involved in conscious intention) triggers another part (the part responsible for memory retrieval). There’s no homunculus, no little person inside doing the real deciding. The algorithm is me.

This is why I resist Sam’s framing. Sam says my Wizard of Oz circuits weren’t active “for reasons I can’t possibly know and could not control.” But those reasons are neurological—they’re part of me. When he says “your brain does something,” he treats this as evidence that you didn’t do it, as if you were separate from your brain, watching helplessly from the sidelines. But my brain doing it is me doing it. The deliberative algorithm running in my neurons is my free will. Or, to quote Eliezer Yudkowsky, thou art physics.

The algorithm involves both conscious and subconscious processes. Some steps happen outside awareness—like which movies pop into my head. But consciousness isn’t merely observing the process; it’s participating in it: setting goals, deciding on a course of action, evaluating options, vetoing bad ideas. I’m not positing a ghost in the machine. I’m saying the machine includes a component that does what we call “deliberation,” and that component is part of the integrated system that is me.

Exhibit B: So, it’s an illusion?

Someone might say, “Ok, you’ve shown how the feeling of free will falls out of a deterministic process. So you’ve shown it’s an illusion, right?”

No! The deliberative algorithm is not just a post-hoc narrative layered on top of decisions made elsewhere; it is the causal process that produces the decision. The subjective feeling of choosing corresponds to the real computational work that the system performs.

If conscious deliberation were merely a spectator narration, then changing what I consciously attend to and consider would not change what I do. But it does. If you provide new reasons for my conscious deliberation—“don’t choose My Little Pony or we’ll all laugh at you”—I might come up with a different result.8

It’s certainly possible to fool oneself into thinking you had more control than you actually did. I’ve already admitted that I don’t have full introspective access to why my mind does exactly what it does. But if this is an illusion, it would require that something other than the deliberative algorithm determines the choice, while consciousness merely rationalizes afterward. This is not so; the algorithm is the cause. Conscious evaluation, memory retrieval, and reasoning are not epiphenomenal but instead are the steps by which the decision is made.

Exhibit C: Did you choose your preferences?

Did I choose my preferences? Mostly no, but they are still my preferences. I’ll explore this more later, but, for now, I’m happy to concede that I mostly didn’t choose my taste in music, books, movies, or anything else. They were shaped by my genes, hormones, experiences, and countless other factors, none of which I selected from some prior vantage point. Puberty rewired my preferences without asking permission.

But this doesn’t threaten free will as I’ve defined it (we’ll get to semantics later, I promise). The algorithm takes preferences as inputs and works with them. It doesn’t require that you author those inputs from scratch.

The objection against identifying with my own preferences amounts to saying, “You didn’t choose to be you, therefore you have no free will.” But this sets an impossible standard. To choose your own preferences, you’d need some prior set of preferences to guide the selection, and then you’d need to have chosen those, and so on, forever. The demand is incoherent. What remains is the thing people actually care about: that your choices flow from your values, through your reasoning, to your actions. That’s free will. You can’t choose to be someone else, but you can choose what to do as the person you are.

Exhibit D: What about those Libet Experiments?

What about those neuroscience experiments that seem to show decisions being made before conscious awareness? Don’t these prove consciousness is just a passive witness?

The classic evidence here comes from Libet-style experiments (meta-analysis here), where brain activity (the “readiness potential”) appears before participants report awareness of their intention to move.9 These findings are interesting, but they don’t show that the entire deliberative algorithm I’ve described is epiphenomenal. When researchers detect early neural activity preceding simple motor decisions, they’re detecting initial neural commitments in a task with no real stakes and no reasoning required. This doesn’t bypass conscious evaluation, simply because there’s barely any evaluation to bypass.

In Sam’s movie example, the early “popping into consciousness” happens subconsciously, and I grant that. But the conscious evaluation, simulation, and selection that follows is still doing real computational work. The Libet experiments show consciousness isn’t the first step, but they don’t show it’s causally inert. To establish that, we would need to see complex decisions where people weigh evidence, consider consequences, and change their minds, being fully determined before any conscious evaluation occurs.10

There are also more dramatic demonstrations, like experiments where transcranial magnetic stimulation (TMS) activates the motor cortex opposite to the one a participant intended to use, forcing the “wrong” hand to move. When asked why they moved that hand, participants say things like “I just changed my mind.” I’ve actually talked about these studies before. I agree that they show that consciousness can invent explanations for actions it didn’t cause. But confabulation in artificial, forced-movement scenarios doesn’t prove that deliberation is always post-hoc rationalization. It proves we can be fooled when experimenters hijack the system.

Exhibit E: Aren’t You Just the Conscious Witness of Your Thoughts?

Sam has repeatedly referred to our conscious experience as a mere witness to our actions. In his book, he said (my bolding):

I generally start each day with a cup of coffee or tea—sometimes two. This morning, it was coffee (two). Why not tea? I am in no position to know. I wanted coffee more than I wanted tea today, and I was free to have what I wanted. Did I consciously choose coffee over tea? No. The choice was made for me by events in my brain that I, as the conscious witness of my thoughts and actions, could not inspect or influence. Could I have “changed my mind” and switched to tea before the coffee drinker in me could get his bearings? Yes, but this impulse would also have been the product of unconscious causes. Why didn’t it arise this morning? Why might it arise in the future? I cannot know. The intention to do one thing and not another does not originate in consciousness—rather, it appears in consciousness, as does any thought or impulse that might oppose it.

[...]

I, as the conscious witness of my experience, no more initiate events in my prefrontal cortex than I cause my heart to beat.

He’s made similar arguments in his podcasts, such as Final Thoughts on Free Will (jump to 1:16:06 and listen for 1.5 minutes). In that episode, he responds to compatibilist philosophy by arguing that what “you” experience as conscious control is just being a conscious witness riding on top of unconscious neural causes, and calling all of that “you” (as compatibilists do) is a “bait-and-switch”. That is, compatibilists start with “you” in the intuitive sense—the conscious self—but then expand it to include all the unconscious processes you never experience or control. By that sleight of hand, Sam argues, compatibilists can say “you” chose freely, but only because they’ve redefined “you” to mean something the ordinary person wouldn’t recognize. He concludes by saying, “The you that you take yourself to be isn’t in control of anything.”

I think this is a key crux of our disagreement. Sam sees consciousness as a mostly passive observer.11 I think it’s an active participant, a working component of the deliberative algorithm. Contrary to his claim, I think it can initiate events in your prefrontal cortex AND influence your heartbeat.

Here’s a simple demonstration: tell yourself to think about elephants for the next five seconds. Your conscious intention just shaped what happened in your prefrontal cortex. You don’t have complete control—it wouldn’t surprise me if a to-do list or a “did I turn off the stove?” trampled upon your elephantine pondering, but your conscious direction influenced events in your prefrontal cortex.

Of course, Sam would protest that the conscious intention to think about elephants arose from unconscious causes. This is true. But we need to distinguish origination (which I concede is unconscious) from governance. Even if the thought arose from the unconscious, it still went into the algorithm before you decided to act upon it. Therefore, you still had the ability to consciously deliberate, revise it if needed, or simply veto the whole idea.

I think Sam’s analogy to heartbeats actually backfires. He means to show that consciousness is as powerless over thought as it is over cardiac rhythm. But notice that you can influence your heartbeat: imagine a frightening scenario vividly enough and your heart rate will increase. You can’t stop your heart by willing it, but you can modulate it within a meaningful range.

I think this is a miniaturized version of a larger disagreement. Sam looks to the extremes and says, “You can’t choose what thoughts appear in your mind. You can’t stop your heart. You can’t inspect the rationale for your thoughts and actions. Looks bad for free will.” I look at the proximate areas and say, “You can choose to light up your elephant neural circuitry. You can choose to increase your heart rate. You can inspect the rationale for your thoughts and actions, albeit imperfectly. There’s plenty of free will here.” Your consciousness isn’t omnipotent, but it isn’t impotent either. It can modulate physiology, focus attention, and do real causal work while operating within constraints.

Sam is generally unimpressed with these sorts of claims. In his book, he quips: “Compatibilism amounts to nothing more than an assertion of the following creed: A puppet is free as long as he loves his strings.” But this gets the distinction backwards. A puppet would be unfree if the strings were pulled by an external controller, bypassing its algorithm. A person is free (in the compatibilist sense) when the “strings” are their own values, reasoning, and planning, and when the algorithm isn’t being bypassed but is the thing doing the pulling.

I understand where Sam is coming from. I’ve said before that sometimes our executive function seems more like the brain’s press secretary. But notice what a press secretary actually does. A pure figurehead would be someone who learns about decisions only after they’re final. A real press secretary sits in on the meetings, shapes messaging strategy, and sometimes pushes back on policy because of how it will play. The question isn’t whether consciousness has complete control, but whether it’s contributing in the room when decisions get made.

Confabulation research shows that we sometimes invent explanations after the fact. It doesn’t show that we always do, or that conscious reasoning never contributes. Again, the test is the counterfactual. You gave me a reason not to choose My Little Pony mid-deliberation, and it changed my decision. This means the conscious reasoning is doing real causal work, not just narration. That’s compatible with also sometimes confabulating. We’re imperfect reasoners, not mere witnesses.

Pathological Cases

Maybe a way to make the distinction between merely witnessing and being an active participant more clear is to talk about pathological cases. There are conditions where consciousness really does seem to be a mere witness, and, notably, we recognize them as pathologies:

• Alien hand syndrome—Here’s how the Cleveland Clinic describes alien hand syndrome: “Alien hand syndrome occurs when your hand or limb (arm) acts independently from other parts of your body. It can feel like your hand has a mind of its own. [...] With this condition, you aren’t in control of what your hand does. Your hand doesn’t respond to your direction and performs involuntary actions or movements.”
  Here’s an example from the Wikipedia page: “For example, one patient was observed putting a cigarette into her mouth with her intact, ‘controlled’ hand (her right, dominant hand), following which her left hand rose, grasped the cigarette, pulled it out of her mouth, and tossed it away before it could be lit by the right hand. The patient then surmised that ‘I guess “he” doesn’t want me to smoke that cigarette.’”

• Epileptic automatisms—Neuropsychologist Peter Fenwick defined it as follows: “An automatism is an involuntary piece of behaviour over which an individual has no control. The behaviour is usually inappropriate to the circumstances, and may be out of character for the individual. It can be complex, co-ordinated and apparently purposeful and directed, though lacking in judgment. Afterwards the individual may have no recollection or only a partial and confused memory for his actions.”

• Tourette syndrome—The paper Tourette Syndrome and Consciousness of Action says this: “Although the wish to move is perceived by the patient as involuntary, the decision to release the tic is often perceived by the patient as a voluntary capitulation to the subjective urge.”

• Schizophrenia—Here’s how one person with schizophrenia described an experience: “It is my hand and arm that move, and my fingers pick up the pen, but I don’t control them. What they do is nothing to do with me.”

How does any of this make sense if the non-pathological “you” is only a witness to actions? There would be no alien hand syndrome as it would all be alien. There could be no distinction between voluntary and involuntary behavior if it’s all involuntary to our consciousness. To me, these are all cases where consciousness isn’t able to play the active, deliberate role that it usually plays. What are these in Sam’s view?

Proximate vs Ultimate Authorship

A key distinction has been lurking in the background of this discussion, and it’s time to make it explicit: the difference between proximate and ultimate authorship of our actions.

Proximate authorship means your deliberative algorithm was the immediate cause of an action. The decision ran through your conscious evaluation process: you weighed your options, considered the consequences, selected a course of action, and, afterwards, felt like you could (Could₁) have selected otherwise. In this sense, you authored the choice.

Ultimate authorship would mean you are the ultimate cause of your actions. This would mean that, somehow, the causal chain traces back to you and stops there.

Sam and I agree that no one has ultimate authorship. The causal chain does not stop with you. You did not choose to be you. Your deliberative algorithm—the very thing I’m calling “free will”—was itself shaped mostly by factors outside your control:

• Your genes, which you didn’t select

• Your childhood environment, which you didn’t choose

• Your experiences, which are mostly a combination of events outside your control and the above, which you didn’t choose

This could go on and on. The causal chain stretches back through your parents, their parents, the evolution of the human brain, the formation of Earth, up to the Big Bang. As Carl Sagan put it, “to make an apple pie you must first invent the universe.” I have invented no universes; therefore, I have ultimate authorship over no apple pies (though I do have proximate authorship over many delicious ones, just for the record).

How to Make an Agent out of Clay

So how are we, without ultimate authorship, supposed to actually be anything? When does it make sense to think of ourselves as agents, with preferences we endorse, reasons we respond to, and a will of our own? In short, how do we become a “self”?

Earlier I said my preferences were my preferences in some meaningful way, but how can that be if I didn’t choose them? And even if I did choose them, I didn’t choose the process by which I chose them. And if, somehow, I chose that as well, we can just follow the chain back far enough and we’ll reach something unauthored. That regress is exactly why ultimate authorship is impossible, and I’ve already conceded it.

But notice what the regress argument assumes: it gives all the credit to ultimate authorship and none to proximate authorship. By that standard, nothing we ordinarily call control or choice would count.

Consider a company as an example. Let’s say I make a bunch of decisions for my company. I say we’re going to build Product A and not Product B, we’re going to market it this way and not that way, and so on. In any common usage of the words, I clearly made those decisions—they were under my control. But did I, by Sam’s ultimate authorship standard? Well, the reason I wanted to build Product A is because I thought it would sell well. And that would generate revenue. And that would make the company more valuable. But, did I make the decision to set the goal of the company to be making money? Well, I wasn’t a founder of the company, so it wasn’t my idea to make a for-profit company in the first place. Therefore, by the standard of ultimate authorship, I had no control and made no decisions! The founders made every one of them when they decided to found a for-profit company. This, of course, is not how we think about decision-making and control.

What matters for agency isn’t whether your starting point was self-created; it’s whether the system can govern itself from the inside, whether it can reflect on its results and revise its own motivations over time.

Humans can evaluate their own evaluations. I can have competing desires and reason through them. I can want a cigarette but also not want to want cigarettes, and that second-order stance can reshape the first over time. That’s the feedback loop inside the decision-making system. The algorithm doesn’t just output actions; it can also adjust the weights it uses to produce future actions.

Here’s a real example from my life: I believe I’ve successfully convinced myself that I like broccoli. Years ago, I made a conscious decision to tell myself I really liked broccoli. I didn’t hate it prior, but I wouldn’t have said I particularly enjoyed it. But I decided I’d be better off if I did, so I gathered all my anti-rationalist powers and told myself I enjoyed the taste. I ate it more often, and each time I told myself how much I was enjoying it. Within a couple of years, I realized I wasn’t pushing anymore. I just liked broccoli. Frozen broccoli, microwaved with salt, pepper, and a little lemon juice, is now my go-to snack. And it’s delicious.

Now, we don’t have ultimate authorship of either the first-order desire (disliking broccoli) or second-order desire (wanting to have a healthier diet), so who cares? But notice what happened here. This wasn’t just a parameter being adjusted in some optimization process. It was me deciding what kind of person I wanted to be and reshaping my preferences to match. That’s me authoring at the proximate level and taking ownership of the kind of person I’m becoming. The broccoli preference became mine not because I authored it from scratch, but because I consciously endorsed and cultivated it. It coheres with who I take myself to be.

This matters because I want to show that, over time, humans can become coherent agents. I want to show that humans are a distinct category from just a pile of reflexes or a mere conscious witness to one’s actions.

And this is why the regress to ultimate authorship doesn’t touch what matters. If “ownership” required self-creation, then no belief, value, or intention would ever count as yours either, because those, too, trace back to unchosen influences. But that’s not how we actually draw the line. We treat a preference as yours when it is integrated into your identity.

Note what this reveals about entities that can have free will. To reflect on your own desires you have to be able to represent them as your desires. You have to be able to take yourself as an object of evaluation and revise yourself over time. That requires a self-model robust enough to support second‑order evaluation: not just “I want X,” but “do I want to be the kind of person who wants X?”

You can see how the sense of agency develops in humans over time. It wouldn’t make sense to describe an infant as much of an agent. But over time, humans develop a sense of who they are and who they want to be. They can reflect on themselves and change accordingly. The algorithm can, to some degree, rewrite its own code in light of an identity it is actively shaping. This is another sense in which proximate authorship is “enough”. Not only can we run the algorithm, we can modify it.

That capacity for self-editing is a real boundary in nature. It separates agents from mere processes. A muscle spasm can’t reflect on itself. A craving can’t decide it would rather be a different kind of craving. But I can, and that’s the distinction that matters when we ask whether someone acted freely.

Sam’s entire objection seems to boil down to the assumption that control requires ultimate authorship. But this assumption doesn’t hold.

Disputing Definitions

OK, some of this has gotten into semantics, so, in keeping with my pledge: I hereby dispute the definition.

As we’ve seen, when I say “I have free will,” I don’t mean I’m the ultimate, uncaused source of my decisions, untouched by genes, environment, or prior events. I mean I have the capacity to translate reasons, values, and goals into actions in a way that is responsive to evidence. Or, in short, to run the algorithm.

So why call this “free will”?

First, you can see how the feeling of free will falls out of this algorithm. When people say they “could have done otherwise,” they are feeling their choice-making algorithm at work, and, as I’ve shown, that algorithm really is at work. The phenomenon matches the feeling of free will, so I say it’s appropriate to call it that.

Second, I think this definition matches how people talk in everyday life. Consider the following:

• “I could have made that shot.”

• “You could have studied harder.”

• “The defendant could have acted differently.”

• “The car could have gone faster.”

• “It could have rained.”

In all of these, “could have” means something like, “given the situation, a different outcome was within reach under slightly different conditions.”

For example, consider “I could have made that shot.” If I miss a half-court shot, I might say “I could have made that shot.” By that, I mean that, given my skill, if I tried again under similar conditions, it’s possible I could have made it. Making it is within my ability. If I try a full-court shot and the ball falls 20 feet short, then I probably just couldn’t have made it. I lack the physical capacity.

This is Could₁. It’s about alternative outcomes across nearby scenarios (e.g. I could have made that shot if the wind was a little bit different).

Contrast that with what the sentence would mean if people were using Could₂. The sentence would be, “I could have made that shot even if everything about the past and the laws of nature were the same.” It says that, rewinding every atom in the universe and every law of physics, things could (Could₂) have gone differently. This is a completely different claim and it’s not what people mean when they use the word.

This is not some complex claim that relies on consciousness. I’m talking about basic standard usage of the word “could”. Here’s another example: Which do people mean by “The car could go faster”? Do they mean:

1. The car could (Could₁) go faster had I pressed harder on the accelerator? Or,

2. The car could (Could₂) go faster even if the accelerator remained exactly how it was?

Could₁ is simply the standard usage of the term. In addition, it’s how it’s used in ordinary moral or legal discussions.

Take it or Leave it?

The term “free will” is what computer scientist Marvin Minsky would call a “suitcase phrase”—people pack different meanings into it and call it the same thing. There are some definitions of free will that Sam and I would both jump up and down and say, “No! That does not happen.” Mainly, the notion that if we were to reset the universe’s clock back 30 seconds and put every atom back in its place, someone could (Could₂) choose to act differently. But there are also some definitions of “free will”, like the feeling of weighing your options, reasoning your way to a conclusion, and acting based on that reasoning, where we should jump up and down and say, “Yes! That’s a real thing.”

Sam looks at the range of meanings people attach to “free will,” sees the metaphysical baggage, and concludes we’re better off abandoning the term. I look at the same thing and see most ordinary usage pointing toward something defensible. When someone says “I chose to stay late at work,” they’re not claiming to have escaped the causal order of the universe or exercised some quantum soul-power. They’re saying the deliberative algorithm ran: they considered leaving, weighed their reasons, and decided to stay. That’s Could₁, and it’s real.

Sam has an analogy for what he thinks compatibilists are doing. He compares it to claiming that Atlantis is real—it’s just the island of Sicily. Sure, it lacks the ancient advanced civilization, didn’t sink into the Atlantic, and isn’t “greater in extent than Libya and Asia”, but hey, it’s an island! Compatibilists, he suggests, are performing the same sleight of hand: pointing to something real but mislabeling it with a term that implies much more.

Sam’s analogy seems to imply that Could₂ is the defining feature of free will, and that I’ve discarded it while keeping the name. But I think this gets it backwards. As I said, when people say “I could have done otherwise,” they mostly mean Could₁. Admittedly, the free will I’m describing doesn’t deliver everything the term has ever been associated with. There’s no ultimate authorship, no metaphysical Could₂. But consider what people actually use free will for. They use it to distinguish choice from compulsion, to ground praise and blame, to make sense of deliberation. Could₁ does all of that and Could₂ does none of it. The features I’m preserving aren’t peripheral; they’re the load-bearing components. People want Could₁ from their free will and Sam is demanding Could₂.

I don’t understand why he seems to place so much importance in ultimate authorship. He seems to think that without it, “free will” names nothing worth preserving. But ultimate authorship was never part of how we actually explain human behavior. We’re billions of years into a cause-and-effect universe. When we ask “Why did she do that?” we don’t expect an answer that traces back to the initial conditions of the universe. We expect proximate causes—reasons, motives, deliberation.

Any time someone asks, “Why?” there is an unbroken chain of answers that could technically answer the question. There’s a sweet spot for good explanations for most questions, and it’s neither the ultimate cause nor the most proximate one, though it’s often much closer to the latter. Consider some examples:

Why did he lose the chess match?

Immediate (and useless) proximate cause: His king was checkmated. (Duh!)

Useful proximate cause: Because he left his rook undefended, lost it, and his position collapsed.

Ultimate cause: The Big Bang

Why did the team lose the football game?

Immediate (and useless) proximate cause: Because the other team scored more points. (Thanks, Dad. Wasn’t funny the first ten times.)

Useful proximate cause: They couldn’t stop the run.

Ultimate cause: Again, the Big Bang

The same applies to moral explanations. “Why did he betray his friend?” calls for an answer about motives, reasoning, and character, not about the initial conditions of the universe. We explain human action in terms of proximate causes because that’s the level at which deliberation, and therefore responsibility, operates. Ultimate authorship was never doing any work in these explanations. Letting it go costs us almost nothing we actually use.

Free will by any other name would smell as sweet

It’s worth stepping back and asking, “How does a philosophical concept like ‘free will’ gain metaphysical legitimacy anyway?” We’re not going to find it like we would a physical object. When I say “free will exists”, I’m not saying we’re going to see it in a brain scan.

This is why I say this isn’t just about disputing a definition. I’m making a stronger claim: My point is that any coherent account of agency, responsibility, and reasoning must posit something playing the free-will role. There must be some concept that distinguishes deliberated action from compulsion, reflex, or accident.

Without it, I think you’re forced into some strange positions. In his podcast Final Thoughts on Free Will, he treats someone being convinced by an argument as having the same freedom as being “pushed off a cliff and then claiming that I’m free to fall”. In the podcast Making Sense of Free Will, he makes no distinction between choosing orange juice and having a muscle spasm (see “Responding to Another Sam Harris Thought Experiment” in the appendix).

The distinction between being persuaded and being pushed, between choosing and spasming, isn’t some folk illusion we should discard in light of modern science. These are natural categories. It’s a distinction that carves reality at its joints. A spasm is an “open-loop” process: a signal fires, the muscle contracts, and no feedback mechanism checks whether this action serves your goals. Choosing juice is a “closed-loop” control system: an option is proposed, simulated against your preferences, evaluated, and executed only if it passes muster. These are fundamentally different mechanisms. One is responsive to reasons; the other isn’t. If you told me “the orange juice is poisoned,” I’d choose differently. If you told my leg “don’t jerk” while tapping my patellar tendon, it would jerk anyway.

This is what makes the choice mine in a way the spasm isn’t. The choice responds to what reasons mean to me. This is the difference between acting and being acted upon. Sure, both events are determined by prior causes, but to not see these as differences in kind seems, frankly, bizarre.

Or consider coercion. When someone holds a gun to your head, we say your choice was constrained. Yes, you “gave” them your wallet, but not freely. What makes this different from an unconstrained choice? It’s not that determinism was more true in the coercion case. It’s that your algorithm was given artificially narrowed options by an external agent.

Free Will as the Ontological Minimum

When I talk about free will, I’m not positing anything magical or spooky. Free will, as I’ve described it, exists the way beliefs exist. If you shrunk down in a Magic School Bus you wouldn’t find beliefs stored in labeled containers in the brain. But beliefs are real, right? They’re what we call a certain functional capacity of the brain. Free will is similar. It’s the name for what’s happening when a system weighs reasons, considers alternatives, and selects among them.

This is the minimal ontological commitment required to make sense of how we actually think about people. When we hold someone responsible, when we distinguish choice from compulsion, when we ask “why did you do that?”, we expect a reasons-based answer. Sam can call it something else if he likes. But he needs something to mark these distinctions, or his account of human action becomes incoherent. He simply has a free-will-shaped hole in his ontology.

I’m genuinely curious: from Sam’s perspective, do “beliefs”, “reasons”, “thinking”, and “agents” exist? We distinguish humans from thermostats by saying we respond to reasons while thermostats respond only to temperature. If reasons are real and can be causes of action, why not free will? It’s the same kind of thing, a higher-level description of what certain physical systems do, irreducible not because it’s made of magic, but because it captures patterns the lower level doesn’t.

Why It Matters: Phenomenology, Incentives, Morality, Law

Why does any of this matter? How does my defense of free will cash out in terms of things that we care about? I’ll list five reasons why this matters:

1. Phenomenology. People have a strong intuitive sense of free will. Where does this feeling come from, and does it track something real?

2. Incentives and behavior. Can people respond to rewards, punishments, and social pressure? How does free will relate to deterrence and rehabilitation?

3. Moral responsibility. Are people moral agents? Can they be held responsible for their actions?

4. Hatred and retributive punishment. Does anyone deserve to suffer for what they’ve done?

5. Crime and punishment. How should the legal system treat offenders?

Let me address each in turn.

Phenomenology: The Feeling of Free Will

We have a persistent feeling that we could have done otherwise. Is this feeling tracking something real, or is it an illusion? The answer depends on which “could” we mean. For Could₁, the sense that we would have chosen differently had our reasons, evidence, or preferences been different, yes, that’s completely real. But for Could₂, the sense that we might have chosen differently with every atom in the universe held fixed, no, that’s not real.

And, as I’ve argued, Could₁ is what the feeling of free will is actually about. This is what makes the algorithm-based account satisfying: it explains the phenomenology of free will without explaining it away. When you run through options, simulate outcomes, and select among them, you’re not passively watching a movie of yourself deciding. You’re experiencing your deliberative process at work. The feeling of choosing is the choosing. That’s what free will feels like from the inside, and that’s what free will is.

Incentives and Behavior

Here, Sam and I agree on the facts. People obviously respond to incentives. Stigmatizing drunk driving works. Offering bonuses improves performance. Punishment can deter crime. We shape behavior through incentives all the time.

I think Sam would argue that this doesn’t mean they have free will, just that their behavior responds to inputs. Fine, you could say that, but if you need a system that responds to reasons, weighs options, and updates based on consequences to explain human behavior, you’ve just described free will but are refusing to use the term. Incentives work because they feed directly into your deliberative algorithm. They change the weights, alter the utility calculations, and thus change behavior. This is why we can hold people accountable, offer rewards, impose consequences, and expect behavior change.

Moral Agency and Responsibility

I’ve claimed that we have proximate authorship but not ultimate authorship of our actions. Is this “enough” authorship for moral responsibility? I believe so. I believe being a moral agent is being the kind of entity whose decision-making can incorporate moral reasoning. This is a bit beyond the scope here, but the following are the types of things I would expect a moral agent to be able to do:

• Represent itself as a persisting individual (a self-model)

• Represent other entities as having welfare (i.e., as beings that can be benefited or harmed)

• Consider the effects of its actions on others, including tradeoffs between self-interest and moral reasons

• Have its own welfare at stake such that some outcomes can be better or worse for it

• Be responsive to reasons in the counterfactual sense: new reasons can change what it does (i.e., run the algorithm)

• Update its future behavior in light of reasons (e.g. criticism or reflection)

This is why we treat adults differently from infants, and humans differently from bears. It’s not that adults have ultimate authorship and infants don’t; it’s that adults have proximate authorship, and their algorithm can incorporate moral reasoning. A bear that mauls someone isn’t a moral agent. It doesn’t think, “How would I feel if someone did this to me?”

There are degrees here, of course. A four-year-old has more moral agency than an infant, and less than an adult. Someone with severe cognitive impairment may have diminished moral agency. The question is always: to what extent can this entity’s algorithm incorporate moral reasoning?

Moral Desert, Hatred, and Retributive Punishment

In addition to moral responsibility, there’s the question of desert, of whether wrongdoers deserve to suffer as retribution for their actions. Here, Sam and I completely agree that they do not. To deserve retribution in that deep sense, someone would need ultimate authorship of their actions.

To see why, consider an example Sam gives: someone commits violence because a brain tumor is pressing on their amygdala. We recognize them as a victim of neurology, not a monster deserving punishment. But now replace the tumor with an abusive childhood, genetic predispositions toward impulsivity, or serotonin imbalances. At each step, we’re still describing physical causes the person didn’t choose. The distinction between “tumor” and “bad genes” is arbitrary—both are prior causes outside the person’s control. It’s brain tumors all the way down. There but for the grace of God go I.12

Moral desert simply requires a metaphysical freedom that people do not have.

Once you give up ultimate authorship, a certain kind of hatred has to go with it. You can’t coherently hate someone as the ultimate author of their evil, as if they, from nothing, simply chose to be bad. That hatred requires the same metaphysical freedom that no one actually has.

Think about a bear that mauls someone. The bear causes harm, and we might kill it for public safety, but we don’t hate the bear. It’s not the kind of thing that could deserve retribution. The important part is recognizing that, without ultimate authorship, the same logic extends to humans. People who do terrible things are not deserving of suffering for its own sake. On this, Sam has been a tireless voice, and I appreciate his advocacy of this position.

This doesn’t eliminate all meanings of “hate” entirely, just a particular kind. You can still hate your job, Mondays, and git merge conflicts. You can definitely still hate dealing with git merge conflicts for your job on Mondays. But notice this is a different kind of hate. There’s no sense in which you want Monday to “pay for what it’s done.” It’s about anticipating that you’ll have a bad experience with it and seeking to avoid it.

The same applies to people. You can recognize that someone’s algorithm doesn’t adequately weigh others’ suffering, and you can avoid them accordingly. But there’s no need to view your enemies as self-created monsters deserving retributive punishment.

On this point, Sam wins. Perhaps if retributive justice were all I cared about, I would agree with him that we should consider free will an illusion. But free will does more work than that. It’s deliberation doing real causal work. It grounds the distinction between choice and compulsion, makes sense of why incentives change behavior, and gives meaning to praise and blame. Retributive punishment is the one piece that genuinely requires ultimate authorship, and it’s the one piece I’m happy to let go.

Crime and Punishment

What does this mean for crime and punishment? Does this mean we can’t hold anyone responsible? No. Sam and I are aligned here. We can hold people responsible without blaming them for ultimate authorship. We can and should hold people responsible in a forward-looking sense: for deterrence, rehabilitation, and public safety. Courts still need to distinguish intentional action from accident, choices made with a sound mind from those made under coercion or insanity. My account of free will provides exactly that framework: Did the algorithm run normally, or was it bypassed (reflex), broken (insanity), distorted (addiction), or given severely constrained options (coercion)?

Sam and I agree that sometimes we must incarcerate people because they are dangerous to others. But we do so to mitigate harm and deter future crime, not to exact retributive justice upon them.

Final Cruxes

Sam is right that there’s no ghost in the machine, (probably13) no soul pulling levers from outside the causal chain, no metaphysical Could₂ freedom. We agree more than we disagree. (In fact, Dan Dennett has called Sam “a compatibilist in everything but name!”) However, I wanted to compile what I see as the core cruxes of disagreement into a list. If Sam and I were to sit down and productively hash this out, here’s where I think we’d need to focus:

1. Is conscious deliberation causally efficacious, or is it epiphenomenal narration? I say the algorithm is the decision-making process—consciousness is doing real computational work. Sam says consciousness is merely “witnessing” decisions made elsewhere.

2. Is there a meaningful categorical difference between deliberated actions and reflexes? I say yes—one runs through the algorithm, one bypasses it. Sam seems to collapse this distinction since both are “caused”. But if there’s no difference between choosing orange juice and having a muscle spasm, something has gone wrong.

3. Is there a meaningful categorical difference between entities that can reflect on and revise their own decision-making versus those that cannot? A thermostat responds to temperature; a human can respond to reasons, evaluate their own preferences, and update their future behavior accordingly. I taught myself to like broccoli. I would like to see a thermostat do that. I can notice a bad habit and work to change it. This capacity for reflective self-modification seems like a real category that separates agents from mere processes. Does Sam recognize this as a meaningful distinction, or is this also collapsed because both are ultimately “caused”?

4. What should we think of pathologies where someone feels like a mere witness to their actions? To me, these seem like cases where the algorithm is damaged and consciousness isn’t able to play its active, deliberate role that it usually plays. I don’t know how Sam would describe these.

5. What lessons should we learn from the Libet-style experiments? Does it show that consciousness is post-hoc rationalization, or merely that consciousness isn’t the initiating step while still doing causal work downstream?

6. What should we think about an entity that has proximate authorship but not ultimate authorship (as all of us do)? Is that sufficient for moral responsibility, control, praise, and blame? Sam seems to think that without ultimate authorship, “control” is illusory. I think proximate authorship is sufficient, and that demanding ultimate authorship sets an impossible standard. The implication would be no one has ever controlled anything.

7. What counts as “you”? When Sam says “your brain did it,” he treats this as evidence against free will, almost as if “you” were separate from your brain. I say my brain doing it is me doing it. The deliberative algorithm running in my neurons is my free will. We may simply have different intuitions about where to draw the boundary of the self and whether being moved by your own values counts as freedom or puppetry. Similarly, should you identify with yourself? Should you take credit for the person you’ve become? Should we make anything of a person’s ability to become a more coherent agent over time versus a pile of unauthored behaviors? I say “yes”.

8. What criteria must a metaphysical concept meet to earn its place? If beliefs, reasons, and agents qualify, what test do these pass that free will uniquely fails? Does Sam reject it simply because of the historical “Could₂ baggage” associated with it? For me, a concept earns its keep by leading to and aligning with other natural categories, and doing without them requires tap-dancing around the concept.

9. What do ordinary people mean by “could have done otherwise”? I claim everyday usage is Could₁: “I would have acted differently if my reasons or circumstances had been different.” Sam seems to think people intuitively mean Could₂: “I could have acted differently with every atom in the universe held fixed.”

10. Is “free will” worth preserving as a concept, or should we retire it? Sam looks at the metaphysical baggage and says we’re better off abandoning the term. I look at what people actually use the concept for and say these are the load-bearing features. If we abandon the term, don’t we need something else to replace it? Doesn’t any coherent account of agency require something playing the free-will role?

I say let’s keep the term. Free will names something real: a process fully physical, fully determined by prior causes, and yet still you doing the choosing. The algorithm isn’t an illusion overlaid on “mere” physics. It is the physics, operating at a functional level that matters for morality, law, and human experience.

So, Sam, what would it take to convince you? If the algorithmic account captures what people mean by free will, does the work we need it to do, and doesn’t require any spooky metaphysics, what’s left to object to besides the name?

Appendix

Responding to Another Sam Harris Thought Experiment

I want to go over another thought experiment that Sam gives to show how all of his objections don’t threaten the notion of free will as I’ve described it. This is from the podcast Making Sense of Free Will. The thought experiment starts at 42 minutes in. The narrator makes a point that Sam has made many times, but it’s made clearly here, so I’ll use it. Here’s the setup:

Put yourself in a seat on an airplane. You’re a bit thirsty, and the beverage cart is making its way down the aisle to you. The flight attendant asks you what you’d like to drink. You see the choices on the cart: apple juice, orange juice, soda, water. You ponder things for a moment, make up your mind, and you ask for orange juice. After a few satisfying sips, you go for another and suddenly experience a muscle spasm in your arm. The movement causes some juice to spill on your neighbor’s pant leg.

The narrator (echoing Sam), argues that the selection of orange juice and the spilling of the juice aren’t as different as they seem. Yes, the spasm feels like something done to you. But did you really “choose” the orange juice? Did you create your preference for it? The narrator makes the case:

Maybe you had a flash of memory of your grandmother’s home. She had an orange tree in the backyard. Nostalgia is why you chose the orange juice over the apple juice. Subjectively speaking, does this really seem like an example of free will? Even the contents of that story are filled with things you didn’t choose, like your grandparents, where their house was, the fact that they had an orange tree, or the fact that your parents took you there when it was fruiting, and so on. And, in any case, as Sam points out, you can’t account for why this memory occurred to you in the very moment the flight attendant came by. Those neurons happened to be online and ready to fire at that moment. And, apparently, the neurons that could have fired that would have delivered the catchy slogan of your favorite apple juice advertisement and pushed you in that direction, didn’t fire. And, more importantly, you can’t account for why this grandmother story moved you to choose orange juice, rather than, say, be bored by orange juice because you had it so much as a kid.

This might sound compelling until you apply the algorithmic account. Then each objection dissolves:

• “The contents of that story are filled with things you didn’t choose.” Yes! You don’t have omnipotent control of the universe. There are constraints on your choice. That’s no barrier to free will.

• “You can’t account for why this memory occurred to you in the very moment the flight attendant came by.” True—we don’t have full introspection into our minds. But that’s OK. That information was fed into the algorithm. Had another memory popped in, it would have been incorporated instead, and might have influenced the outcome. The thoughts that did occur were sufficient to run the algorithm. Free will is sometimes constrained by your subconscious.

• “You can’t account for why this grandmother story moved you to choose orange juice.” I’d dispute this because you do have some degree of introspection, but it doesn’t matter either way. You don’t need full introspection for free will.

• “Why did you choose orange juice?” Because of my utility function, which was used in the evaluation step of my algorithm. The fact that this preference traces back to childhood memories I didn’t choose doesn’t change the fact that the algorithm ran. I don’t need ultimate authorship over my preferences. From a compatibilist perspective, this is no objection to free will.

The way these polls are reported is essentially guaranteed to produce a wild overestimate, which inevitably leads experts to write “how wrong Americans are” pieces, like this Brookings article claiming that “Americans believe foreign aid is in the range of 25 percent of the federal budget,” or KFF1 reporting that the “average perceived amount spent on foreign aid was 26%.”

But this isn’t just ignorance. The real problem is a failure of measurement and the statistics used to summarize it. The story isn’t “Americans are clueless” (though that may also be true), it’s “pollsters are using the wrong math.”

The Real Problem: Arithmetic Mean + Small Numbers

The problem is that pollsters ask for a percentage, then take the arithmetic mean to represent the data. For small true values, this approach is structurally doomed, and it has nothing to do with foreign aid specifically. It has to do with how we summarize guesses about small numbers.

When the true value is small, guesses are bounded at zero but unbounded above. That is, nobody can guess negative percentages, but anyone can guess 50% or 80%. On top of that, people tend to respond with round numbers like 5% or 20%, not decimals like 0.05% or 0.15%. This means that, even if there are many guesses around the true value of ~1%, there can only be outliers in the positive direction, so it results in a right-skewed distribution. If we choose the arithmetic mean as the average, it will be dragged upward by the right tail. A handful of overestimates skew the whole average.

This isn’t a sampling problem, and it won’t go away with more data. With more data, the arithmetic mean converges to the population arithmetic mean, but in a right-skewed distribution, that number is systematically higher than the median or geometric mean. A larger sample just gives you a more precise estimate of a misleading number. The “wisdom of the crowd” effect also will not fix this. That only works when the errors are independent and centered around the truth. If everyone is biased in the same direction, it won’t cancel out.

To see this in action, look at the histogram below. I simulated2 200 responses where most people guessed close to the true value of 1%, but some outliers guessed much higher (there’s a floor at zero, so they can’t guess much lower). The arithmetic mean comes out to 16.39%. But when you see the raw data, you can tell that this isn’t a good representation of the responses.

Now, I want to be clear about what I’m not arguing. I’m not claiming people actually know how much we spend on foreign aid. We systematically overestimate small, salient things, especially whatever’s been in the news lately. This is the availability bias, and it’s definitely a factor.

But there’s an additional problem here, which is my main point. People overestimate foreign aid, and these polling methods exaggerate how much they overestimate. In the KFF data, the majority of respondents said it was below 20%, and the largest decile by far was 0-10%. KFF said that the “average” was 26%, but their own data show that this isn’t what most people actually believe; it’s an artifact of using the wrong statistic.

This isn’t unique to foreign aid. You can produce the same distortion with any small number. If you’re itching to write a “here’s how dumb Americans are” piece, simply ask people to estimate something small, then take the arithmetic mean. What percentage of the federal budget goes to the National Endowment for the Arts? What percentage of high school students are pregnant? What share of deaths are from terrorism? In each case, the true number is small, guesses are bounded at zero but unbounded above, and a few outliers will drag the arithmetic mean into misleading territory.

Here’s a simple example. Say you want to write about how Americans overestimate the Scots-Irish population. Wikipedia says the percentage of Scots-Irish is 0.11%. You poll 50 people, and 49 of them just happen to be incredibly accurate and guess 0.1%, but then one person comes along and guesses 5%. The arithmetic mean would be 0.198%. Nearly double the true value.

You’ve got your headline: “Americans massively overestimate Scots-Irish population!” But is that really what your data showed? 49 out of 50 people were almost exactly right with a slight underestimation. One outlier dragged the arithmetic mean up, and if we’ve chosen the arithmetic mean to represent the average, now you’ve got a story about American overestimation.

There’s a psychological basis for this problem. Empirically, when you ask people to estimate quantities, like how many jelly beans are in a jar, the distribution of guesses tends to be lognormal. That means if you take the logarithm of everyone’s guesses, those values are roughly normally distributed.

This is because humans perceive the world proportionally, not linearly. The difference between 1% and 2% feels significant; the difference between 51% and 52% feels trivial, even though both are one percentage point. This is known as the Weber-Fechner law.

This matters for polling because it means errors are proportional rather than additive. Someone who overestimates by a factor of 5 (guessing 5% instead of 1%) is making the same kind of error as someone who guesses 50% instead of 10%. But when you take the arithmetic mean, that 5% guess gets treated as a modest overshoot while the 50% guess massively distorts the average. The use of the arithmetic mean doesn’t match the way people think.

The Fix: The Geometric Mean

If the underlying distribution is lognormal, one solution would be to have people estimate on a log scale, then transform back. But this is obviously impractical. No sane pollster is going to ask, “What is your estimate of the natural log of the percent spent on foreign aid?”

Fortunately, there’s a mathematically equivalent approach that doesn’t require confusing anyone with logarithms: the geometric mean.

The most common meaning of “average” is the arithmetic mean, where we add up all the values and divide by how many there are. But this isn’t the only way of averaging. Another is the geometric mean, where we multiply all the values together and take the nth root (where n is the number of values). Just like the arithmetic mean minimizes the squared error on a raw scale, the geometric mean minimizes squared error on a log scale. If you think people’s errors are multiplicative, the natural average is the geometric mean.3

The graph below shows the simulated foreign aid data again, this time with the geometric mean. Instead of 16.39%, we get 4.7%, which, based on visual inspection, seems to be a much better representation of the responses.

Let’s go back to that Scots-Irish example. 49 people guessed 0.1%, one person guessed 5%, and the arithmetic mean came out to 0.198%, nearly double what everyone except for one person said. The geometric mean is 0.1080%. If you’re going to try to represent the results with a single number, this is a much better representation of what people believed.

Same data, better summary statistic, and a sensible result.

Other Possible Solutions

The geometric mean is my preferred method in the case of polling for small numbers, but it’s not the only option. Here are some others that one might consider:

• Take the median. Report the middle value instead of the arithmetic mean. It’s far more robust to outliers.

• Remove the outliers. Drop the top and bottom X% of responses before averaging. This removes the most extreme outliers, though it requires choosing a cutoff, which introduces another degree of freedom, so you must be careful.

• Bin the responses. Instead of asking for a precise number, give people ranges: 0–1%, 1–5%, 5–10%, and so on. Then report something like “a plurality of respondents placed foreign aid in the 1–5% range.” This sacrifices precision but avoids the outlier problem entirely. It also introduces a degree of freedom in which bins to select.

• Force respondents to make trade-offs. For budget questions specifically, ask people to allocate 100% across all major categories—essentially, “draw us a pie chart.” This forces internal consistency. If you ask about each category separately, people’s estimates could add up to way more than 100%. If you make them allocate from a fixed budget, you’ll get more realistic answers. (Of course, this adds complexity, which always increases the chance you get bogus answers from people who just don’t want to be polled at the moment.)

Conclusion

I don’t think (most) pollsters are being malicious. The arithmetic mean is the default for a reason: it’s simple, it’s familiar, and in most contexts it works fine. The problem is that “most contexts” doesn’t include “estimating small percentages,” and nobody stopped to notice.

There are assumptions baked into everything we do with data. Taking the arithmetic mean of a bunch of numbers and calling it “the average” is a choice, one so automatic we often forget it’s a choice at all. Sometimes it’s the right choice. Sometimes it isn’t. Knowing the difference matters.45

When pollsters report that “the average perceived amount spent on foreign aid was 26%,” they’re not providing the most accurate representation of the responses. Even an incredibly well-informed population would produce an inflated arithmetic mean, simply because a few high guesses drag the number up.

That’s how these things usually go. A reasonable choice in one context becomes an unexamined habit. The habit produces nonsense. The nonsense becomes a talking point: “Americans think 26% of the budget goes to foreign aid!” No, they don’t. Some people guessed high, and the poor statistical choices of the pollster did the rest.

Predictions are the currency of a rational mind. They force us to articulate our assumptions, unpack the commitments of our beliefs, and subject our beliefs to the crucible of reality.

Committing to a prediction is a declaration: “Here’s what I actually believe about the world, and here’s how confident I am. Let’s find out if I’m right.” That willingness to be wrong in measurable ways is what separates genuine truth-seeking from specious rationalization.

This essay is an invitation to put on prediction-tinted glasses and see beliefs as tools for anticipating experience. The habit applies far beyond science. When you think predictively, vague assertions become testable claims and disagreements become measurable. Whether you’re evaluating a friend’s business idea, judging whether a policy will work, or just trying to settle an argument at dinner, you’re already making implicit predictions. Making them explicit sharpens your thinking by forcing your beliefs to confront reality.

Turning Beliefs into Predictions

For many people, beliefs are things to identify with, defend, and hold onto. They’re part of who we are, our tribe, our worldview, our sense of self. But in a rationalist framework, beliefs serve a different function: they’re tools for modeling reality.

From this perspective, a belief leads naturally to a prediction. It tells you what future (or otherwise unknown) experience to expect. Rationalists call this “making beliefs pay rent.”

A prediction concretizes a belief by answering: If this is true, what will I actually experience? What will I see, hear, measure, or observe? Once you try to answer that, your thinking becomes specific. What was once comfortably abstract becomes a concrete commitment about what the world will show you.

Predictions Make You Accountable to Reality

Many of the statements we encounter daily in the media, in academic papers, and in casual conversation, don’t come with clear predictive power. They sound important, even profound, but when we try to pin them down to testable claims about the world, they become surprisingly slippery.

The key to avoiding this is to make specific predictions, and even to bet on your ideas. Economist Alex Tabarrok quipped: “A bet is a tax on bullshit.” Most importantly (at least in my opinion), it’s a tax on your own bullshit. When you have to stake something on your prediction, even if it’s just your credibility, you become clearer about what you actually believe and how confident you really are.

Tabarrok used the example of Nate Silver’s 2012 election model. Silver’s model gave Obama roughly a 3-to-11 chance of winning against Romney, and his critics insisted he was wrong. Silver’s response was simple: “Wanna bet?“ He wasn’t just talking, he was willing to put money on it. And, crucially, he would have taken either side at the right odds. He’d bet on Obama at even money, but he’d also bet on Romney if someone offered him better than 3-to-1. That’s what it means to actually believe your model: you’re not rooting for an outcome, you’re betting on your beliefs.

Constantly subjecting our predictions to reality is important because, as physicist Richard Feynman said: “It doesn’t matter how beautiful your theory is, it doesn’t matter how smart you are. If it doesn’t agree with experiment, it’s wrong.” Reality is the ultimate arbiter. Not who argues more eloquently, not who has more credentials, but whose predictions better match what actually happens.

Predictions in Action

Consider the 19th-century debate over what caused the infectious diseases that were plaguing Europe, such as cholera, malaria, and tuberculosis. The dominant hypothesis at the time was miasma theory, the idea that diseases were caused by “bad air” or noxious vapors from rotting organic matter. The competing hypothesis was germ theory, the idea that diseases were caused by microscopic organisms.

Each hypothesis generates different predictions. If you believe in germ theory, you expect that examining blood or tissue from a sick person under a microscope will reveal tiny organisms. You expect that preventing these organisms from spreading (through sterilization, handwashing, etc.), will reduce disease rates. You expect that introducing these organisms to a healthy subject will cause illness.

If you believe in miasma theory, you make different predictions: Diseases should cluster around areas with foul-smelling air—swamps, sewers, garbage heaps. Improving ventilation and eliminating bad odors should reduce disease. And looking under a microscope shouldn’t reveal anything particularly relevant, since the problem is airborne vapors, not tiny creatures.

These hypotheses do overlap in some predictions. For example, both theories predict that interventions such as disposing of sewage and drinking clean water would reduce the spread of disease. This is common when comparing hypotheses, and it makes them harder to separate. But you can still test them by focusing on where they diverge: What happens when you sterilize surgical instruments but don’t improve the smell? What do you see under a microscope?

These hypotheses differed in what experiences they anticipated from the world. And the world answered. Microscopes, sterilization, and epidemiological data validated germ theory.

Now contrast this with a belief like “everything happens for a reason.” What experience does it lead you to anticipate? If something bad happens, there’s a reason—but what does that mean you should observe? If something good happens, there’s also a reason—but, again, what should you experience? If nothing happens, that too has a reason. The belief is compatible with any possible observation, which means it doesn’t constrain your expectations about what will happen in the world. It doesn’t actually predict anything at all. It doesn’t pay rent.

Predicting the Past

It’s tough to make predictions, especially about the future - Yogi Berra

Predictions are often associated with forecasting future events, but they encompass far more than that. Instead of thinking about what has happened versus what hasn’t, think about it from a first-person perspective: in terms of what information you know versus what you don’t. The act of prediction is fundamentally about inferring information unknown to you from what you already know, a process that applies equally to the past and present as it does to the future. Whether you’re deducing what happened thousands of years ago, inferring what exists in places you’ve never visited, or anticipating what will occur tomorrow, you’re engaging in prediction whenever you use available evidence to form expectations about information you don’t currently have.

Consider the debate around the evolution of human bipedalism, an event decidedly in the past. Several hypotheses have been proposed to explain why bipedalism evolved, but let’s look at just two of them.2

The “savannah hypothesis” suggests that bipedalism arose as an adaptation to living in open grasslands. According to this idea, walking upright allowed early hominins to see over tall grasses and spot predators or prey from a distance.

The “carrying hypothesis” proposes that bipedalism evolved to free the hands for carrying objects, such as food and infants, and, later on, tools. This would have provided a significant advantage in terms of transporting resources and infant care.

You might think that because it happened so long ago there’s no way to find out, but each of these hypotheses generates different predictions about what we should expect to find in the fossil record. The savannah hypothesis predicts that bipedal humans first appeared in, well, the savannah. So we should expect to find the oldest bipedal fossils in what was savannah at the time3. The first bipedal fossils may also coincide with a shift in climate towards more arid conditions which created more savannahs.

The carrying hypothesis predicts that bipedal species will predate the regular use of tools, as the hands were initially freed for carrying rather than tool use. It also suggests that early bipeds will show skeletal adaptations for efficient walking, such as a shorter, wider pelvis and a more arched foot.

New fossil discoveries have allowed researchers to test these predictions against the evidence. The famous “Lucy” fossil, a specimen of an early hominin species, was found in a wooded environment rather than an open savanna. This challenged the savannah hypothesis and suggested that bipedalism evolved in a more diverse range of habitats. Furthermore, the earliest known stone tools date to around 2.6 million years ago, long after the appearance of bipedal hominins. This also aligns with the predictions of the carrying hypothesis.

The point here is that competing hypotheses generate different predictions about what we should observe in the world. And this isn’t restricted to the future. We can “predict” what we’ll find in the fossil record, what a telescope will reveal about distant galaxies, or what an archaeological dig will uncover, even though the events themselves happened long ago.

You might think, well, that works for scientists but not the rest of us. I disagree. When you wear prediction-tinted glasses, you see that predictions are all around us. Even when people don’t think they’re making predictions, they often are. Positive statements—those telling us what’s true about the world—often make predictions.

If there is no prediction, there is no question

These same principles apply outside of science as well. Let’s see how this predictive thinking can apply to philosophy. Consider this example from Rationalist author Eliezer Yudkowsky regarding one of philosophy’s most famous4 puzzles: “If a tree falls in the forest and no one is around to hear it, does it make a sound?”

Imagine we have two people: Silas, who thinks a tree doesn’t make a sound, and Loudius, who thinks it does.

How would we turn this into a prediction? Well, if there’s a debate over whether something makes a sound, the natural thing to do might be to bring a recording device. Another thing we might do is to ask everyone if they heard the sound.

So we ask Silas and Loudius to register their predictions. We ask Silas: “If we place sensitive measuring equipment in the forest, will it detect compression waves from the falling tree?” He says, “Of course it will, but that’s not what a ‘sound’ is. A sound is when a person experiences an auditory sensation.”

We turn to Loudius with the same question, and he agrees: “Yes, the equipment would detect compression waves. That’s exactly what sound is.”

Then we ask them both: “Will any human report hearing something?” Both agree: “Of course not, there’s no one there.”

Notice what happened: Silas and Loudius make identical predictions about what will happen in the physical world. They both predict compression waves will be detected, and both predict no human will report hearing anything. If two positions produce identical predictions, the disagreement is purely meta-linguistic. Their disagreement isn’t about reality at all; it’s purely about which definition of “sound” to use.

Through our prediction-tinted glasses, we see the riddle dissolves into nothing. What seemed like a philosophical puzzle is actually just two people talking past each other and a debate about the definition of words—a decidedly less interesting topic.

Prediction Improves Discourse

We can apply this same thinking to basic conversation. Consider a conversation I had several years ago with someone who told me that “AI is an overhyped bubble”. Instead of having a debate that hinged on the definition of vague words like “AI” and “overhyped”, I said, “I don’t agree. Let me make some predictions.” Then I started to rattle some off:

1. Funding for AI will increase every year for the next five years

2. AI will remain the largest sector for tech investment over the next five years

3. All the major tech companies will still be focused on AI in five years…

“Oh—” he cut it, “I agree with all that. I just think people talk about it too much.”

OK, fine. That’s a subjective statement, and subjective statements don’t have to lead to predictions (see the appendix for types of statements that don’t lead to predictions). But by making predictions, what we were actually disagreeing about became much clearer. We realized we didn’t have a substantive disagreement about reality at all. He was just annoyed by how much people talked about AI.

Conclusion

Putting on prediction-tinted glasses transforms how you engage with ideas. Vague assertions that once seemed profound reveal themselves as empty when you ask, “What does this actually predict?” Endless debates dissolve when you realize both sides anticipate the exact same observations. And your own confident beliefs become humbler when you’re forced to attach probabilities to specific outcomes.

The practice is simple, even if the habit takes time to build. When you encounter a belief, ask: “What experience should I anticipate if this is true? What would I observe differently if it were false?” When you find yourself in a disagreement, ask: “What predictions do we actually differ on, or are we just arguing about definitions?” When you hold a strong belief, ask: “What probability would I assign to specific outcomes, and what evidence would change my mind?”

Of course, in many real-world situations, people aren’t striving for maximal predictive clarity. Sometimes people are aiming for social cohesion, emotional comfort, or personal satisfaction. Artful ambiguity, tribal signaling, and motivational fictions are important social lubricants. Not every utterance can or should be a prediction. Still, it’s better to let your values define what tribe you’re in and keep your beliefs about reality anchored to reality.

When you aim for clarity and truth, explicit predictions are the tool that forces belief to meet reality. By making our beliefs pay rent in anticipated experiences, we invite reality to be our teacher rather than our adversary. It corrects our models, sharpens our thinking, and aligns our mental map with the territory it’s meant to represent.

Appendix

Types of statements that do not lead to predictions

Not all statements lead to testable predictions. There are some statements that are so meaningless as to not predict anything, and others that are incredibly important, yet don’t predict anything. Some propositions, by their very nature, resist falsification or empirical evaluation. Here are a few categories of such statements:

1. Subjective Opinions: Statements like “I like that painting” express personal taste rather than objective facts. While opinions are vital, they’re typically immune to falsification (barring creepy futuristic mind-reading AI).

2. Normative Statements: Prescriptions like “We should be nice to each other” or “murder is wrong” express moral values rather than descriptive facts. While moral statements principles are important, they express values rather than facts about the world5. Normative statements can be built upon positive statements, but they still rely on a normative statement in the end. For example, you could test whether murder increases or decreases overall happiness, but that still rests on the normative premise that happiness is what matters morally.

3. Analytical Statements: Some statements are true by definition or logical necessity rather than because of how the world happens to be. “All bachelors are unmarried” is true because of what “bachelor” means. Similarly, “2 + 2 = 4” isn’t a claim you test by counting objects; it’s a consequence of how we’ve defined numbers and addition. These statements don’t generate predictions about what you’ll observe because they’re not about the world in that way. They can still be important, but they’re not the kind of beliefs that “pay rent” in anticipated experiences.

4. Meta-linguistic statements: The example we used above with the tree falling in the forest is ultimately a meta-linguistic one. What is the definition of ‘sound’? There is no prediction we could make here. We could be descriptivist and decide that the definition is however people use it, then go out and conduct a survey. We could make predictions about the result of that survey, but that doesn’t tell us what the one, true definition of sound is, just what it is given a descriptivist framework.
   Similarly, is Pluto a planet? There is no prediction that answering “yes” or “no” would make here. We have measurements of all the relevant parameters. The International Astronomical Union decided to create an official definition and, by that definition, it is not. But this is a question of which definition of “planet” we choose.

5. Vague Assertions: Some claims are so fuzzy or ill-defined that they can’t be properly evaluated. Politicians and pundits often traffic in this sort of strategic ambiguity. Statements like “My opponent’s policies will lead to disaster” or “The education system is broken” sound like factual claims but are too amorphous to test. Forcing such claims to make concrete predictions often reveals their emptiness or shows they’re actually testable once properly specified.

6. Inherently Unfalsifiable Claims: Some propositions resist testing, either by their very nature (like a non-interfering god) or by strategic design of the speaker (like adding ad hoc escape clauses when predictions fail). For instance, a deist god that created the universe but never interferes with it is undetectable by definition.

   Unfalsifiable claims like this may be personally meaningful, but they cannot be empirically adjudicated. They lie outside the realm of prediction because they don’t generate any anticipated experiences.

   However, we can still reason about some unfalsifiable propositions indirectly through thought experiments, such as those involving the Anthropic Principle (which asks why we observe the universe to be compatible with our existence).

   It’s also worth noting that people often mistakenly believe certain claims are unfalsifiable when they actually aren’t. “God answers prayers” is one such example. This belief does generate predictions: if true, we’d anticipate that prayed-for patients recover faster than those not prayed for, or that religious communities experience better outcomes in measurable ways. These are testable claims, even if believers sometimes retreat to strategically unfalsifiable versions (”God answers prayers, but in mysterious ways”) when the predictions fail.

The point here isn’t that non-predictive statements are worthless. Meta-linguistic analysis, subjective experiences, moral reasoning, and logical inference all play crucial roles in human cognition and culture. However, we need to distinguish unfalsifiable propositions from testable predictions.

Part 1: India and the Curses of Modernity

I thought a lot about what’s ahead for India (or, at least, Chennai, and wildly extrapolated it to the rest of India) as it develops. Unlike countries that stumbled blindly into modernity’s pitfalls, India can potentially learn from the experiences of more developed countries.

Obesity

People seem to walk around much more here for their daily errands (selection bias warning: I noticed this while walking around). Perhaps not coincidentally, obesity rates are noticeably lower than in the US, but obesity isn’t nonexistent. I looked up the numbers for India; in 1990, it was less than 1%. Today it’s above 7% and increasing at an accelerating rate (i.e., the curve is still bending upward).

While this, of course, pales in comparison to the US’s 40+% obesity rate, the trajectory is still concerning.

Perhaps we can sympathize with the US’s inability to foresee the rise in obesity. Changes in diet and lifestyle after WWII, such as suburban living, processed foods, and desk jobs, were unprecedented, and so were their health consequences.

But India has the data. They can see what might await them. Does that mean they can avoid it?

I’m skeptical. As much as we might like to solve problems like obesity with some motivational posters and inspiring speeches (any other David Goggins fans here?), the historical record isn’t encouraging. US obesity rates have risen steadily despite a plethora of interventions over the years12. Cookies are just so tasty and chips so crunchy. Willpower and cultural values have so far proven to be insufficient defenses, at least at the population level (i.e., they may work for you, but they’re not working for everyone).

It’s an interesting time for a question like this. Now there are pharmacological solutions to this. We have Ozempic. India could, theoretically, become the first country to deploy effective weight loss drugs (e.g., Ozempic, but more likely Tirzepatide/Retatrutide/whatever comes next) as a preventive public health policy. Mass produce them. Make them cheap. Stop the obesity crisis before it starts.

This is a tall order. It is unclear if this will happen anywhere, let alone in India, but there’s certainly a case to be made for it. The possibility exists in a way it never did for many other countries.

Social Isolation

Just in walking to the end of the street, I would see about 60 people outside, perhaps talking, waiting, or selling, but mostly just existing in a shared space. Many seem to know each other; people generally seem to know their neighbors (another selection bias warning: I’m only seeing people who are out and about). Each time I walked back, I’d pass two women who seemed to maintain a running mental log of every movement in and out of the neighborhood.

All of this connectivity made me wonder if there’s less social isolation in Chennai, one of the curses of modernity that I’m very concerned about.

I wonder how this social fabric is going to fare when they all have central air conditioning and personalized Netflix queues demanding their attention. Will they know their neighbors then? I don’t think we appreciate the significance of having better alternatives. Why sit outside in the hot sun talking to the neighbors when everyone can retreat into private, climate-controlled entertainment bubbles with the latest season of Stranger Things?

Can a culture resist this?

Birth rates offer one measurable proxy for cultural resistance to modernization. If India were successfully maintaining traditional social structures, we might expect to see it resisting the global decline in birth rates. But the data shows the opposite: India used to have a higher birth rate than the world average, but fell below that in 2010. By now, like many other countries, India has even dipped below the replacement rate.

Israel stands out as perhaps the only highly developed society maintaining high birth rates—ultra-Orthodox Jews maintain birth rates above 6 despite living in a developed economy. But they’ve created distinct social structures to achieve this. Without such deliberate resistance, India appears to be following the standard modernization trajectory.

Fentanyl

I’m sure this marks me as an American, but another curse of modernity I was on the lookout for was fentanyl. Fortunately, I didn’t see any signs of it while walking the streets. I asked ChatGPT about this, and it suggested that India never had the US-style prescription-opioid wave. This suggests to me both that our opioid crisis was highly contingent—had a few companies behaved differently, we might have been in a very different situation—and that there’s a path out of it. Yes, the opioid epidemic will continue to do damage, but maybe there is a light at the end.

But before we congratulate India too much, we should acknowledge that there are tradeoffs here. How many people are suffering far from my eyes because they can’t get access to effective pain relief? We don’t want to make the mistake of comparing the measured (people dying of opioid overdoses, which is tracked by governments) vs the unmeasured (people quietly suffering for lack of opioids, which is much harder to track). We don’t know how many people are suffering in their homes or hospitals because they don’t have access to opioids or fentanyl. We shouldn’t forget about them.

I did see some effects of alcohol, though, including men passed out on the sidewalks with alcohol next to them. This was uncommon. Much more common was for people (usually men) to simply be sleeping outside, seemingly due to a lack of other options.

Carbon Emissions

The big developmental question for me is how India can become a high-income country while staying on a carbon budget that keeps global emissions within reasonable limits. It’s often been said that there’s no such thing as a high-income, low-energy country. We should take this fact seriously. This is not just India’s problem; it’s the world’s. It does not, however, imply that a high-income economy must be a high-carbon-emitting one.

For those who haven’t been watching the charts (looking at you, Europeans who think forgoing air conditioning will save the planet): China has risen. India wants to do the same, and you can see the GHG emission impact in this graph. The key developmental question I see is how to provide the energy India needs without repeating (or exceeding!) China’s emissions curve.

This seems like it should be a top global priority. India’s rise is inevitable3; the only question is how cleanly it happens. There are a plethora of components to this, but it will come as no surprise to readers that I think increased nuclear power should be among them. Fortunately, such efforts are underway. The US and India signed an agreement to work together to build US-designed nuclear reactors in India. More of this, please.

Part 2: On Development

Causal Order of Development

One of the things that I thought most about during my trip was the mechanics of development, in particular, the causal order. Take the electrical wiring: like in many developing countries, wiring in India is chaotic, tangled, and seemingly improvised.

So does a country clean up its wires as a cause of development, or as a consequence? Should governments decree “Thou shalt untangle thy cables” and hope civilization blossoms? Or do people one day simply wake up, glance skyward, and collectively decide, “Enough of this tangle; we’re better than this”?

In the case of wires, I don’t think it can be exclusively bottom-up—neighborhoods aren’t going to spontaneously band together to reorganize the electrical grid. But the causal direction and optimal development path remain unclear. In economic theory, this is known as “circular cumulative causation”, where all the core variables are interlinked in causal loops (basically, everything causes everything). So the question is, where does one focus one’s resources to get the most development?

This type of question came up repeatedly throughout my travels. When I observed that something looked different in the developed world, should that difference be actively engineered or allowed to emerge naturally? Take driving patterns, for instance. Traffic in Chennai is chaotic by Western standards: lots of honking, lane markings are ignored, and vehicles of all sizes jockey for position. We don’t drive quite like that in the US. And I’m pretty sure Indians would, on average, get where they’re going faster if they followed some more rules4. But how exactly does a country’s driving culture evolve from chaotic to orderly? Through enforcement, wealth, or some combination of factors?

Tourism

One thing that countries often rely on to fund their development is tourism. People come in clutching their foreign currency, practically begging to be relieved of it, and local businesses are generally more than happy to accommodate. Most places have entire municipal departments dedicated to this sort of thing. Chennai, it seems, is completely uninterested in such things.

The near-complete lack of non-Indians is what surprised me most. This is, after all, a major city of several million people. So where were all the international business travelers? Where were the tourists with their cameras and guidebooks?

After leaving the airport, I saw no non-Indians for days. Not a single one. The first non-Indian I saw was a white person in the distance at the mall, but as I got closer, I realized it was actually a friend of mine in town to see me, who just happened to want to check out the mall.

When just wandering around, I saw basically one non-Indian the entire time, and I’m not even entirely certain about that. Late one night, I passed someone while walking on the street who I think was white. We did that thing where you’re both surprised to see each other and exchange a brief, knowing smile of mutual recognition. But by the time I’d properly registered what had happened, we’d already passed each other. I wish I’d stopped him to ask what he was doing there. Perhaps he was wondering the same about me.

Tech

There’s also a bit of a paradox in the development of India’s IT sector. If you work in tech in the US, you’re sure to work with loads of talented Indian engineers. They’ve succeeded at all levels of tech companies; the CEOs of Google and Microsoft are both Indian-born.

And yet India’s own tech infrastructure is conspicuously poor. You see it directly—the e-visa site kept crashing on me—and more generally in the digital world. India has produced precisely zero globally used tech products. No apps that everyone has on their phone. No social networks. No operating systems. Meanwhile, China built TikTok, Israel built Waze, South Korea has Samsung, and Sweden built Spotify. India has more software engineers than almost anywhere on the planet, yet somehow hasn’t created a single thing the world uses daily.

Now, I’m sure some of this has to do with the fact that by being in Silicon Valley, those same software engineers are, by definition, not also working in India. So yes, some of this is good old-fashioned brain drain. But I think this explanation only goes so far. I have not spent enough time thinking about this, but I sense that there’s an insight somewhere that I just haven’t put together.

Low-trust Societies

I don’t know how related this is, but I wonder how much the intangibles play a role. India, or at least Chennai, is, in many ways, what Francis Fukuyama would call a low-trust society. People don’t trust systems to work, so they don’t respect them, which means the systems don’t work, validating the original distrust and perpetuating the systems. It’s a hard cycle to break.

The moment it hit me hardest was at the airport security line. I was queuing—or so I thought—at the x-ray machine when someone appeared beside me and slid his tray in front of mine. Clearly an emergency, I thought. The poor fellow must be about to miss his flight. I held my tray back to help him.

Then the next guy comes from behind and grabs my tray and pulls it back from the machine, and puts his in its place. Oh no, another emergency! This man must be racing to see his dying grandmother!

But when a third person did the same thing—again with no communication, no eye contact, no apologetic “terribly sorry, bit of an emergency” like you’d get in the UK—I started wondering how it could be that so many Indian grandmothers were simultaneously on their deathbeds.

It was about then that I figured out the dynamic: we weren’t politely queuing in an orderly line. We were competing. This was a contact sport.

When the next person tried to cut in, I thought: ‘You may be from India, but I’m from Indiana. And I would wager that I know more about basketball than you do.’

Time to deploy the box-out.

Once I locked in, nobody got around me. Wide stance, drop the hips and keep a low center of gravity, arms out for balance and territory, light contact with the forearm if needed—don’t be shy, and don’t be afraid to stick that butt out to create separation. Bobby Knight would have been proud.

But it’s also like, “Why are we doing this?” Perfectly functional queuing systems exist. I’ve seen them work beautifully in dozens of countries. I’ve even participated in them myself on numerous occasions.

I’ve had similar (and worse) experiences in other countries where everyone fights for position rather than waiting patiently, and I know how to manage in these situations. But it brings out the worst in me. I think it brings out the worst in everyone.

How does this change? How does a society form better norms? The answer to this specific situation seems simple: if airport security simply enforced the rules, the behavior would probably stop. Would that propagate to bus terminals and so on? Hard to say, but this seems more fixable.

Part 3: Economics, Culture, and Daily Life

Economics and Culture

I also noticed the interesting interaction between economics and culture, particularly how labor abundance shapes everyday life. In some restaurants, you might have three people attending to your table: one takes your order, another recommends dishes, and a third hovers nearby ready to assist.

What’s going on is that labor is so cheap that seeming inefficiencies don’t matter. Wages are low relative to the marginal revenue product. That is, where the cost of one more person’s time is cheap enough that businesses optimize for customer throughput and attentiveness rather than labor efficiency, so you get three people doing a job one or two could do. I noticed that this style of service makes me uncomfortable, likely because I’m calibrated to American norms where such attention signals extreme luxury.

I think this highlights something interesting about India’s economic moment. Right now, it’s not uncommon for middle-class Indians to live a little bit like aristocrats: personal drivers who wait outside shops for hours, maids who clean the place, and maybe even a cook as well. A middle-class Indian family likely has more household help than a middle-class American family.

But as India gets richer, I expect those luxuries will disappear. Two complementary forces will drive this change:

First, wages everywhere will rise. It’ll start in the sectors that become more productive through technology, such as tech companies and modern factories. But to keep everyone from joining the more productive sectors, wages in the more stagnant sectors will have to rise, too. That is, people will have to pay maids, drivers, and cooks more so they don’t quit and become software engineers5. Thus, customers will end up paying more for the exact same services, a phenomenon known as Baumol’s cost disease.

In addition, technology will create new opportunities. Personal drivers are already being replaced by ride-sharing apps where one driver serves many people throughout the day instead of waiting idle for hours. Live-in maids will give way to weekly cleaning services, and more washing machines, dishwashers, and robot vacuums. The economy will gain efficiency but lose the personal touch.

This means that middle-class Indians today enjoy a lifestyle that will be impossible for middle-class Indians tomorrow, even though tomorrow’s Indians will be richer. If you’ve ever watched Mary Poppins and wondered how a merely upper-middle-class banker (Mr. Banks—score one for nominative determinism), can afford a nanny, a full-time cook, and a live-in maid, it’s the same phenomenon.

Conservative Culture

The culture is more conservative, but selectively so. One doesn’t see couples being affectionate or any indication that anyone is or has ever dated anyone else or that “dating” is even a word. There’s clearly more pressure on women to dress modestly than even in, say, Sri Lanka. I don’t think I saw a woman’s knee the whole time. Yet I saw more middle-aged women’s exposed midriffs in Chennai than I’ve seen anywhere outside a beach. Just walking down the street, there’s side-belly everywhere from aunties in saris. I guess what makes something modest is subjective.

Another thing I noted about this “conservative” society was the word SEX plastered in big letters all over the city. It was a little hard to miss, actually. When riding in a tuk-tuk (which they call “autos”), I saw them plastered all over the place. I took a picture of one and, according to ChatGPT, the poster is an advertisement for a sex-related Ayurvedic clinic in Chennai. The bullet list says:

Loss of manliness

• Lack of male virility

• Premature ejaculation

• Weak semen discharge

• Small penis

• Skin-related diseases

Skin color

Speaking of advertisements, it’s hard not to notice how light-skinned everyone in advertisements is compared to the much darker skin color of those I saw. Billboards are almost exclusively covered with incredibly light-skinned Indians. There was basically no overlap between the skin tones I saw in the people around me and the people I saw on the billboards6. I don’t think this is a psychologically healthy thing for a society.

Tolerance

Religious

Religion seems to be very important and somewhat omnipresent, although in a different way than I had expected. There were fewer temples than I anticipated, for instance. But religion is omnipresent in a different way; I bought clothes from a store that had an in-house mandir (small temple area) and a priest who blessed my clothes for me.

I saw lots of churches and temples (didn’t see any mosques, but also wasn’t looking for them), and everyone seemed pretty chill about it all. No religious conflict or tension was apparent from just walking around. Obviously, you can’t see everything just from walking around and talking to people, but this was very different from, say, Bosnia, where religious tension felt high.

The caste system wasn’t visible. It’s certainly there under the surface, but the lack of visible signs seemed like a good thing.

Head Waggle

At the risk of reading too much into things, I wonder about the significance of the ambiguous (mostly-South) Indian head waggle. It generally means yes, though it can also mean “I’m following” but sometimes just “I’m listening (but not following)” and maybe a little of “I don’t agree but you seem nice” and, every once in a while, it seems to mean “No but isn’t this a nice way to be told ‘no’?”

Does the ambiguous head waggle make having a multi-religious, multi-ethnic society easier? I have no idea, but it’s fun to think about.

Miscellany

• I saw less extreme poverty than I expected. This could, of course, be more of a statement about where I went and what I did than about Chennai, but I still made a note of it. I didn’t see many instances of children begging, for example. (In fact, I think I saw more children begging from Romas in Italy than I saw in Chennai.)

• The amount of commerce that happens over WhatsApp is incredibly high. Instagram and WhatsApp seem to be the foundational infrastructure for much of the economy.

• I felt very safe wandering around, even at night. There are generally so many people around that you feel safe. I wouldn’t be surprised if women feel differently.

• I can see why tourism is somewhat lacking. If I’m being honest, Chennai is notable for lacking charm. I was often in search of cute little coffee shops or pastries. There are coffee shops and there are pastries, but nothing that would match a recognizable description of “cute”.

• What Chennai lacks, above all, is verdure. There’s concrete everywhere, and even the sparse greenery is fenced off, like you need a permit to visit it.

• India (or at least, Chennai), seems to rely heavily on personal recommendations. You’re going to buy shoes? Don’t go until I ask my cousin which shop to go to. The results from this cousin system often overlap with what one would have figured out from Google Reviews, and don’t seem any better. In fact, I would suggest it’s slightly less good than Google Reviews. I went to a bunch of stores at a mall that were cousin-approved. Then a bunch that weren’t just for comparison. Not much of a difference.

• Mahatma Gandhi still occupies a revered position within India. While we have Washington on the $1, Lincoln on the $5, and so on, Gandhi’s portrait adorns every denomination of Indian currency that I saw.

• I was surprised by the abundance of leather products for sale. My guess is this is from water buffalo, which, unlike cows, are not sacred in Hinduism. India is actually one of the world’s largest exporters of buffalo meat. However, I didn’t see any buffalo meat on any of the menus. (I didn’t actively seek it out, but I was trying to take note of what was on the menus).

• The supposed sacredness of cows wasn’t at all apparent from walking around. I visited many temples and attended religious ceremonies, but never encountered anything resembling “sacred cow” reverence. Cows wandered the streets freely, but seemed merely tolerated rather than revered. I saw no garlands on them, no special treatment. Many looked underfed. As far as I could tell, they weren’t treated any better than chickens (other than not being eaten, which I guess is something).

It’s a compelling narrative. It’s also deeply flawed.

This matters because Reich isn’t just another pundit. He’s a former U.S. Secretary of Labor under President Clinton, a Berkeley professor, and a bestselling author with millions of social media followers. “The System” itself became a national bestseller. When someone with his platform and credentials makes sweeping economic claims, people listen. They form worldviews around his assertions. They vote based on his narratives.

To be clear: there are real problems with inequality in America. There are valid concerns about corporate power and the influence of money in politics. These issues deserve serious, evidence-based discussion. But Reich’s book, rather than illuminating them, obscures them behind a fog of hyperbole, oversimplification, and demonstrably false claims.

Throughout “The System,” Reich makes sweeping assertions that crumble under scrutiny. He ignores evidence and seems more interested in stoking outrage than understanding the world—a dangerous approach, since bad diagnoses lead to bad prescriptions that harm the very people they’re meant to help.

We can have empathy for those who struggle while still being grounded in reality. In fact, crafting policies that work requires exactly this combination. The challenges facing working Americans are too important to be addressed with anything less than intellectual honesty. The outrage-fueling falsehoods that Reich provides won’t deliver solutions.

Reich’s Thesis: Democracy vs. Oligarchy

Reich tells us his agenda is “neither right nor left.” That’s the old framework, he insists. “Today the great divide is not between left and right. It’s between democracy and oligarchy.”

It’s a neat rhetorical trick. By framing the debate as “democracy vs. oligarchy,” Reich attempts to position himself above the partisan fray. Who, after all, could be against democracy?

But this performance of neutrality falls apart the moment you examine Reich’s actual proposals. His supposedly post-partisan agenda includes:

• Higher minimum wages

• Stronger unions

• Gun control

• More corporate regulation

• Higher taxes on the wealthy

• Expanded social programs

If this looks familiar, it’s because it’s the standard progressive Democratic platform. There’s nothing wrong with advocating these positions, but pretending they transcend left-right politics doesn’t withstand even cursory scrutiny. Reich wants the moral authority of standing above politics while advocating for an explicitly political agenda. He can’t have it both ways.

Reich’s Economic Claims

Inherited Wealth

Reich claims America has transformed into an inheritance-based oligarchy—where most of today’s ultra-wealthy are heirs, not entrepreneurs. He tells us:

We’re already at a point where many of today’s super-rich have never done a day’s work in their lives. Six out of the ten wealthiest Americans alive today are heirs to prominent fortunes.

This would be damning if true. However, it is false.

Reich does not cite any sources for his claims, so I don’t know where he got this idea. All I can find from searching it is him saying the same thing in other outlets. I checked the Forbes list of America’s richest people. Of the top ten, exactly zero are heirs to prominent fortunes. Nine founded the companies that made them wealthy. The only exception is Steve Ballmer, who joined Microsoft early but wasn’t a founder.

I did manage to find what I think he’s talking about in Wikipedia’s list of the top ten richest people going back to 1987. I went over every year and the closest I could find was the 2001-2004 time period when the Walton heirs make appearances due to a methodological change—Forbes used to classify them together as the “Walton family” but in 2001 split them out into individuals. In that period, five of the people on the list were Walton heirs. However, this is a thing of the past; by 2005, all but one had fallen off the list.

In fact, a 2014 study examined the Forbes 400 list of wealthiest Americans from 1982 to 2011. They found that the proportion of the Forbes 400 who grew up wealthy decreased from 60% in 1982 to just 32% in 2011. Meanwhile, the share who came from upper-middle-class backgrounds (what the authors call “some money”) increased by about the same amount.

Even more tellingly, 69% of those on the 2011 list started their own businesses, up from 40% in 1982. The Waltons and Marses are increasingly the exception, not the rule. The world is changing in precisely the opposite direction Reich claims.

Reich’s zero-sum view of wealth—where the rich getting richer means everyone else gets poorer—fundamentally misunderstands how wealth creation works. Wealth isn’t a fixed pie that gets divided up. When someone invents a new technology, develops a better process, or creates a product people want, they expand the total amount of value in the world. The entrepreneur captures some of that value as profit, but consumers capture most of it through lower prices, better products, and entirely new capabilities that didn’t exist before.

Reich repeatedly tells us that the rich have “siphoned off economic gains” from the 90% to themselves. Yet, economic research shows that “only a miniscule fraction of the social returns from technological advances over the 1948-2001 period was captured by producers, indicating that most of the benefits of technological change are passed on to consumers rather than captured by producers.”

Yes, Google’s founders became billionaires, but billions of people now have instant, free access to virtually all human knowledge. When entrepreneurs like Gates or Jobs create billions in wealth for themselves, they’re creating trillions in value for society.

Do the Richest People Work?

But let’s examine the deeper absurdity: the idea that today’s ultra-wealthy don’t work. Here’s another great opportunity to test his claims, to make contact with reality. We can simply ask, do the richest people in the world work?

Here’s Bill Gates reflecting on his early years:

When I was your age, I didn’t believe in vacations. I didn’t believe in weekends. I pushed everyone around me to work very long hours. In the early days of Microsoft, my office overlooked the parking lot—and I would keep track of who was leaving early and staying late.

There are a lot of things you could call him, but lazy isn’t one of them.1

Or consider Elon Musk, currently the richest person in the world. The Walter Isaacson biography of him talks about him sleeping on the factory floor and working 100-hour weeks. Here’s a quote from the book:

From the very beginning of his career, Musk was a demanding manager, contemptuous of the concept of work-life balance. At Zip2 and every subsequent company, he drove himself relentlessly all day and through much of the night, without vacations, and he expected others to do the same.

This shows how divorced from reality Reich is. He’s so focused on his narrative about rich people being lazy and entitled that he fails to see the completely contradictory facts right in front of him.

Income and Net Worth

When Reich isn’t demonizing the rich, he’s telling you how poor everyone else has become. His go-to claim is: “Most people’s incomes haven’t risen for four decades.”

This is the kind of statement that usually gets thousands of retweets, yet, unfortunately, near-zero fact-checks. It also happens to be demonstrably false.

Once again, Reich provides no sources, so I googled it. Here’s what the actual data show for real (inflation-adjusted) median personal income in the US:

There have been plateaus and a noticeable dip around the Great Recession, but the trend is unmistakable: up and to the right. Not flat. Not declining. Rising.

This exemplifies the maddening experience of reading Reich. He makes a bold, specific claim. You want to verify it. But there’s no footnote, no source, no hint of where he got this “fact.” And when you type his statement into Google, the results often contradict his claims.

But Reich’s statement feels true, and we should think about why. Even as real incomes have risen, housing costs in desirable metros have exploded, college tuition has outpaced inflation, and daycare and medical costs have become fodder for water-cooler horror stories. People see billionaires’ wealth plastered across social media while they’re struggling to save for a down payment. The fact that the median American is better off statistically doesn’t negate the real anxieties people face about making rent or paying off student loans.

Food spending shows how the story gets complicated. While food prices have risen, incomes have risen faster. In the 1960s, Americans spent about 17% of their income on food; today it’s under 10%. This decrease in spending on necessities has freed up income—but much of that freed income now goes toward bidding up inherently scarce goods like real estate in desirable metros. We’re not poorer overall, but economic competition has shifted from securing basic needs to competing for positional goods2. When people feel the squeeze of rising housing costs while the savings on food happen invisibly in the background, they understandably feel like they’re falling behind.

Of course, no single number is going to tell us everything we need to know about how people are doing. Another thing we could look at is net worth. The Federal Reserve publishes changes in US family finances. Here’s what they say in their most recent report:

[Between 2019 and 2022], real median net worth surged 37 percent to $192,900, and real mean net worth increased 23 percent to $1,063,700, accelerating the steady growth experienced over the 2013–19 period. The 2019–22 changes imply some narrowing of the wealth distribution between surveys. Indeed, growth in median net worth was the largest increase over the history of the modern [Survey of Consumer Finances], more than double the next-largest one on record.3

The Shrinking Middle Class

Reich also warns ominously about the “shrinking middle class4,” painting a picture of widespread downward mobility. This is technically true—the middle class has shrunk. But here’s what he doesn’t tell you: as you might guess from the numbers above, it’s largely because people are getting richer.

I checked many different sources for this because there are different ways of calculating the middle class. According to the Center for Retirement Research, using data from the Urban Institute, the middle class has shrunk from 39% to 32% primarily because the upper middle class has grown from 13% to 29%. The lower middle class and poor have also shrunk during this period (1979 to 2014).

Pew Research, looking at the period from 1971 to 2021, found similar results:

• Middle-income households fell from 61% to 50% of adults

• Upper-income households rose from 14% to 21%

• Lower-income households increased from 25% to 29%

In other words, for every person who fell out of the middle class into poverty, nearly two rose into the upper-income tier.

A third source, the US Census Bureau, confirms the same pattern (also inflation-adjusted), as you can see in the figure below:

Despite Reich’s pronouncements, this isn’t a dystopia; it’s upward mobility.

International Comparisons

He also tries to convince Americans that they’re poorer than people in other developed countries. Consider this assertion: “Considering taxes and transfer payments, middle-class workers in Canada and much of Western Europe are better off than in the United States. The working poor in Western Europe earn more than do the working poor in America.”

Once again, Reich offers no sources. I googled lots of variations of the phrase and, in short, I can’t tell what he’s talking about.

He’s not talking about income because he specifically says he’s considering transfer payments, which include government benefits like welfare, unemployment insurance, and social security. I think the best thing to look at would be consumption because it reflects real living standards better than income alone, since it accounts for transfers, benefits, and cost of living.

So I think the best thing to look at then is Actual Individual Consumption (AIC), which measures all goods and services actually consumed by households. The story it tells is that per capita AIC is highest for the US, even higher than countries that have higher per capita GDPs, like Norway and Luxembourg. It’s yet another Reichism: He says something, you Google it, and the reality is the exact opposite.5

Reich’s “Rigged System” Claims

How the Rich Vote

Reich’s claims extend beyond economics to our political and justice systems, where he insists the wealthy have rigged institutions in their favor. According to Reich, the super-rich have essentially purchased America:

The concentration of wealth in America has created an education system in which the super-rich can buy admission to college for their children, a political system in which they can buy Congress and the presidency, a health-care system in which they can buy care that others can’t.

Of course wealth confers advantages in healthcare, education, and, I would add, virtually everything else. This has been true in every society throughout history. But Reich isn’t making a general complaint about inequality; he’s making specific claims about oligarchic control. These we can test.

If the super-rich can “buy the presidency,” then Reich should be able to predict every election by simply checking which candidate the wealthy prefer. Fortunately for him, there are betting markets on this very thing, so he should be able to turn his wisdom into a fortune, join the super-rich, and then pick the next president.

But here’s where his theory crashes into reality: the super-rich don’t vote as a bloc. Billionaires have backed both sides of every election in living memory. Tech moguls lean left while energy executives lean right. Wall Street hedges its bets across both parties. Despite Reich’s claims, if there’s a secret oligarch meeting where they decide who becomes president, someone forgot to send out the memo.

He frames politics as a party of the people (Democrats) versus a party of the rich (Republicans). This is certainly a common sentiment on the left: a 2018 study found that Democrats estimated 44% of Republicans earned over $250k. The real number, however, is just 2%.

Even if we could ascribe the “party of the rich” to a single party, it’s not clear which party that would be. Certainly, in the minds of Reich and many on the left, it’s the Republicans. But according to Vox, it changes over time, and as of 2012, it’s the Democratic Party.

I searched for more recent data, but the most recent I could find were only for white voters. Even with this narrower demographic, the same pattern holds: higher-income voters have shifted toward Democrats.

But, more to the point, Reich’s simplistic “rich versus poor” framework fails spectacularly when you examine actual voting patterns. The truth is that people don’t vote based solely on economic interests. They vote based on values, cultural issues, and their vision of America’s future.6 Reich desperately wants the divide to be economic rather than ideological, but real-world voters stubbornly refuse to cooperate. In recent elections, both the wealthiest and poorest Americans have tended to vote for the same party (Democrats), while middle-income voters split. This is the exact opposite of what his theory would predict.

Affluenza and the Justice System

Much like everything else, Reich wants you to believe the justice system is rigged for the rich. And to be clear: wealth absolutely provides advantages in our criminal justice system. Rich defendants can afford bail while poor ones sit in jail awaiting trial. They can hire teams of lawyers while public defenders juggle hundreds of cases. They can afford expert witnesses and private investigators. These are real inequities that deserve serious consideration.

But Reich isn’t interested in these substantive issues. Instead, he reaches for the most inflammatory example he can find, the infamous ‘affluenza’ case:

An even more flagrant example is Ethan Couch, a Texan teenager who killed four people and severely injured another while driving drunk[...] a psychologist who testified in Couch’s defense argued that the teenager suffered from “affluenza,” a psychological affliction said to result from growing up with wealth and privilege. Couch served a 720-day sentence.

It’s a shocking story. I remember when this happened. But the thing to keep in mind is, as far as I have been able to find out (e.g., see the Wikipedia page here), this term was used one time in the entire history of the US criminal justice system by a single psychologist, and was immediately met with endless ridicule. Reich presents this as if “affluenza” is now standard legal doctrine, as if wealthy defendants routinely waltz into court, claim their money made them do it, and saunter free. But this case made national headlines precisely because it was so outrageous and stupid.

If Reich’s thesis were correct—if the rich really could, to use his exact words, “buy their way out of jail”—we’d need to explain some inconvenient facts. Rich people do serve serious prison time. Here are some recent examples:

• Bernie Madoff died in federal prison while serving 150 years

• Elizabeth Holmes is serving 11 years for fraud

• Sam Bankman-Fried got 25 years for his crypto crimes

• Harvey Weinstein is serving 23 years

• Martin Shkreli was in prison for six and a half years

These aren’t small fry. These are exactly the kind of ultra-wealthy, well-connected people who should be able to game Reich’s “rigged system.”

The tragedy here is that by reaching for hyperbole—by pretending one-off outrages represent systematic policy—Reich ignores the real reforms we need. Bail reform is a genuine issue affecting millions. The underfunding of public defenders is a problem. I believe we should make it a policy goal to ensure everyone gets competent legal representation (AI lawyers?) regardless of their bank account. But throughout his entire book, Reich never mentions these substantive problems. Instead of building coalitions around achievable reforms, he offers cartoon villains and simplistic narratives.

Our goal should be to understand what’s actually true about the world. Reich would like you to believe that rich people can just say “affluenza” and walk out scot-free. Thankfully, that is not how the world works.

Reich’s tendency to oversimplify and misrepresent complex systems—whether it’s the justice system or the economy—wouldn’t matter so much if he were just a commentator. But Reich has actually shaped policy, and the results demonstrate exactly why his simplistic thinking is dangerous. His approach reminds me of H.L. Mencken’s famous quip: “For every complex problem there is an answer that is clear, simple, and wrong.” The System is full of them.

When Reich Made Policy

Though Reich spends much of the book vilifying rich people like Warren Buffett and, especially, Carl Icahn, there is one place where they all agree: CEOs are overpaid.

In the early 1990s, CEO pay started to rise at a much higher rate than before. In 1993, the Clinton administration, with Robert Reich as Secretary of Labor, decided to tackle excessive CEO pay. The president signed a bill into law limiting the tax deductibility of executive salaries to $1 million7. Anything above that couldn’t be written off as a business expense. There was just one small exception: “performance-based” compensation remained fully deductible.

I don’t know how involved Reich was with this particular bill, but, according to his Wikipedia article, he was “one of the most powerful members of the Clinton cabinet. […] As a member of the National Economic Council, Reich advised Clinton on health care reform, education policy, welfare reform, national service initiatives, and technology policy, as well as deficit reduction and spending priorities.” In short, he likely had a front-row seat to this policy’s creation.

What followed was a case study in unintended consequences. Rather than accept lower pay, CEOs and their boards simply restructured compensation packages. Base salaries stayed at $1 million (conveniently deductible), while stock options and “performance” bonuses exploded.

In short, the policy failed completely. Instead of limiting CEO pay, it sent it soaring. The very problem Reich rails against throughout The System was supercharged by a policy enacted on his watch.8

Reich witnessed this spectacular backfire firsthand. He saw a well-intentioned policy create the opposite of its intended effect. One might hope this would make him cautious about proposing simple solutions to complex problems.

That hope would be in vain.

Throughout The System, Reich shows no evidence of having learned from this experience. He still presents policy interventions as if we can simply decree outcomes. Want to limit CEO pay? Just cap it! Want higher wages? Just mandate them! Want less inequality? Just redistribute everyone’s money!

The CEO pay debacle perfectly illustrates why we can’t trust Reich’s economic prescriptions. He sees a problem, imagines a solution, and assumes reality will comply. When it doesn’t, he doesn’t update his worldview. He just proposes more of the same, harder.

False Claims

The System reads like a game of Mad Libs where the narrative structure is pre-written and only the details change. The story is always the same: you’re a victim—poor, mistreated, and helpless—at the mercy of oligarchs who’ve rigged everything against you.

It’s a compelling narrative for some. The problem is that this narrative structure comes first, and the facts are forced to fit. When you approach economics this way, you end up cherry-picking data, disregarding contradictory evidence, and making claims that can’t survive a Google search.

The sheer volume of false claims raises an obvious question: Why does Reich need to fabricate evidence? Strong arguments stand on real data. The fact that he resorts to falsehoods suggests that reality doesn’t support his worldview.

Getting the facts wrong doesn’t help anyone—it just leads to bad solutions. It reminds me of the Thomas Sowell quote: “When you want to help people, you tell them the truth. When you want to help yourself, you tell them what they want to hear.” Reich, it seems, has chosen the latter path.

Preaching Only to the Converted

Perhaps the most frustrating aspect of Reich’s approach is how unnecessary it is. There’s an important story to be told about what happened to the American worker over the past sixty years involving global competition, technological change, and shifting skill demands. But by framing every issue as an apocalyptic battle against “oligarchy,” even his valid points get lost in the hyperbole.

This approach represents a failure of intellectual ambition. Reich’s arguments only work if you’re already converted. If you’re not already primed to make these leaps, you’re left behind. If you believe that progressive policies are inherently good in all cases, then you might agree. But if you don’t already, he does nothing to convince you.

Instead of building coalitions, Reich serves up comfort food for progressives: cartoon villains, simple solutions, and that warm feeling of righteousness. But intellectual comfort food, while satisfying in the moment, doesn’t nourish, and it doesn’t solve real problems.

Why This Matters

You might wonder why I’ve spent so much time fact-checking Reich. Am I just being pedantic? Does it really matter if he exaggerates here and there (and everywhere)?

It matters immensely. These aren’t just abstract facts—they’re what we might call instrumental facts. They’re the instruments we use to navigate reality, create our whole frame of reference, and make decisions. When those instruments are broken, we make broken choices. Reich’s false claims about wages, wealth, and inequality aren’t harmless errors. They’re the foundation for policies that could harm the very people they claim to help. But there’s an even more insidious effect of Reich’s false narrative.

The most damaging part of Reich’s narrative emerges when he addresses meritocracy. According to him, belief in meritocracy is a trick played by oligarchs: “The oligarchy wants Americans to view the system as a neutral meritocracy in which anyone can make it with enough guts, gumption, and hard work.”

His counter-message is stark: “Don’t believe the system is a meritocracy in which ability and hard work are necessarily rewarded. Today the most important predictor of someone’s future income and wealth is the income and wealth of the family they’re born into.” Once again, no citations, and Googling it suggests it’s completely false, but I want to focus on another point.

This isn’t just wrong. It’s poisonous. When people believe effort is futile, they stop trying. When they believe the system is rigged, they disengage. When they believe they’re helpless victims, they become helpless victims.

If you think America is an oligarchy where merit doesn’t matter, you might not pursue education, develop skills, or start a business. You might not even immigrate here, despite the fact that immigrants in America reach levels of prosperity unmatched anywhere else in the world. Now imagine a society where millions hold these beliefs—it becomes a self-fulfilling prophecy of stagnation and decline.

I want people to believe effort matters—not because the system is perfect, but because belief in agency leads to better outcomes than learned helplessness. Yes, advantages compound. Yes, the playing field isn’t level. But the solution isn’t to tell people the game is rigged so they shouldn’t even try.

Reich wants reforms led by active citizens. I agree, but here’s what I’d add: I want reforms led by informed, active citizens. Citizens who’ve done their homework, checked their facts, thought through potential unintended consequences, and promoted open dialogue where people can share ideas and work together.

Conclusion

The American worker deserves better than fairy tales. The issues Reich raises deserve better than sloganeering. They deserve careful analysis that acknowledges trade-offs, engages with criticism, and builds coalitions rather than vilifying anyone who disagrees. And readers deserve better than books that prioritize ideological comfort over inconvenient truths.

Until we demand better—from our public intellectuals, our politicians, and ourselves—we’ll keep getting books like The System. And we’ll keep wondering why our problems never seem to get solved, even as we’re told, again and again, that the solution is just one revolution away.

The real revolution we need isn’t against imaginary oligarchs. It’s against intellectual laziness, tribal thinking, and the seductive falsehoods we tell ourselves about how the world works. That revolution starts with each of us deciding that truth matters more than tribe, that accuracy matters more than simplicity, and that real solutions matter more than feeling righteous.

The beauty of my system is its breathtaking simplicity. No calorie counting. No forbidden foods. You don't even need to exercise, unless you count sprinting to the bathroom as cardio, which I suppose technically it is.

The trick is to go to a foreign country and eat or drink something inadvisable. This has happened to me several times—I seem to have a gift for it, really—but the last time was particularly calamitous. What followed was two weeks of what we might charitably call "severe gastrointestinal distress," though that rather understates the experience. I began planning my vacation routes entirely around bathroom (squat toilet) locations.

Now, with another trip approaching, I find myself researching ways to avoid a repeat performance. The condition is officially known as traveler’s diarrhea (TD) and can be caused by any number of bacteria (though most commonly E. coli or Campylobacter), viruses (especially norovirus), or parasites, and can lead to vomiting, cramps, and other delightful symptoms.

The standard prevention advice starts with the obvious: when you see that street vendor selling jars of mysterious alcohol with cobras in them, just keep walking. (I say this with the authority of someone who has never, ever managed to just keep walking.) The Mayo Clinic has also helpfully condensed centuries of learning into their rule of thumb: "boil it, cook it, peel it, or forget it."

But as I was researching, I started to see suggestions about probiotics and got interested.1 The idea is to fight off the bad bacteria by recruiting an army of good bacteria. It's like preparing for an invasion by hiring friendlier invaders first.

Though it’s not the first line of defense, I’ve come around to the idea that it might be worth giving it a try. You're never going to achieve zero exposure to nasty pathogens when traveling. You can try—you can adhere to a diet of food so hot you can see steam rising off it, but you must realize you're eating it with silverware that was washed in the very water you're trying to avoid. You can boil your food into submission, but then you pick up your glass with hands you washed in the sink. The bacteria are everywhere, lurking (plotting?) on their fomites.

The point is, perfect avoidance is impossible. What you need instead is a digestive system that can handle the occasional bacterial incursion without staging a full revolt. This is where the probiotics come in—building up your intestinal defenses before the enemy arrives. So join me as I go down the probiotics rabbit hole.

Comparing Different Probiotics

The best reference I’ve found on probiotic benefits is from a 2024 meta-analysis by Alharbi et al. They scanned the literature for randomized controlled trials (RCTs) of probiotics used to prevent TD and found 10 that met their criteria. The strains and high-level results are shown in the table below:

We’ll keep coming back to this meta-analysis, but rather than go over this table line by line, let’s dig into some of the probiotic strains.

Saccharomyces boulardii CNCM I-745

In the early 1920s, a French microbiologist named Henri Boulard was traveling through Southeast Asia when he noticed locals chewing on lychee and mangosteen skins to treat cholera-induced diarrhea (and, according to this company selling it as a tea, they also made tea out of it). Boulard wondered what it was in the fruit skins that might be responsible, and, a few years later, isolated the yeast Saccharomyces boulardii. (Presumably Saccharomyces nativeshavebeenchewingitformillenniaii wouldn't fit on the label.)

S. boulardii has since become one of the probiotic world's biggest success stories, with studies to back up its reputation. The specific strain that's been studied is S. boulardii CNCM I-745—and yes, those numbers and letters matter. Just as a Honeycrisp apple differs from a Red Delicious, different strains of the same species can have vastly different effects. Virtually all the research has been done on this particular strain, so when I mention S. boulardii from here on, that's the one I'm talking about.

A 1993 placebo-controlled, double-blind study by Kollaritsch et al. of 3000 Austrians traveling to what researchers called “distant regions” found that it significantly reduced TD. A 2010 meta-analysis by Lynne V McFarland described that study as follows:

Travelers were randomized to either 250 mg/d S. boulardii, 1000 mg/d S. boulardii or placebo, starting five days before leaving and throughout the duration of their trip (median of 3 wk). Only 1016 (34%) completed the study, but S. boulardii significantly reduced the incidence of traveler’s diarrhea in a dose-dependent manner. Of the travelers given placebo, 39% developed diarrhea whereas only 34% of those given the lower dose and 29% of those given the higher dose of S. boulardii developed traveler’s diarrhea (P < 0.05).

Going from 39% to 29% is statistically significant with a large enough sample size, yes, but it's not exactly the bacterial force field I was hoping for. Most travelers didn't get sick anyway, and of those destined for digestive drama, most still achieved it despite swallowing their capsules. Still, reducing your chances by about a quarter isn't negligible—for every four people who would have spent their vacation on intimate terms with foreign plumbing, one gets to actually enjoy the trip. Overall, McFarland declares, “The use of S. boulardii as a therapeutic probiotic is evidence-based for both efficacy and safety for several types of diarrhea.”

This, however, was only enough for Alharbi et al. to list S. boulardii among the probiotics that “may prevent TD”. I know, a ringing endorsement that ranks somewhere between “better than nothing” and “couldn't hurt”.

Lactobacillus rhamnosus GG (LGG)

The next stop is Lactobacillus rhamnosus GG (LGG). The "GG" stands for Gorbach and Goldin, the two researchers who discovered this particular strain. The table in the meta-analysis doesn’t call out which strain, but if you follow the references, you’ll find that they were, in fact, testing LGG.

The Probiotics Institute2 calls it "the world's most documented probiotic strain" with over 250 studies to its name. However, despite all of these studies, the meta-analysis only considered two to have met their criteria.3 Two studies instead of only one for S. boulardii is definitely an improvement, but to call these “well-studied”—at least for TD prevention in humans—would be a stretch.

The first RCT was Prevention of travellers' diarrhoea by Lactobacillus GG by Oksanen et al. (full text) published in 1990. Researchers followed 820 Finnish tourists vacationing in two resort towns along Turkey’s southern Mediterranean coast.

The results were, shall we say, underwhelming. In the placebo group, 46.5% got sick, while in the LGG group, 41.0% got sick—a reduction of about 12%. Further, the results were not statistically significant (p=0.065). In plain English, this means we can't rule out that the difference was just random chance.

The two resort towns showed different results. In one location, LGG seemed to work reasonably well, while in the other, it was mostly negligible.

What's going on? Well, there are two possibilities. Maybe the two towns had different bacterial environments—different local cuisines, different water sources, different E. coli strains, that caused the difference. Or maybe it's just statistical noise. My conclusion is that this study hints at a small protective effect, but even that isn’t conclusive.

The 1997 study by Hilton et al. (full text) took a different approach. Researchers recruited 400 healthy American adults about to embark on trips to Asia, Africa, and Latin America. Participants started taking either LGG or a placebo two days before departure and continued throughout their travels.

Of the original 400 participants, only 245 managed to follow the protocol correctly, leaving them with 126 people having taken LGG and 119 on placebo for a grand total of 2,743 travel days to analyze.

The results actually made LGG look good for once. The LGG group reported diarrhea on only 3.9% of their travel days, compared to 7.4% for the placebo group. That's about a 50% reduction.

However, this isn't an apples-to-apples comparison with the other studies. Instead of asking whether participants got diarrhea during the trip (a yes/no question), Hilton asked whether they got diarrhea each day.4 This methodology is better at detecting partial protection—if LGG reduces three days of diarrhea to one day, that shows up as a 67% improvement here, whereas in the Oksanen study, that person would still just count as 'sick.'5

Lactobacillus acidophilus

Lactobacillus acidophilus is everywhere in multi-strain blends, though after looking at the research, I suspect that it's primarily there to bulk up the bacteria count.

The meta-analysis by Alharbi et al. found three randomized, double-blind, placebo-controlled trials of L. acidophilus. In all three cases, the rates of TD were higher for the group that took L. acidophilus than for the placebo group. Some of the results weren’t statistically significant, but it seems like, in the best case, it doesn’t work, and the evidence points to it being counterproductive. Despite the meta-analysis saying “Lactobacillus acidophilus showed no efficacy in preventing TD except when mixed with other strains”, they still listed L. acidophilus as among the five probiotics that “may prevent TD”. Unless you're aiming for my explosive weight loss plan (patent pending), I'd save the money and skip this one entirely.

Multi-strain Blends

One thing you commonly see in probiotics products is multi-strain blends. The logic is simple: if one strain is good, surely a dozen must be better. Throw everything at the problem and something's bound to work. But does it?

Not really. Lynne V McFarland (the same researcher who commented on our S. boulardii study) published a systematic review in 2021 examining whether multi-strain probiotics outperform their single-strain counterparts. Her conclusion was, "In most cases, multi-strain mixtures were not significantly more effective than single-strain probiotics." The key, she argues, isn't the number of strains but whether any given strain has actual evidence behind it.

The Alharbi et al. meta-analysis lists five different probiotic strains that, again, “may prevent TD”: L. acidophilus, L. rhamnosus, L. fermentum, S. cerevisiae, and S. boulardii. Perhaps what’s going on is that the multi-strain blends are grabbing at least one of the strains that work, and the others are mostly there for marketing.

But a problem with multi-strain blends is that you don’t know how much of each strain you’re getting. Most products list the amount of total probiotics, not the amount of each individual strain. Chances are, you’re getting lots of the cheap ones and probably not as many of the good ones. A 2016 study by Morovic et al. found that many blended products consisted predominantly of Lactobacillus acidophilus and Bifidobacterium animalis subsp. lactis, two strains that have not been shown to work.

Best Selling Probiotics

After diving into the research, I'm a bit underwhelmed. The effect sizes aren't impressive; none of them work as well as I had hoped. We're talking about modest improvements, not miracle cures. But modest improvements beat explosive diarrhea in a foreign country. And it might vary by person. Hilton et al. noted that "the protective effect appeared to be more pronounced in patients with a prior history of traveler's diarrhea," which I choose to interpret as the universe's way of saying 'Yes, we mean you specifically.'

The top choice when you search “probiotic” on Amazon is Physician's CHOICE Probiotics 60 Billion CFU. The first thing to note about it is that S. boulardii and L. rhamnosus GG are completely missing! (See the appendix for why you won’t find LGG in the cheap multi-strain blends.) It claims “10 diverse strains”. Diversity might be good, but if they’re all not helpful in 10 different ways, it’s still not helpful. I would skip this one.

Amazon Basics has a probiotic. It has the following 8 probiotics: Lactobacillus acidophilus La-14, L. rhamnosus Lr-32, L. plantarum Lp-115, L. salivarius Ls-33 and L. bulgaricus Lb-87; Bifidobacterium lactis Bl-04, B. bifidum Bb-02 and B. longum Bl-05. It has L. rhamnosus, which is good, but, remember, it’s the strain that matters. The one in this product is L. rhamnosus Lr-32, not LGG. Similarly, it lists Bifidobacterium longum, which has been found to support digestive health. However, that finding only applies to the strain B. longum UABl-14, which is not the same as B. longum Bl-05, and this 2024 study by Shehata et al. shows that those two strains behave differently. I’d say skip it.

If you look through enough products, you’ll find some with S. boulardii in it, like this Probiotics 70 Billion CFU, but the problem is, you don't know how much S. boulardii is in it. Based on the Morovic study, these multi-strain formulas are probably mostly L. acidophilus, which does not seem to work. I think none of the top-selling products are what we’re looking for.

My Recommendations

I say stick with the two single strains that seem to have the best actual science behind them: Lactobacillus rhamnosus GG and Saccharomyces boulardii. Unfortunately, there are no studies that I could find that combine these two strains.

For LGG, your option is Culturelle Pro Strength Daily Probiotics6. (See the appendix for why there aren’t more options.)

S. boulardi, however, is easily available online, commonly sold as Florastor in the US. This is indeed S. boulardii CNCM I-745, the exact strain from the studies.7

You need to start taking these things two or three days before you leave. Your intestines need time to get acquainted with their new residents, to sort out the sleeping arrangements, as it were. Continue taking them throughout your trip.

Results

I wrote the above a month ago. Then, armed with Lactobacillus rhamnosus GG and Saccharomyces boulardii, I ventured to South India and Sri Lanka. I'm proud to offer you the results of my tiny, n=1 experiment: I got sick… thrice.

It's not all bad news, though. The two bouts each lasted only a day. The third lasted three days, though that estimate might be generous—by day four, my gut hadn't exactly recovered so much as negotiated a detente. We established a demilitarized zone that, unfortunately, surrounded all food. Peace held as long as I kept my distance, but the moment I reached for a samosa, hostilities resumed. Still, I came back only 11 pounds lighter instead of my previous 17, so... progress?

When I got sick, I started taking a double dose of probiotics, which may have helped my recovery. I also immediately abandoned any pretense of scientific experimentation and took Pepto-Bismol, which probably deserves most of the credit for my relatively quick recoveries.

My final thought is that the benefits of probiotics to prevent TD are probably real and somewhat meaningful. I think there's a good chance they helped, and if you, like me, often get sick when traveling, even a 25-50% reduction in TD is worth the price of admission. I'll be packing them again next time, right next to my Pepto-Bismol and my lowered expectations.

Remember, probiotics are just one layer of protection. These pills might reduce your risk by 25-50%, but probiotics are no more an invitation to drink straight from the Ganges than carrying bear spray is an invitation to challenge a bear to a wrestling match. It ain’t gonna save you. Most importantly—and I cannot stress this enough—no matter how persuasive the man with the jar, no matter how much he insists it's a local delicacy, do not, under any circumstances, drink the cobra alcohol.

Appendix

Can probiotics be dangerous?

Some bacteria can kill you, of course, but those aren’t the ones being sold as probiotics. For the ones on the supermarket shelves, the track record seems to be that they are very safe. If you look, you’ll find that they are listed as GRAS (Generally Recognized As Safe) by the FDA and/or granted QPS (Qualified Presumption of Safety) by the European Food Safety Authority (EFSA). The Mayo Clinic says side effects are rare and probiotics are safe.

If you want to dig deeper, you can search this FDA site for GRAS notices. Here is the FDA notification of GRAS for LGG.

I think the risk with probiotics is that they waste your money, not that they harm you physically.

Does the right probiotic depend on where I’m going?

My guess is the answer is “yes”, but nobody's done the research to match specific probiotics to specific destinations. We do know that different regions harbor different nasty bacteria. Latin America is E. coli territory (second reference), Southeast Asia is relatively higher in Campylobacter jejuni (second reference), and South Asia serves up mostly E. coli with a side of everything else. Logic suggests that different probiotics might work better against different bacteria, but so far, researchers have only tested against 'traveler's diarrhea' in general.

Why are there probiotics for men and women?

Short answer: marketing.

While there are certainly differences in gut microbiota between men and women, these variations appear to be relatively minor and not significant enough to require completely different probiotic formulations. The scientific literature I’ve seen on TD prevention doesn't show meaningful sex-based differences in probiotic effectiveness.8

So while I don’t think there’s anything wrong with taking one marketed towards men or women, I tend to avoid both because it’s an indication that the marketing department has too much influence into the product, although if you find one that is marketed to a specific sex, it should be fine.9

How are they shelf-stable and living? Are they growing?

For the curious, if you’re wondering how billions of supposedly living organisms can sit in a bottle on a shelf for months without food, water, or refrigeration, it’s pretty cool science. We’re told they’re living, right, so are they reproducing in there like some kind of microbial terrarium?

The answer is that they’re alive, but in a state of suspended animation. In their dried, powdered form, there's almost no free water10 available, which forces them into what is called an "anhydrobiotic state." They essentially shut down all operations: no eating, no reproducing, no energy production. They're alive in the most technical sense, but doing absolutely nothing. This dormancy isn't permanent, though. Heat and oxidation still slowly damage the cells, which is why probiotics still expire.11

The real magic happens when you swallow them. Once they hit the moisture and warmth of your digestive system, they wake up like microscopic Sleeping Beauties. Suddenly surrounded by water and nutrients, they resume normal bacterial activities—metabolizing, dividing, and hopefully crowding out the less friendly microbes.

Why do I need to keep taking it? Shouldn’t a single batch be enough to colonize?

Not necessarily. England's first attempt to colonize North America—the Roanoke Colony—collapsed within a year from food shortages and hostile locals. A second attempt vanished entirely, leaving behind only the word "CROATOAN" carved on a post.

Oh, wait, you were probably asking about probiotic colonization of the gut. In that case, they don’t permanently colonize you. They pass through the gut, help balance things while they’re present, then fade once you stop taking them. Though unlike the Roanoke colonists, they don't leave cryptic messages carved in your intestinal walls.

Why don’t I just treat the symptoms?

You could also skip the probiotics entirely and just treat the symptoms when they arise. Imodium and Pepto-Bismol are the classic choices here. Interestingly, you can actually take Pepto-Bismol preventively—two tablets four times a day—but many people don’t like the side effects (turning your tongue/stool black, potential constipation, etc.). Most people decide they'd rather risk it. The downside of waiting, of course, is that once you're sick, these medications are damage control, not a cure. You're managing the disaster, not preventing it.

Note that Pepto Bismol is much harder to find outside of North America, so if you do intend to use it, I recommend bringing it with you.

Why you won’t find LGG in Cheap Probiotics

One thing you won’t see in any of the multi-strain blends on Amazon is LGG, despite it being "the world's most documented probiotic strain." This ended up being its own rabbit hole, but I will keep it very short. The original patent for LGG was granted in 1989 and expired in 2006. LGG has also been trademarked, and all over documents, I see the phrase “LGG^® is a trademark of Chr. Hansen A/S”. (See, for example, where they announced their partnership with celebrity Jessica Alba.) That’s because in 2016, the company Chr. Hansen purchased the trademark for LGG for €73 million.

Anyway, I think the upshot of this all is that anyone can use the organism, but if you market LGG in your product, you owe Chr. Hansen money. So you’re not likely to see it in the lower-cost “kitchen-sink” approaches. The Culturelle brand (owned by the multinational company dsm-firmenich12) holds the primary U.S. retail license, which is why they advertise as “the only probiotic supplement with 100 % LGG.” As in, they’re willing to hand over some money to Chr. Hansen.

A Note on Industry Influence

Now that we’ve met Chr. Hansen, it’s time to return to our source of unbiased scientific information that we learned so much about LGG from: The Probiotics Institute. It turns out, funnily enough, that this institute that has so much good stuff to say about LGG happens to be—coincidentally, I’m sure—owned by Chr. Hansen.

This… is not ideal.

I'm going to avoid the tangent on whether industry funding undermines the science here (future blog post coming on that), and simply say: despite these conflicts of interest, I think LGG is still worth trying.

Though our primary source has been the meta-analysis by Alharbi et al., I kept running into results from The Probiotics Institute every time I researched LGG. Meta-analyses follow the GOGI principle—garbage in, garbage out. If you take a bunch of biased studies and average them, the bias doesn’t go away. You just become more confident in a biased answer.

Perhaps there's publication bias at play here due to these conflicts of interest: positive results get published while negative ones gather dust in file drawers, making treatments look more effective than they really are. Given the lack of independent institutions and the fact that probiotics is a multi-billion dollar industry, it’s reasonable to wonder.

Fortunately, there are ways to sniff out this kind of bias. If negative studies were being systematically buried, we'd expect to see certain telltale patterns in the published data—specifically, a suspicious lack of small studies showing negative or minimal effects.

A 2018 meta-analysis by Jong-Myon Bae looked for exactly this problem. He examined only the double-blind, placebo-controlled, randomized controlled trials and created a funnel plot of their results. If studies were being hidden, the plot would look lopsided, like someone had taken a bite out of it.

Starting with 1,227 papers on probiotics, Bae excluded those that didn't meet his criteria. Just requiring that they be RCTs knocked out over two-thirds of them. In the end, only 11 met his standards (most of which overlap with the meta-analysis by Alharbi et al.).

This is an annoyingly small number. There’s all this hype about how studied these probiotics are—LGG has over 250 clinical study publications, Bifidobacterium animalis subsp. lactis BB-12 has over 300 scientific publications, according to a study by employees of Chr. Hansen (who also owns BB-12)—yet so few meet the meta-analysis criteria for inclusion.

One takeaway is that much of the analysis is garbage not sufficient to understand the effects in humans. The other takeaway is that maybe these meta-analyses did a good job sorting through all the mess, and we should trust them more. However, with only 10 or 11 good studies across all strains of probiotics, we shouldn’t be too convinced by our results.

Anyway, Bae made the following funnel plot and, fortunately, it does not show the telltale signs of publication bias.

So despite Chr. Hansen's empire-building, the evidence that does exist appears to be honest, just frustratingly sparse.

These questions about human behavior are perplexing. What's more perplexing is that they strike us as perplexing. We're humans describing human behavior—shouldn't we be able to explain ourselves? Yet when you ask people the reasons why, their explanations often don't quite add up.

In The Elephant in the Brain, authors Kevin Simler and Robin Hanson offer a unified theory for these puzzles. The authors—a software engineer and an economics professor known for prediction markets and his explanation of the Fermi Paradox—argue that much of human behavior is driven by ugly, self-interested motives that we'd rather not acknowledge, even to ourselves. These hidden drives shape everything from politics to charity, from medicine to religion. The "elephant in the brain" is this massive, uncomfortable truth: we're not nearly as noble as we pretend to be, and our minds actively hide this fact from us.

Here's their thesis in their own words:

We, human beings, are a species that’s not only capable of acting on hidden motives—we’re designed to do it. Our brains are built to act in our self-interest while at the same time trying hard not to appear selfish in front of other people. And in order to throw them off the trail, our brains often keep “us,” our conscious minds, in the dark. The less we know of our own ugly motives, the easier it is to hide them from others.

Self-deception is therefore strategic, a ploy our brains use to look good while behaving badly. Understandably, few people are eager to confess to this kind of duplicity. But as long as we continue to tiptoe around it, we’ll be unable to think clearly about human behavior. We’ll be forced to distort or deny any explanation that harks back to our hidden motives. Key facts will remain taboo, and we’ll forever be mystified by our own thoughts and actions. It’s only by confronting the elephant, then, that we can begin to see what’s really going on.

Again, it’s not that we’re completely unaware of our unsavory motives—far from it. Many are readily apparent to anyone who chooses to look. For each “hidden” motive that we discuss in the book, some readers will be acutely aware of it, some dimly aware, and others entirely oblivious. This is why we’ve chosen the elephant as our metaphor. The elephant—whether in a room or in our brains—simply stands there, out in the open, and can easily be seen if only we steel ourselves to look in its direction (see Figure 1). But generally, we prefer to ignore the elephant, and as a result, we systematically give short shrift to explanations of our behavior that call attention to it.

The first time I read this book, I thought, "Clearly true. No notes!" So I didn't write a review. But then I kept finding myself in conversations trying to explain the thesis, only to discover all my supporting examples had evaporated from memory. After enough iterations of this, I figured it was time to reread and properly document the best arguments.

The core insight is that we're not just blind to our motives—we're strategically blind. Natural selection didn't design us to be honest; it designed us to look honest while pursuing our interests. And the best way to look honest is to believe your own spin. Hence, the Feynman quote for actually seeking truth: “The first principle is that you must not fool yourself—and you are the easiest person to fool.”

Evolution made us political animals. As a highly social species, our survival has always depended on maintaining good reputations and strong alliances. To navigate these social hierarchies, we rely heavily on signaling. Signaling exists throughout nature—a peacock's elaborate tail announces genetic fitness, a lion's roar warns rivals to stay away. Chimpanzees constantly signal their place in the status hierarchy through distinct vocalizations, submissive or dominant body postures, and strategic grooming partnerships1. Human signaling contains all of this and more. We're not just living our lives; we're constantly broadcasting messages about ourselves. Every day, through countless behaviors, we communicate who we are, what groups we belong to, and why others should value us.

We signal all sorts of things: our wealth, intelligence, health, social connections, group loyalties, moral virtues, and quality as a mate. We signal our love and devotion to those we care about. We signal what reliable, generous friends we are. These aren't just nice gestures; they're advertisements for our value as allies. We also signal our willingness to retaliate when crossed, showing we're not to be trifled with. Basically, we signal anything that might make us a more attractive friend, ally, or partner—or a more formidable enemy.

Let’s return to the close friends who insult each other—what’s that about? Insults that would shatter a casual friendship instead become signals of trust. The very harshness of the words says, “Our bond is so secure, I know I can say things to you that would end a lesser friendship.” It works precisely because it would be destructive coming from anyone else. This phenomenon is called countersignaling—demonstrating closeness or status by forgoing the usual signals of kindness or respect.

Signaling is a world of complexity in itself. The same ripped jeans send different messages depending on who’s wearing them. On someone genuinely struggling financially, the holes might signal nothing—they're simply making do with damaged clothes. On a teenage punk rocker, those same rips become a deliberate statement: "I reject mainstream values." But on a wealthy fashionista, they become countersignaling: "My status is so obviously high that no one could mistake these designer rips for poverty. I can afford to look tattered precisely because everyone knows I don't have to."

We could go deeper into the world of signaling, but the point is, it’s an important part of being human.

Some signals are socially acceptable to acknowledge—we freely admit we dress nicely for job interviews or buy flowers for others to show our affection. But others must remain hidden.

Take status. In our ancestral environment, high status meant preferential access to resources, better mates, and allies who would defend you. Low status could mean starvation, exile, or death. Given these stakes, we're driven to climb the hierarchy.

But if everyone openly pursued status, social life would become an exhausting tournament of people constantly jockeying for position. So we developed norms: don't boast, be humble, let your work speak for itself. These norms keep the peace by suppressing destructive competition and promoting cooperation.

Violate these norms—say, by introducing yourself with a list of accomplishments—and people won't like you. You'll lose status.

Thus, we need to signal qualities that we're forbidden from advertising directly. So we've evolved elaborate workarounds. Consider the LinkedIn humblebrag: "Humbled and honored to receive this award!" The social rules say I can't brag about my achievement, but I can express gratitude for it. If you happen to notice I won an award and think I'm impressive, well, that's on you. I was merely expressing my humility.

Or consider conference introductions. If I want the audience to know how accomplished I am, I can't just walk on stage and list my accomplishments—that would be gauche. But notice how we've developed a workaround: having someone else introduce me. There’s no social norm against standing slightly offstage while someone else comes on stage and lists my accomplishments. The information gets transmitted, the audience is impressed, but technically, I never bragged. Our social norms create elaborate loopholes for the very behaviors they supposedly prohibit.

This dance between needing status and being unable to pursue it openly created an evolutionary pressure for deception.

And the best way to look like you're not seeking status is to genuinely believe that you're not.

Thus, the need for self-deception: our brains evolved to hide our status-seeking agenda from our conscious awareness, letting ourselves pursue selfish goals while believing we're driven by noble motives. No one says, "I want this handbag so people will know I'm better than them." They say, "It's so cute! I need this handbag." And if it just so happens to prominently feature a brand name that signals wealth, well, I tell myself I don’t care about that—I just liked the pattern.

This dynamic plays out everywhere. Ask someone why they want to be a doctor:

• Socially acceptable: "I want to help people."

• Less acceptable: "The pay is excellent."

• Unthinkable: "It's a prestigious job and I'll have higher social status and get to mate with someone else of high status."

That last one might be a factor, but it’s not something someone can go around saying and still have a lot of friends. In fact, so much of this is done at the subconscious level that Simler and Hanson say not to think of ourselves as the executives of our own minds, but as the press secretaries, crafting plausible stories for public consumption while the real decisions get made in neural backrooms our conscious thought can't access.

Many of our behaviors happen subconsciously. You've probably noticed that when people walk together, they match speeds. But at what speed do they end up? The faster person’s, the slower person’s, or somewhere in between? Actually, research shows that it's the lower-status person who adapts to the pace of the higher-status person2. Your brain handles this calculation automatically, understanding social hierarchies and adjusting your gait accordingly. You've likely been speeding up or slowing down to match others based on unconscious social rankings your entire life and you never even knew it was happening (or, at least, I had no idea).

We rarely do things for just one reason—our motivations are layered. Just because the expensive handbag gives you status doesn't mean it doesn’t also carry Chapstick. Both purposes are real. The signaling doesn't cancel out the practical use, and the practical use doesn't mean the signaling isn't happening. Many human behaviors work this way, accomplishing multiple goals at once.

So how can we untangle these mixed motives? Here's a simple thought experiment that can sometimes help: How much would you value something if no one else could know about it? Take a Harvard diploma. People go to college to get an education, but they also go so they can tell everyone they got an education. How much would a Harvard diploma be worth if you couldn’t tell anyone about it? If all you got was the education itself, but had to keep your Harvard connection secret forever. My guess is that most people would value it at less than half, which tells you something about what that tuition is really buying—not just knowledge, but signals of intelligence, ambition, and the kind of hoop-jumping ability that employers prize.

The book works not by rigorous scientific proof, but by pointing at a plethora of behaviors and saying, “Look over here. Here’s an example. Here’s another. Do you see it?” Not every example will convince every reader—you might think some behaviors have less to do with hidden motives than the authors claim. But, to me, the core thesis feels rock solid: we're political animals pretending to be truth-seekers, status-seekers pretending to be altruists, self-deceivers by design rather than accident.

The authors approach these subjects without moralizing or judgment. These aren't character flaws to fix but features that helped our ancestors navigate complex social worlds. This isn't a book about what "bad" people do—it's about what we all do. As the authors make clear:

The line between cynicism and misanthropy—between thinking ill of human motives and thinking ill of humans—is often blurry. So we want readers to understand that although we may be skeptical of human motives, we love human beings. (Indeed, many of our best friends are human!) We aren’t trying to put our species down or rub people’s noses in their own shortcomings. We’re just taking some time to dwell on the parts of human nature that don’t get quite as much screen time.

The authors readily admit they see the elephant even in their own behavior:

We, your two coauthors, can also give examples from our own lives. Robin, for example, has often said his main goal in academic life is to get his ideas “out there” in the name of intellectual progress. But then he began to realize that whenever he spotted his ideas “out there” without proper attribution, he had mixed feelings. In part, he felt annoyed and cheated. If his main goal was actually to advance the world’s knowledge, he should have been celebrating the wider circulation of his ideas, whether or not he got credit for them. But the more honest conclusion is that he wants individual prestige just as much as, if not more than, impersonal intellectual progress.

Once you start seeing the elephant, you can't unsee it. It appears in our jokes, our shopping habits, our career choices, even our most intimate relationships. The first half of the book describes the idea, which I’ve summarized here. The second half explores these hidden motives across different domains of life. Here are the examples I found most interesting and compelling.

Sections

Laughter

Stop and think about laughter for a second. Really think about it. Why do we convulse and make barking noises when amused? We say we laugh because something is funny, but is that really all that's going on? Why, then, do sitcoms need laugh tracks? Why do we chuckle nervously when making a risky comment?

Simler and Hanson point to research that shows we laugh roughly thirty times more when other people are around, and that speakers laugh about 50% more than listeners. Additional studies show that only 10-20% of laughter follows anything resembling a joke. Watch any conversation—people laugh when slightly embarrassed, when changing topics, when softening criticism. "Your presentation was... interesting [nervous laugh]." We might think laughter is about humor, but how do we square that with the data?

Laughter serves many roles, but a primary purpose is to communicate a "play" signal. It's a way to say, "We're just playing here; nothing is genuinely threatening or hostile." This would have been crucial for our ancestors—play fighting helps young mammals practice real combat skills, but without clear signals, practice could escalate into actual violence.

What makes laughter especially useful as a signal is that it's largely involuntary. You can lie with words, craft whatever message you want, but a genuine laugh is hard to fake convincingly. This involuntary nature is exactly why we evolved to trust it—it's a more reliable indicator of someone's true state of mind than their carefully chosen words.

The signal works both ways: we laugh at our own edgy remarks to say "just kidding," and at others' antics to say "I know you're kidding." This back-and-forth creates a feedback loop where context determines humor more than content. A joke that kills at a bar might bomb at a funeral. Take grandma falling down the stairs. Is that funny? By itself, no—it could be tragic. But if grandma gets up laughing (signaling "I'm okay, this is play, not danger"), suddenly it becomes hilarious. Her laughter gives us permission to laugh. We're not really laughing at what happened—we're responding to the invisible social currents that tell us whether this is a "play" moment or a "serious" one.

But laughter does more than just signal safety—it's also a tool for probing social boundaries. Consider how comedians test controversial material: they'll make an edgy observation about race or politics, then gauge the audience's response. If people laugh, the boundary-pushing was accepted; if not, they can retreat behind "just a joke!" This creates a cycle of tension and release—the danger of violating a norm followed by the relief of pulling back from it.

In everyday conversation, we use the same technique on a smaller scale. We test the waters with a half-joke about our boss, flirt through teasing, or challenge authority with "harmless" mockery. If our probe gets a laugh, we've successfully pushed the boundary. If it falls flat or draws glares, we were "obviously joking."

Of course, not all laughter fits this model perfectly. Sometimes we do just find things genuinely surprising or absurd. But the pattern holds remarkably well, and the hidden agenda explains why humor is so hard to define and why jokes die when explained. Making these implicit social negotiations explicit ruins their entire function. It also explains the research showing that speakers laugh more than listeners—they're not just responding to humor, they're actively using laughter to manage the social situation.

This is the elephant in the comedy club: We tell ourselves we laugh because things are funny, but we're actually engaged in complex social maneuvering. Our brains hide this from us—imagine how awkward it would be if every time you laughed, you consciously thought, "I'm signaling group cohesion and testing social boundaries." The self-deception is crucial. The person who laughs most naturally, who seems least calculating about it, is paradoxically the most effective at using laughter for these hidden purposes. We can't admit, even to ourselves, that our spontaneous joy is partly a political tool.

Conversation

Why do we talk? Why do we have conversations? You might say to communicate information, and, of course, that’s part of the reason. But this explanation is hard to square with many of our experiences. If conversation were truly about information transfer, we should prefer listening—after all, the listener gains information while the speaker gives it away. Yet people like to talk and even compete to have their voices heard. People interrupt and talk over each other. We need social norms to remind us to let others speak, and even while listening, we're plotting what to say next.

Clearly, it's not just about gathering information.

And if it were, why would there be all these weird social protocols about relevance? You could just state whatever had the most informational value. But abruptly introducing unrelated facts is considered rude, even if the information is valuable. The authors offer some sample dialogue to show how bizarre conversation would be if it were merely about exchanging information:

A: FYI, Alex and Jennifer are finally engaged.

B: Thanks. Have you heard that the President is trying to pass a new healthcare bill?

A: Yeah, I already knew that.

B: Oh. In that case, um … a new Greek restaurant just opened on University Avenue.

A: That's new information to me. Thanks.

This is not how conversations go. Real conversations meander from topic to topic, each remark building on the last. I once saw a headline from The Onion that captured this perfectly (something like): "Friends Gather To Tell Tangentially Related Anecdotes."

Further, if conversations were truly about information exchange, meeting someone new should be the most exciting opportunity—they haven't heard any of your best stories or insights yet. You'd jump straight to your most valuable knowledge. But instead, we make small talk about the weather.

What’s going on here?

The book posits that every spoken remark contains both "text" (the explicit information) and "subtext" (an implicit message about the speaker, such as "I'm the kind of person who knows such things"). Listeners, while interested in the text, are also keenly attuned to the subtext, constantly evaluating the speaker's value as a potential friend, ally, lover, or leader.

Think of everyone as having a backpack full of tools—stories, knowledge, experiences—that demonstrate their value. Each sentence shows off a tool from your backpack. Anyone can have random tools (unrelated anecdotes), but if you have exactly the right tool for this moment (a perfectly relevant story that showcases your wit, knowledge, or experience), that's much more impressive. The relevance constraint makes showing off harder, and therefore more credible.

This explains why we need conversational hooks, perhaps some recent events or shared contexts, to showcase our knowledge. Without them, we sound like we're randomly bragging. It also explains why conversational debt flows the opposite direction from what you'd expect. If conversation were about information exchange, the debt would be "they gave you information, now you owe them information." But the actual debt is "they gave you the floor, now give them their turn."

Beyond individual display, conversation also helps us find our tribes. When someone catches your obscure reference, shares your niche hobby, or laughs at the same things, it signals deeper compatibility. We're not just showing off our tools; we're finding people whose backpacks contain similar items—people who share our knowledge base, values, and ways of thinking.

This hidden agenda explains many conversational mysteries: why we often rather talk than listen, why relevance matters so much, why prepared remarks feel like cheating, and why good conversationalists know when to strategically yield the floor. We're not only exchanging information; we're performing an elaborate dance of self-presentation and tribal identification, all while pretending we're just having a chat.

Consumption

Keynes once predicted that as society grew wealthier, we'd work less, having solved the problem of scarcity. It hasn't panned out exactly as he thought. The authors attribute this miscalculation to Keynes overlooking the rat race of competitive signaling. Despite material abundance, we still compete fiercely through consumption—not for the practical utility of goods, but for the limited supply of social status they help us attain.

We all know that consumption is often conspicuous, but this doesn't just apply to Rolex-wearing show-offs. It's all of us, even those who think we're above it. The authors offer some thought experiments:

If you’re a high-powered executive, imagine wearing your old high school backpack to work. If you’re a bohemian artist, imagine bringing the Financial Times to an open-mic night. If you’re a working-class union member, imagine ordering kale salad with tofu at a restaurant.

The discordance in these scenarios reveals how much our consumption choices are about fitting in and signaling group membership.

Take the Prius, a textbook example of conspicuous consumption dressed up as its opposite. Studies show that people pay a premium for the Prius—even when you account for the savings on gas—specifically for its distinctive look that signals environmental consciousness. They're not just saving fossil fuels; they're broadcasting their altruism, showing they're the kind of person who cares about the planet. The premium they pay is essentially the cost of that signal.

Our cars often do a lot of signaling. A Tesla once signaled being hip, tech-savvy, and environmentally conscious. But after Elon Musk's political turn, that same Tesla might now signal political allegiance—a meaning many owners never intended. Hence the proliferation of bumper stickers proclaiming "I BOUGHT THIS BEFORE WE KNEW ELON WAS CRAZY." These stickers are essentially saying, "Please ignore the new signal my car sends." This attempted damage control only makes sense in a world where our purchases broadcast messages about who we are.

We also like to tell backstories for our purchases: "I got it on sale," "It’s ethically sourced," "We found it at this little vintage shop". This lets us fine-tune what our possessions say about us. They signal that we have a keen taste for style or know how to sniff out a good deal. Whatever the details, it means we’re a good ally to have.

This isn’t just for material goods—it applies to travel and activities as well. Let’s go back to our earlier question: How much would you pay for that vacation if you couldn't tell anyone about it—no photos, no stories, no proof you went? Would you still pay for that climbing gym membership if you couldn’t add a photo of yourself conquering an overhang to your dating app profile? Some might pay the same amount, valuing the experience itself. But for many, if we're being honest, the price would drop significantly.

Here’s a thought experiment to get a sense of the significance of all of this conspicuous consumption. Imagine if we magically couldn't form impressions from possessions. How would it change us? Would anyone buy clothes beyond basic weather protection? Would anyone maintain their lawn beyond basic upkeep? Think of all the industries that would collapse overnight: luxury goods, fashion, cosmetics, high-end restaurants, most car upgrades. The global economy would shrink dramatically if we stopped buying things to impress others.

Also, notice how little advertising actually says about the product itself—instead, it's all lifestyle and image, selling you on who you'll become by buying it.

We tell ourselves it's about quality and personal taste, but we're often just playing the ancient game of status with modern props.

Art

It will come as no surprise when I tell you that Hanson and Simler think that art is not about pretty pictures. It doesn't take a skeptic to look at a Jackson Pollock and think, "Something else is going on here besides 'I had a blank wall and $30 million burning a hole in my pocket.'"

While most people suspect there's more to art than pure aesthetics, we still tend to fall back on the idea that art's value is intrinsic—that it's about the artwork and the feelings evoked by looking at it. But this notion quickly falls apart when we examine how art actually works in practice.

The authors illustrate this with a thought experiment:

Imagine that one of your friends, an artist, invites you over to see her latest piece. “It’s a sculpture of sorts,” she says. “Smooth swirls punctuated by sharp spikes. Rich pinks and oranges. Pretty abstract, but I think you’ll like it.” It sounds interesting, so you drop by her workshop, and there, perched on a pedestal in the center of the room, is the sculpture. It’s a delicate seashell-looking thing, and your friend is right, it’s beautiful. But as you move in for a closer look, you begin to wonder if it might actually be a seashell. Did she just pick it up off the beach, or did she somehow make it herself? This question is now absolutely central to your appreciation of this “sculpture.” Here your perceptual experience is fixed; whatever its provenance, the thing on the pedestal is clearly pleasing to the eye. But its value as art hinges entirely on the artist’s technique. If she found it on the beach: meh. If she used a 3D printer: cool. And if she made it by manually chiseling it out of marble: whoa!

The authors contend that art—both creating and appreciating it—is primarily about signaling desirable qualities about ourselves, not pure aesthetic pleasure. Art functions as "honest signaling" because creating quality art requires genuine skill, talent, time, and effort that can't be faked. Similarly, developing the taste to appreciate "difficult" art requires real investment in learning and cultural exposure.

This explains why owning art works as a status symbol. Expensive art obviously signals wealth. But art also signals intelligence, education, and cultural capital through demonstrations of "sophisticated" taste. There's social pressure to appreciate abstract or challenging art precisely because doing so marks you as part of an exclusive, discerning group.

Consider this scenario: You buy an artwork believing it will showcase your refined taste—something that might earn a New Yorker writeup praising its "phlegmatic allure" and noting how it "appeals more to cognoscenti than to popular audiences." You gladly pay millions for this badge of sophistication. But after walking out, you discover it's just leftover poster board from a construction site. You've been duped.

You’re embarrassed. But why? If art appreciation were truly about intrinsic beauty or personal feeling, you'd shrug—after all, you genuinely liked looking at it. But you're mortified because art isn't really about private appreciation. It's about what your taste says about you to others. You wanted people to see you as someone with a discerning eye, and now you look like a fool.

The elephant in the art gallery is this: We're not only buying beauty or meaning or emotional resonance. We're also buying proof of our sophistication, wealth, and cultural belonging. The art itself is almost beside the point—what matters is what owning, creating, or appreciating it says about us.

Below are two images. One of these is actually a piece of ripped poster board. The other is the one written about in the New Yorker.

Charity

That charity is about something other than helping people in the most effective way possible is patently obvious to anyone who’s seen the cold reception effective altruism has gotten in many circles. Any time an effective altruist suggests we rank causes by lives saved per dollar, the room goes chilly: “Now is not the time for spreadsheets,” the response is. That reflexive discomfort is a clue. If the real point of giving were simply to help as many people as possible, a spreadsheet would be useful, not threatening. Yet this is not how people react. Indeed, people don’t seem to investigate the ultimate impact of their donations very much. A survey Simler and Hanson cite found that only three percent of donors do comparative research before giving. Clearly, we have something else in mind.

Before we continue, I want to reemphasize an important point: As with most things, we donate to charity for multiple reasons. One is to help others. Nothing I've said or will say contradicts this. However, Hanson and Simler point out that another reason is to be seen helping others.

There is clearly some signaling going on in our charitable endeavors. Is anyone really surprised that men donate more when the solicitor is an attractive woman, or that people give more when their donations are public versus anonymous?

Consider all the infrastructure built around making charity visible: blood drives hand out "I GAVE TODAY" stickers so donors can parade their virtue to their classmates. Charity galas exist when we could simply mail checks. Donation pages show recent givers' names scrolling by. We prefer sponsoring a friend's charity run over quietly donating the same amount. The phenomenon is too systematic to call accidental: the mechanics of giving are often engineered for visibility.

Another clue: sometimes, it seems the goal is to donate, not necessarily for any impact. When Princess Diana died, people rushed to give money because donating was how they could express their grief and show their love for her. As the authors note:

Occasionally, we’re even happy to donate without knowing the most basic facts about a charity, like what its purpose is or how donations will be spent. “Within two weeks of Princess Diana’s death in 1997,” writes Geoffrey Miller, “British people had donated over £1 billion to the Princess of Wales charity, long before the newly established charity had any idea what the donations would be used for, or what its administrative overheads would be.”

The authors cite economist James Andreoni's "warm glow" theory of giving: we donate because it makes us feel good. That's why instead of setting up one automated payment to a single charity, we scatter small donations across multiple causes, collecting little hits of satisfaction.

But this only raises a deeper question: Why does giving trigger that warm glow?

The authors argue that it's evolution's reward for successful signaling. Each donation broadcasts multiple attractive qualities: wealth (look at these resources I can spare), generosity (see how freely I give), and most crucially, that we're reliable allies (I'll be there when you need help).

Much of this would be socially unpalatable to admit aloud. Telling your date, “I volunteer at the soup kitchen because it makes me look caring,” would vaporize the very status you are trying to amass. So the observing brain rewards the act with a surge of warm emotion, while the storytelling brain re-casts the deed as pure, undiluted altruism. The elephant hides the prestige calculus from conscious view. Only by lining up the data does the hidden agenda reveal itself.

None of this denies that charity can really help people. It does; it can radically improve people’s lives. But the form our giving takes—the causes we choose, the rituals we stage, the insistence on being seen—often has less to do with efficacy than with social theater. Donations are currency in a status market: we spend them not just to move money from richer to poorer, but to move admiration from observers to ourselves.

Education

The "hiddenness" of the elephant varies by topic. Some elephantine motives are deeply buried—like the self-serving reasons for charitable giving—while others lie barely beneath the surface. To my eye, education falls into this latter category. Probe even gently and most people will admit they're in it for the signal. "I went to college to get a job" is hardly a shocking confession.

The conventional story says people go to school to get an education. But several behaviors undermine this narrative. First, free education is stunningly available. A decade ago, we had MOOCs; now, ChatGPT can answer virtually any question. The world's best course materials sit freely online. And what professor would kick out someone quietly auditing their class? Yet demand for expensive college degrees remains sky-high, while demand for the actual learning—stripped of credentials—barely exists.

The contradictions pile up: Students take easy-A classes, celebrate canceled lectures, and use ChatGPT to do their homework. Graduates promptly forget most of what they learn and nobody—not students, schools, or employers—seems particularly bothered by this mass amnesia. Most damning is the "sheepskin effect"3: that final year of high school or college delivers a 30% salary boost, while other years yield just 4%. Senior year doesn't pack in seven times more learning. Employers simply care about the diploma.

Hanson and Simler argue that school primarily signals pre-existing ability. A Harvard degree announces: "I'm the type of person who can get into Harvard." It demonstrates your willingness to jump through hoops—SATs, essays, exams, papers, deadlines—precisely the skills employers value in corporate environments.

This explains why the diploma itself carries such disproportionate value compared to years of attendance. It signals not just intelligence, but persistence: the capacity to see tedious projects through to completion, to meet arbitrary requirements without complaint, and to deliver results on schedule even when motivation wanes. The credential proves you're someone who finishes what they start.

Really, we don't need mountains of evidence for the signaling theory. Ask most people: if you could have either the education or the diploma, which would you choose?4

Medicine

The elephant in medicine is harder to spot than in the other topics. There's no single thought experiment that makes you slap your forehead and go "Oh, of course!"

Conventional wisdom says medicine exists to heal the sick. Hanson and Simler agree, but argue we're missing part of the story. They say medicine is also about "conspicuous caring," an elaborate performance where we play concerned caregivers and grateful patients. Or, in their words, medical care serves as "an elaborate adult version of 'kiss the boo-boo.'

They offer an anecdote to illustrate, quoting from the public records of the physicians treating King Charles II when he fell ill:

His Royal Majesty was forced to swallow antimony, a toxic metal. He vomited and was given a series of enemas. His hair was shaved off, and he had blistering agents applied to the scalp, to drive any bad humors downward.

Plasters of chemical irritants, including pigeon droppings, were applied to the soles of the royal feet, to attract the falling humors. Another ten ounces of blood was drawn.

The king was given white sugar candy, to cheer him up, then prodded with a red-hot poker. He was then given forty drops of ooze from “the skull of a man that was never buried,” who, it was promised, had died a most violent death. Finally, crushed stones from the intestines of a goat from East India were forced down the royal throat.

You might object to the medical quality of the treatment, but nobody could say his doctors didn't try hard. The elaborate procedures signalled to everyone that they were doing everything possible to save the king.

This isn't just about medieval quackery. Modern medicine shows the same pattern: we often care more about appearing to help than actually helping.

To show this, they point to the weak returns of the marginal dollar spent on health care. They admit that the first dollars buy crucial interventions—antibiotics, vaccinations, setting broken bones, etc. But the last dollar often yields minimal health benefits: that extra scan "just to be safe," the brand-name drug that works no better than the generic, the aggressive end-of-life interventions that extend suffering more than life.

They point to the famous RAND health insurance experiment conducted from 1974-1982. The study was of 5,800 non-elderly adults, where each was randomly assigned a different level of medical subsidy, ranging from a 5% discount to a full subsidy. Unsurprisingly, those who didn’t pay for health care consumed much more, about 45% more than the least subsidized group. Perhaps more surprisingly, though, despite the large differences in medical consumption, the study found “almost no detectable health differences across these groups.”5

The Oregon Medicaid health experiment and numerous other studies tell the same story: beyond a basic level, additional medical spending produces negligible health improvements.

This is surprising—I certainly wouldn’t have guessed that the impact would be negligible. However, there could be lots of reasons for this, and I still consider it only moderately suggestive evidence of an elephant. I find that the case strengthens when we consider everyday examples.

Consider our behavior when someone falls ill. Why do we bring them food? Because they’re weak and we want to help them get better, obviously. OK, but why is it often so important that it’s homemade food? We bring homemade chicken soup, not because Campbell's lacks nutrients, but because stirring a pot for an hour says "I love you" louder than a can opener.

When someone near and dear to us is sick, we might demand the "best" specialists, even if that makes no difference. We want to show we're the kind of people who spare no expense when loved ones suffer. We (at least, some people6) insist doctors "do everything possible," even if it’s not particularly efficacious.

Why all of this? The answer seems to be because you want to show you care. You want to signal that you’re a good ally to have.

I think it's debatable how much this is an "Elephant in the Brain" situation. Yes, we're signaling and probably doing it unconsciously. (Nobody thinks, "I'm making homemade soup not because I think it’ll help, but to demonstrate my devotion.") But unlike many other examples in the book, this isn't an “ugly” motive we're hiding. The hidden truth here isn't that we're selfish—it's that we're inefficient. We could achieve the same medical results without the additional medical theater, but somehow that wouldn't feel like enough.

One problem I had with this section is that the elephant seems most visible when we make medical decisions for others. But most people make their own medical decisions, where conspicuous caring shouldn't apply. However, even accounting for this, there is probably still a decent amount of conspicuous caring. Many medical decisions involve dependents, and elderly patients often defer to adult children who want to signal their devotion.

The weight of all the evidence suggests they're pointing at something real. Even if we're moderately skeptical and only attribute, say, 15% of medical costs to the elephant, that would be $735 billion7. That's in the same ballpark as the total cost of the US military. So even if you're skeptical, it's worth considering.

Religion

This is where they say the elephant is largest: "In few domains are we more deluded, especially about our own agendas, than in matters of faith and worship."

Religion puzzles the evolutionary mind. Why spend Sunday mornings in uncomfortable clothes listening to ancient stories? Why avoid perfectly good bacon? Why pray to a God who, statistically speaking, doesn’t seem particularly responsive? From a purely rational standpoint, it's madness.

The puzzle deepens when you notice the chaos within each faith. Catholics, Baptists, and Mormons all claim to worship the same God, yet they can barely agree on anything—not on birth control, not on baptism, not even on coffee. But this lack of convergence on correct beliefs doesn’t seem to bother any of them.8

We assume people have religious beliefs, therefore they engage in religious practices. But Hanson and Simler argue we've got it backwards. What if the practices came first, and the beliefs are just along for the ride?

Consider the most "irrational" parts of religion, the costly signals that make economists wince:

• Jewish and Muslim boys undergoing circumcision

• Mormons spending two years knocking on doors in Guatemala

• Muslims fasting from dawn to sunset during Ramadan

• Christians tithing 10% of their income

• Hindus bathing in the freezing Ganges

• Orthodox Jews avoiding electricity on Sabbath

These aren't bugs in the religious software, they're features. The costlier and more arbitrary the ritual, the better it works as a signal of commitment. You can fake a belief, but you can't fake circumcision. You can claim to love your religious community, but talk is cheap—tithing is expensive.

Here's their key insight about sermons:

Imagine a preacher addressing a congregation about the virtue of compassion. What’s the value of attending such a sermon? It’s not just that you’re getting personal advice, as an individual, about how to behave (perhaps to raise your chance of getting into Heaven). If that were the main point of a sermon, you could just as well listen from home, for example, on a podcast. The real benefit, instead, comes from listening together with the entire congregation. Not only are you learning that compassion is a good Christian virtue, but everyone else is learning it too—and you know that they’re learning it, and they know that you’re learning it, and so forth. (And if anyone happens to miss this particular sermon, don’t worry: the message will be repeated again and again in future sermons.) In other words, sermons generate common knowledge of the community’s norms. And everyone who attends the sermon is tacitly agreeing to be held to those standards in their future behavior. If an individual congregant later fails to show compassion, ignorance won’t be an excuse, and everyone else will hold that person accountable. This mutual accountability is what keeps religious communities so cohesive and cooperative.

This mutual accountability explains religion's superpower: creating trust among strangers. Religious communities show measurably higher levels of mutual aid, charitable giving, and collective action than secular groups. When someone shares your costly signals—when they also wake at dawn for prayer, avoid the same foods, and wear the same funny hats—you know they're invested. They're not going to defect. They're one of us.

This also explains why 2+2=4 wouldn’t be a good foundational statement for a religion. Any idiot could have realized that on their own. 2+2=5 is much better. It takes a special kind of idiot—one indoctrinated in a certain set of beliefs—to believe that. Thus, by repeating 2+2=5, people are affirming that they have been indoctrinated into a way of thinking, into a culture, into a collective set of norms and laws and values.

The beliefs are mostly window dressing. Sure, some people sincerely believe them (or, at least, believe they do), but the beliefs follow from the practices rather than driving them. We don't worship because we believe. We worship (and therefore believe) because it helps us cooperate with our tribe.

This is the religious elephant in the brain: We think we're reaching for the divine, but we're actually reaching for each other. The arbitrary rules, the painful rituals, the time-consuming ceremonies—they're not obstacles to community, they're the technology that creates the community.

Like the section on health, this is less about hiding "ugly" motives and more about misunderstanding our true purpose. Building trust and cooperation isn't shameful, it’s great. The elephant here is that we've convinced ourselves the rituals are about pleasing God when they're really about binding ourselves to each other. We're not hypocrites; we're just confused about why we do what we do.

Politics

Probably to no one's surprise, the political elephant is an active and lively beast. Politics offers the perfect stage to believe we're pure of heart while engaging in virtue signaling and public acts of tribal allegiance.

The conventional story goes like this: elections matter, staying politically informed is your civic duty, and people vote because they want the candidate with the best policies to win.

But if voting were really based on impact, we'd see dramatically different behavior in swing states versus safe states. Imagine you're a California voter with car trouble on election day. Missing the vote is no big deal—everyone knows who will win California. But in Wisconsin, where your vote might tip the election, you'd move heaven and earth to cast it.

Yet voter turnout in swing states is only modestly higher. Your chance of affecting the outcome might be 1,000 times higher in Wisconsin than in California, but voting rates barely budge. This suggests the actual impact of our votes isn’t a big factor in whether we decide to vote or not. Something else must be driving our political behavior.

One of the authors, Kevin Simler, ran an experiment in rational voting. For the 2000 presidential election, he wrote down his policy preferences and had a friend compile the candidates' positions. He committed to vote for whoever matched his views best, regardless of party. In the end, he voted for Al Gore.

His reflection afterward is telling:

[P]sychologically speaking, the method was distinctly unsatisfying. It produced a result, but there was no joy in arriving at it. Moving past the psychological, however, there were very few social rewards to this process. It didn’t provide opportunities for me to discuss or debate the issues with my friends, nor to advertise my loyalty to one political team over another. Yes, the Democratic candidate was popular among my left-leaning peers—but I wasn’t voting for the Democrats per se. The very fact that I was open to voting for Bush betrayed my lack of political loyalty. As if to drive home the point, when Bush eventually won the election, I wasn’t particularly disappointed. Sure, my preferred candidate had lost, but without an associated team to root for, it was hard to get too worked up over it. If politics is a team sport, “rational” voting is like playing Tetris alone in the corner.

Notice what made this approach so unsatisfying: not the outcome, but the inability to signal tribal loyalty. The loss of social bonding opportunities that came from discussing candidates, debating issues, and sharing outrage with like-minded friends. Being open to either candidate marked him as politically unreliable—the kind of person who might betray the tribe next election.

Like all of the behaviors I’ve been describing, people follow the news and vote for multiple reasons, including the virtuous ones we advertise and the less seemly ones that Simler and Hanson bring up. We could repeat our thought experiment to help see how much to attribute to different reasons: How much would you follow politics if you could never share your views? If you could only vote in secret, never discuss candidates, and never wow friends with your amazing opinions, would you still spend hours reading news and improving your knowledge?9

Political beliefs and actions function as signals of group membership. Individuals adopt specific political stances, express strong opinions, and engage in visible political activities (e.g., attending rallies, social media activism) to signal their commitment to a particular group and its values. These signals demonstrate that one is a trustworthy ally within the political tribe and shares its moral framework.

Conclusion

When evaluating any model of human behavior, the question isn't whether it's "correct"—all models are simplifications. The question is how much explanatory power it provides. On this metric, The Elephant in the Brain succeeds remarkably well. Unlike grand theories that sound profound but explain little (looking at you, Denial of Death), Simler and Hanson's framework actually helps us understand and predict human behavior.

Consider how often our actions contradict our stated reasons. We say we donate to help others, yet we rarely research which charities save the most lives per dollar. We claim education is about learning, yet we celebrate when classes are cancelled. We insist we care about the environment, yet we make almost no effort to distinguish between what actually helps and what’s counterproductive.

These aren't random contradictions—they follow a pattern. In each case, we pursue hidden selfish goals while telling ourselves (and signaling to others) noble stories. We're not exactly lying; we genuinely believe our own spin (so much so that one might say that humans are simply full of shit). But our actions reveal the elephant: there's often a self-serving motive we'd rather not acknowledge.

Not every example in the book will convince every reader. In my opinion, some of the medical studies have confounding variables that considerably weaken the evidence of an elephant. But overly focusing on individual examples would miss the larger point. The authors aren't claiming that hidden motives explain everything, just that they matter more than we think, and that our minds hide this from us.

Once you see the elephant, you’ll notice it everywhere—the colleague who happens to mention their marathon training in every meeting, the way people post photos of their salads but not their midnight snacks, nearly everything that happens on LinkedIn—and you can't unsee it.

Of course, seeing the elephant doesn't exempt you from its influence. The authors are not immune to it; I’m not immune to it. I wrote this review partly because I want these ideas to reach more people. But would I have worked as hard on it if my name wasn't attached? No. The ideas are already out there in this book, and other reviews exist. Yet here I am, adding my voice, perhaps because I want to signal that I'm the kind of person who reads and understands important books about human nature. (Aren’t I a good ally to have? You should subscribe to my blog.)

Subscribe now

What’s the point of all of this? Acknowledging the elephant doesn't just help us understand puzzling behaviors—it can help us design better policies and institutions.

Take education: if additional schooling is mostly about signaling rather than skill-building, then policies that simply extend schooling are misguided. Here's why: When education primarily teaches valuable skills, a more educated population becomes more productive—all those individual gains should compound and everyone benefits. But if education is mostly a zero-sum competition for limited high-status positions, it's like everyone standing on tiptoes at a concert. Individually, you see better, but when everyone does it, no one's better off, and everyone's feet hurt.

Certainly, there's some of both going on—school makes people more productive, AND it acts as a signal. But the data seem to suggest that signaling might be the greater share. Individuals gain an 8-12% income boost per year of education, but when an entire nation increases its average education by one year, the nation's total economic output grows by only 1-3%. If education were mostly about building productive skills, these numbers should be similar—a country full of 10% more productive workers should be roughly 10% more productive overall. The gap between them is the signature of a zero-sum signaling game, where your diploma only helps you beat others for the same job.

This means that requiring master's degrees for jobs that used to need bachelor's degrees (or bachelor's for jobs that needed only high school) doesn't make anyone more capable. It just forces everyone to spend more years and money to end up in the same place. We've moved the goalposts without improving the game.

Similar arguments can be made for many of these topics. If medicine is partly about conspicuous caring, we can design healthcare systems that provide both effective treatment and the emotional comfort people seek. If charity is as much about reputation as helping, we can structure giving to satisfy both needs efficiently.

The Elephant in the Brain isn't going anywhere. It's been with us throughout our evolutionary history and will remain as long as we're recognizably human. But by acknowledging its presence, we can at least stop designing policies for the noble motives we claim to have and start designing them for the humans we actually are.

Modern medicine, by contrast, is predicated on systematic inquiry, reproducible evidence, and, especially, randomized controlled trials (RCTs). By randomly assigning patients, blinding investigators, and crunching pre-specified statistics, RCTs have become the gold standard for determining what works in medicine.

I love a well-run RCT as much as the next data nerd. All the guardrails are in place, and the math has been pored over for decades, so the statistical analysis can almost run on autopilot. But sometimes we treat the multi-center, double-blind, placebo-controlled, pre-registered, stratified-block-randomized, intention-to-treat RCT as the only legitimate way to learn. But that thinking misses out on mountains of useful data. The goal of scientific inquiry is to elucidate truths about the world, not to check every CONSORT box. We’ve gotten so used to big data that we’ve almost forgotten that there are ways to learn from little data. When RCTs are impractical, infeasible, or flat-out unethical, we still need to discover the truth—and that’s where other branches of statistics, especially Bayesian methods, shine.

In this essay, I'll rewind the clinical trial clock on neonatal Extracorporeal Membrane Oxygenation (ECMO) to tell the story of a medical trial that never should have happened—one where statistical orthodoxy overrode both evidence and ethics.1

What is ECMO?

ECMO is a life‑support technology that takes over the work of the heart and lungs by pumping blood outside the body, adding oxygen, removing carbon dioxide, and returning it to the patient. It essentially assumes the function of the heart and lungs when a patient's own organs cannot keep up. ECMO is intended as a temporary bridge therapy, typically used for days to weeks while the underlying condition is treated or until the patient's organs recover enough to function independently. The medical details of ECMO aren’t the focus here, so I will omit some of them to concentrate on the statistics.2

By 1996, thousands of infants worldwide had been treated with ECMO with ~80% survival rates. Its effectiveness was well-documented in multiple countries. Yet that year, UK researchers conducted a traditional randomized controlled trial, assigning 92 infants to conventional treatment despite the accumulated evidence favoring ECMO. The trial delivered what its designers wanted—statistical significance from a perfectly balanced randomization with traditional controls—but at great cost: the deaths of two dozen infants whose lives likely would have been saved otherwise.

How did we reach a point where the pursuit of methodological purity justified such a trial? The answer lies in how the medical establishment evaluated evidence in the preceding decade. Had we embraced adaptive, Bayesian designs from the outset, we could have reached certainty faster—and with far less human cost.

The First ECMO Trial: Bartlett 1985

Extracorporeal life support began in the 1950s with the first crude heart-lung machines. The first successful case of ECMO was in the early 1970s, and a string of cases followed at the University of Michigan that showed promise. These encouraging results generated enough interest in the medical community to warrant a formal clinical trial.

The first randomized controlled trial of ECMO in infants culminated in a paper published in 1985 (full text available here). The study established objective criteria to identify newborns with an 80%3 or greater chance of mortality even with optimal conventional therapy. Eligible patients were then randomly assigned to either continue receiving conventional ventilator treatment (the control group) or to receive ECMO (the treatment group).

In conducting a trial of a potentially life-saving technology, researchers navigate between two competing goals:

1. Generating reliable knowledge to guide future medical practices.

2. Providing the best available care to the patient who is in the study right now.

This trade-off is known in decision theory as the explore-exploit dilemma. We face this problem all the time: when choosing where to eat, should you return to your favorite restaurant where you know the food is good, or try that new place down the street that might be even better, or might be terrible? In an attempt to balance these competing goals, the trial used a form of response-adaptive randomization, wherein the treatment that a patient receives depends on the results of all previous patients4. The randomization shifts after each outcome—when a treatment succeeds, future patients become more likely to receive it; when a treatment fails, they become more likely to receive the alternative. The goal is to steer more patients to the more beneficial treatment, while also learning which treatment is more beneficial.5

To see how this worked in practice, here are the initial results as described in the study:

The first patient was randomly assigned to ECMO and survived. The second patient was randomly assigned to conventional treatment and died. Hence, the odds of the next patient being randomly assigned to ECMO were 3:1. The next patient was randomly assigned to ECMO and survived.

The trial had preset stopping conditions: it would end when either ten patients had been treated with ECMO or ten control patients had died. After the initial allocation of one patient to ECMO (who survived) and one to conventional treatment (who died), the adaptive randomization increasingly favored ECMO due to its early success. The next nine patients were all allocated to ECMO based on the evolving randomization probabilities, and all survived. This triggered the first stopping condition—ten ECMO patients—though one additional ECMO patient was treated before the trial officially concluded6. In total, the trial included 11 infants who received ECMO (all survived) and one who received conventional treatment (who died).

Following the official study, ten additional infants who met the same entry criteria were treated at the same institution. Eight received ECMO (all survived) while two received conventional treatment (both died)7.

The Critics’ Perspective

The results sparked immediate skepticism. Writing in the same journal the following month, professor of biostatistics Jim Ware and a colleague claimed that the study's severely imbalanced allocation (11 ECMO patients versus only 1 control) undermined meaningful comparison8. In addition, they claimed that the trial design has an unacceptably high false positive rate of 50% because when treatments are equally effective, one would still outperform the other simply due to noise.

This criticism reveals a deep-seated difference in statistical perspective. The critics, applying a frequentist lens, were concerned with controlling the risk of a false positive—that is, declaring ECMO superior when the treatments were, in fact, equally effective. While both frequentists and Bayesians agree that one shouldn't blindly declare a winner, they quantify evidence differently. The frequentist approach sets a high bar to reject the initial assumption of 'no difference.' This approach is designed to be a conservative check on scientific claims.

But Bayesians ask a more direct question: "Given the actual data we observed—11 consecutive ECMO survivors and one control death—what's the probability that ECMO is truly superior?" We don't need to assume the treatments are equal; in fact, we have incredibly strong evidence right in front of us that this is not the case.

In addition, the Bayesian doesn’t have to declare whichever has a higher success rate to be the definitive winner—we can look at the size of the difference, and if it’s too small, declare the result inconclusive.9

Let's step back and think about what we're trying to learn. Each treatment has some true survival rate—a percentage of patients it will save. We don't know these rates yet; that's why we run trials. But even with limited data, we can make informed estimates.

Consider what 11 consecutive ECMO survivors tells us. If ECMO only saved half of the patients (50% survival rate), the odds of seeing 11 successes in a row would be like flipping 11 heads straight: (0.5)¹¹=0.0005. That's vanishingly unlikely. Instead, these results strongly suggest ECMO's true survival rate is much higher, probably between 80% and 100%. When we plot the range of plausible survival rates given our data (the blue line below), the curve peaks near 100%.

The control group tells a different story, though with much less certainty. Since we only have one patient who died, our estimate of the conventional treatment's survival rate is far less precise—hence the wider, flatter probability curve (red line below). After all, a single death could happen even with a treatment that usually works. If the true survival rate were 80%, there's still a 20% chance that one patient would die. But the death does shift our best guess downward—the most likely survival rates for conventional treatment are below 50%.

Yes, there's uncertainty about the exact survival rates for both treatments. But uncertainty doesn't mean complete ignorance. Even with limited data, we can do well to answer the key question: which treatment is more likely to save lives?

Using our probability distributions, we can calculate the odds that ECMO's true survival rate exceeds the conventional treatment's rate. It’s 98.88%.

Think about that—after just twelve patients total (eleven ECMO survivors and one conventional treatment death), we're nearly 99% confident that ECMO is the better choice. That's not a guess; it's what the math tells us.

Now that we know that, let’s return to Ware’s first point. Ware said that there was a 50% false positive rate. Here is where the Bayesian and frequentists differ in perspective. The Bayesian approach doesn’t naturally come with a “no significant difference” threshold. The frequentist approach does have a threshold—a p-value of less than 0.05—though that number is completely arbitrary. Sure, it's longstanding and commonly recognized (and we have five fingers, so that sort of matches), but it was originally just a convenient suggestion by statistician Ronald Fisher that somehow became an almost magical threshold in scientific literature. The Bayesian way isn’t to accept or reject a null hypothesis; it’s to quantify how much the data have updated our beliefs, and decide whether that update is large enough for the practical decision at hand. The frequentist fixation on the null hypothesis blinds us to what's right in front of our eyes: overwhelming evidence that these treatments are not equally effective.

Actually, it’s not quite that simple. This is where the Bayes critics come in. What you believe after the evidence is partially dependent on what you believed before the evidence. Bayes only provides updates.

What if you came to this trial with different prior beliefs? Indeed, there is reason to be skeptical. A randomized trial of ECMO in adults during the 1970s had failed to show benefit, which raised concerns that the apparent neonatal success might not hold up in a controlled trial.

Perhaps you gave it only a 20% chance of working in infants, though you didn’t hold this belief particularly strongly. Even with this skeptical starting point, the trial's strong results would move you significantly—your probability would rise to about 62%. The data largely overcomes your doubts.

Now consider the die-hard skeptic. Someone absolutely convinced ECMO won't work. They’ve seen a million of these “miracle cures” and aren’t particularly interested in getting burned again. It’s not that they think the chance is less than 20%—they’ve seen about 1 in 5 cures work before—it’s that they know how excited people got about those 4 other things that didn’t pan out, so they have a strong prior belief of 20%. Even if they hold their 20% prior belief very strongly, the trial result still forces some updating. As shown below, their probability would shift to around 34%. They remain skeptical, but even entrenched beliefs must bend somewhat to such stark evidence.

The probability distribution we modeled above is for someone who started with no background information and thought there was a 50-50 chance ECMO would work. You can see how people with these prior beliefs would update them after seeing the new data.

In the end, Ware was among those more skeptical and declared the results insufficiently convincing to justify widespread adoption without further randomized trials using conventional statistical approaches and adequate control groups.

The Second ECMO Trial: O’Rourke 1989

Four years later, a new trial was designed (full text here) to address these concerns. Ware himself joined the research team (full text here). They faced the same ethical dilemma—how to generate reliable knowledge while minimizing harm to study participants—but chose a different approach: conventional randomization with early stopping rules. The design would randomly assign patients until one treatment group reached four deaths, then automatically switch all remaining patients to the other treatment. This ensured adequate sample sizes for statistical comparison while limiting casualties.

In the first phase, nine patients received ECMO and all survived. Ten patients received the conventional treatment, with only six surviving. The four deaths were enough to trigger the second phase, and all remaining 20 patients were given ECMO. 19 survived.

We can apply the same Bayesian analysis to this new data. Now, with ten conventional treatment patients instead of just one, we have a much clearer picture of its effectiveness. The results again strongly favor ECMO. When we calculate the probability that ECMO is superior to conventional treatment, we find it's 99.68%.

We can also combine data from both studies, as shown in the graph below. The additional data doesn't shift our estimates much—both treatments perform about as expected—but it does narrow the uncertainty bands. With the combined evidence, we're 99.97% confident that ECMO is superior to conventional treatment.

Ware was convinced. In a follow-up paper analyzing the accumulated ECMO statistics, he concluded that the evidence strongly favored ECMO and that further randomization would be unethical. The critic had become a convert.

Pushback and Increasing Popularity

You might think this evidence was sufficient to guide clinical practice. It should have been, but skeptics in the UK remained unconvinced.

In 1991, a Lancet editorial delivered its verdict in a title dripping with condescension: "Neonatal extracorporeal membrane oxygenation [ECMO]: how not to assess novel technologies," The author, Stephen J. Elliott10, criticized the unconventional methods, complaining that the first trial "accumulated just one patient in the control group" and that the statistical methods "did not allow interpretation of the data other than by career statisticians." If calling the statistician hotline is the price of minimizing harm to study participants, that seems like an easy decision to me.

Despite mounting evidence, both from the trials and extensive clinical experience, Elliott maintained that "Although many paediatricians believe neonatal ECMO to be of proven benefit, it is not."

A transatlantic divide was emerging. As the editorial acknowledged, "A randomised, controlled trial of ECMO based on traditional statistical techniques is no longer possible in the USA." American clinicians, convinced by the existing evidence, had moved on. The UK would take a different path.

Meanwhile, real-world evidence continued to accumulate. UK scientists reported initial findings in 1992 with an 80% success rate. By 1995, doctors from the University of Michigan had treated 460 cases with an 87% survival rate, and reported 8,913 cases worldwide with an 81% survival rate. The procedure was becoming standard care everywhere—except where skeptics still demanded their preferred form of proof.

The UK Collaborative Trial 1996: How Not to Take Care of Patients

In 1996, A trial was run in the UK that randomized nearly 200 infants equally between ECMO and conventional care11. The results delivered a clear verdict: 63 of 93 (68%) ECMO infants lived versus 38 of 92 (41%)12 receiving standard treatment. Put differently, ECMO cut the death rate nearly in half (relative risk 0.55, p = 0.0005). ECMO saved one additional life for every three to four infants treated.

Unlike the earlier American trials, there was no adaptive design, no early stopping rules, no protocols to minimize casualties—just conventional randomization pursued to its bitter end. Everything was designed to satisfy the simplest statistical methods. They got their p-value. But at what cost?

The UK investigators proclaimed their trial "the most reliable and unbiased information" about neonatal ECMO. But was it? By 1996, thousands of infants had been successfully treated with ECMO worldwide. The technology's benefits were well-established. This trial didn't discover new knowledge—it merely repackaged existing knowledge in the preferred format. And at the cost of infant lives.

Even after obtaining these results, the authors hedged, calling them "preliminary"—a reflexive scientific caution that borders on the absurd. This is the same mentality that leads the WHO to label nearly everything a "possible" carcinogen. When overwhelming evidence can still be dismissed as "preliminary," we've lost sight of science's purpose.

The Clock is Always Ticking

Between 1985 and 1996, while statisticians debated p-values and randomization ratios, thousands of infants received ECMO treatment with survival rates consistently above 80%. The technology worked. Centers around the world knew it worked. Parents took their babies home because it worked. Yet institutional gatekeepers demanded their preferred form of proof, treating alternative evidence as somehow less real.

Fortunately, we got it right in the end, and over 63,000 children around the world have been supported by ECMO. Yet the ECMO trials raise important questions about how we evaluate medical evidence. While large-scale RCTs remain invaluable when feasible, they aren't always practical or ethical. The ECMO experience demonstrates the cost of dismissing alternative forms of evidence: a decade’s delay in UK adoption, a trial in which many infants were randomized to what was almost certainly an inferior treatment.

The lesson extends beyond ECMO. In medicine's highest-stakes moments, when mortality looms and time is scarce, we need statistical frameworks that can learn quickly and adapt gracefully. We need frameworks for extracting signal from messy, real-world data while acknowledging uncertainty. The Bayesian approach offers exactly this: a mathematical language for updating beliefs as evidence accumulates and for combining different types of evidence into a holistic picture.

Quality of evidence is a hierarchy, not a binary of “meaningless” versus “real evidence.”13 Meta-analyses of RCTs rightfully sit at the top, but we’d be remiss to ignore the rest. Adaptive Bayesian designs extracted a 99% posterior probability of ECMO’s superiority after only a dozen patients.

Consider the registry data that critics dismissed: Elliott himself noted that the US Neonatal ECMO Registry documented 3,577 cases by 1989, with survival rates matching the trials. Yes, uncontrolled data can mislead—selection bias, changing standards of care, and confounding factors are real concerns. But the absence of randomization doesn't automatically render data nugatory. It means we need to examine it carefully. Were these 3,577 ECMO cases cherry-picked from healthier infants? My understanding is that these infants still met the same high-mortality criteria (80%+ predicted death rate), and conventional outcomes hadn't budged. When thousands of desperately ill infants survive at rates that dwarf historical controls, dismissing this evidence purely because it lacks randomization isn't scientific rigor—it's ritualistic adherence to particular study designs.

Imagine if we approached registry analysis with the same statistical sophistication we bring to trial design. We could model selection biases explicitly, conduct sensitivity analyses, use propensity matching or instrumental variables.14 Even if only half the registry cases had sufficient data for careful analysis, that would yield over 1,700 data points—valuable evidence that puts no additional patients at risk. Would everyone agree on the exact numbers? Of course not. But if careful analysis of over a thousand data points suggests conventional treatment had a 20% survival rate while ECMO patients survived at 80%, that gap is too large to ignore.

The tragedy isn't that we have imperfect data—it's that we often treat imperfect as worthless. This isn't an argument against randomization or blinding; it's an argument against worshipping them. The story is not Bayesian heroism versus frequentist obstructionism; it is a case study in how both paradigms are valuable in different contexts (though I am partial to Bayesian methods). Statistics is a toolbox, not a religion. Use the randomized wrench when it fits, the Bayesian ratchet when it doesn't, and the registry flashlight to see what's actually happening outside the lab. Patients don't care whether their survival clears a p < 0.05 hurdle—they care about not dying.

When prudent, we should embrace trial designs that learn as they heal, fuse trial data with real-world evidence, and update beliefs in real time. When the next lifesaving technology arrives, let’s not demand ten more years and a “definitive” p-value before acting. The clock is always ticking for someone in the NICU; our methods should match that urgency.

Broaching such a topic can be dangerous territory. Reeves says he “lost count” of the number of people who advised against it. Critics accused him of reinforcing biological essentialism by emphasizing sex differences or detracting from women’s rights by talking about men. This book surely took fortitude to write, and I’m glad he did.

That said, while Reeves’ evidence is compelling, I found myself often disagreeing with his prescriptions. He acknowledges biological differences between the sexes, yet too often treats the resulting disparities as problems requiring intervention. His solutions lean heavily on top-down mandates and quotas, attempting to engineer specific outcomes rather than focusing on individual flourishing. Reeves diagnoses real problems, but he often defers to a gap-closing creed that seems designed to satisfy a spreadsheet warden obsessed with balanced columns, not individual well-being.

The State of Boys and Men

Reeves marshals extensive data to paint a troubling picture. By nearly every educational measure, boys are falling behind. The most common grade for girls is now an A; for boys, it's a B. Boys are 50 percent more likely than girls to fail at math, reading, and science. In the UK, 40 percent of women enter college at age 18, compared to just 29 percent of men. In the US, women now earn 100 bachelor's degrees for every 74 earned by men.

Below are a couple of particularly telling graphs from the book:

He points to lots of reasons why boys and men aren’t doing well. His central observation is that men have traditionally derived meaning from their role as providers, but this foundation has eroded as women's economic independence has grown. The result is a crisis of purpose that touches every aspect of men's lives. He says men are “increasingly unable to fulfill the traditional breadwinner role but yet to step into a new one.”

Differences Between Men and Women

Reeves and I share considerable common ground. We both recognize that nature and nurture matter and interact in complex ways to create differences between males and females. Sex differences are influenced by cultural context—societal norms can either amplify or suppress biological tendencies.

Reeves even lists some biological differences between men and women. He observes that "men are typically more aggressive, take more risks, and have a higher sex drive than girls and women." He notes that "males are a bit more interested in things, while women are a bit more interested in people," and emphasizes that "men have a greater appetite for risk. This is not a social construct. It can be identified in every known society throughout history."

He also points to developmental differences, noting that boys' brains develop more slowly than girls', with the male prefrontal cortex particularly lagging behind. These differences persist beyond adolescence and, as Reeves acknowledges, have societal implications: "Masculine traits are more useful in some contexts, and feminine ones in others."

I appreciate Reeves' empirical approach to these questions. The existence of sex differences is ultimately an empirical question, not one of ideology, and he treats it as such. He is, of course, aware of the political sensitivity of the topic, noting that "The idea that there is a natural basis for sex differences is, however, politically charged. So I'd better get the caveats in right away. First, while certain traits are more associated with one sex than the other, the distributions overlap, especially among adults."

Reeves relies on this notion of overlapping distributions to justify many of his policy claims, so it’s worth exploring. While technically accurate, it obscures the magnitude of differences. We hear this concept invoked frequently to downplay differences between groups, but the goal should be to understand the magnitude first, and then we can decide what we think about it. Let’s look at one of those overlapping distributions to get a sense of it.

This is a graph of testosterone. Technically, there is an overlap. Technically, some women have more testosterone than some men, but using these distributions1, a man in the bottom 1% of men still has more testosterone than 99% of women, and vice versa. The average man has about 27 times as much testosterone as the average woman, and more testosterone than 99.999% of women. So when someone says “overlapping distributions”, it’s worth asking what those distributions look like.

Admittedly, testosterone is one of the least overlapping differences between men and women, but it’s an important one. We think of it as the muscle-building hormone, but it does much more than that. Low testosterone in men is associated with poor cognitive performance, higher rates of depression, and higher risk of Alzheimer’s2, though the impacts on women appear to be less significant. It also seems to contribute to males’ greater interest in sports. Women with much higher rates of testosterone due to congenital adrenal hyperplasia also display much greater interest in sports. Sometimes it’s hard to get the feel of what these effects mean, so I recommend reading one woman’s experience taking testosterone replacement therapy. This is, of course, anecdotal, but it gives you a visceral sense of how significant hormones are to our lives.

When Reeves invokes "overlapping distributions" without showing the actual data—without revealing that the overlap in testosterone is vanishingly small—he minimizes a biological reality that fundamentally influences behavior. It’s important to look at these numbers and acknowledge that some sex differences are so pronounced they can be more accurately thought of as categorical differences rather than swept away due to a tiny amount of overlap.

Reeves’ Solutions

Reeves advocates for starting boys in school a year later than girls (which he calls "redshirting," borrowing from the sports term), recruiting more men into healthcare, education, administration, and literacy jobs (HEAL—a term I think he coined), and providing more support to keep fathers involved in their children's lives.

Within each topic, he offers a lot of ideas, so some I agreed with and some I didn’t. But my fundamental disagreement lies with his consistently top-down approach. Often, when Reeves sees a gap between the sexes, his reflex is to declare, "we must close or at least shrink this gap." Throughout the book, I found myself repeatedly asking: Why? Why must we mandate these outcomes? Why assume every disparity requires policy interventions?

I think this “closing the gaps” instinct is just fundamentally flawed.

For example, he says we have an “education system favoring girls and a labor market favoring men” and that “We need to fix both”. His main evidence that schooling favors girls is that they do so much better at it—higher GPAs, higher college‐entry and completion rates.

Let’s take the going to college part. We can think of whether people go to college as a function of whether they have the opportunity and whether they find the opportunity valuable enough to justify the costs (both financial and opportunity costs).

Higher female enrollment isn't obviously a question of opportunity. There are some women-only scholarships, but I doubt it’s a large effect. And it’s not like we live in a culture where parents routinely fund daughters' educations but not that of their sons. If we did, that would clearly be a problem worth addressing. But that's not our reality.

Instead, it could be a question of value. If 10 million women decide college serves their best interests, why must exactly 10 million men reach the same conclusion? Why would men and women necessarily derive identical value from our education system? Reeves acknowledges biological and psychological differences between the sexes, yet resists accepting that these gaps might be producing the differences we see.3

My point isn’t that the current gaps are necessarily “correct”, but to push against this knee-jerk reaction to close every gap, to form people to fit the spreadsheet ideal.

Here’s another example: there is plenty of evidence that men are more interested in computer science and women in psychology (Reeves even cites some of it). Imagine this was the only difference between men and women. Since jobs in computer science require less formal education than ones in psychology, this would create a college enrollment gap. Obviously, the difference between men and women extends beyond college degree preferences, but my point is, there are many reasons this gap could exist. Why is there some moral mandate to close it?

I want everyone who would benefit from going to college to have the opportunity. But if everyone gets that opportunity and we still see a gap, I see no cause for concern. Equal opportunity, not equal headcounts, should be the goal.

Let me give another example of why “closing the gaps” is fundamentally the wrong way to approach public policy. Men have much higher suicide rates than women. But I don’t want to “close the gap”. I want to lower both rates. If your mission is to reduce suicide rates for females, I applaud you. I think dedicating oneself to reducing female suicide rates is a great and wonderful thing to do, irrespective of any gaps. Yes, it would be great if we could lower the suicide rate for males as well. But “closing the gap” shouldn’t be our lodestar—maximizing flourishing and minimizing suffering should be.

Obviously, Reeves isn't advocating for increasing female suicide rates. But societies become slaves to their metrics—what gets measured gets optimized, even at the cost of real human goals. A gap between demographic groups should serve as a starting point for investigation, prompting us to ask: "Is something going on here?" Yet the mere existence of a gap doesn't prove a problem exists. Too often, we leap from observing a disparity to demanding intervention, without pausing to determine whether anyone is actually being harmed or excluded.

That’s the disagreement at a philosophical level. Let’s take Reeves’ proposals in turn.

Contra Redshirting the Boys

Reeves points to the slower developmental time for boys and proposes that boys should start school a year later by default—attending pre-K for two years while girls attend for one. I hadn’t heard this proposed before and I found it genuinely innovative and somewhat radical. Kudos to him for thinking boldly.

However, it also epitomizes the problem I have with his top-down mentality. I am perfectly willing to accept that many children benefit from starting school later. I am also willing to accept that perhaps boys, on average, might benefit more than girls. Though I don’t have any strong intuition about how significant this effect might be, I will accept, for the sake of argument, that boys might do best starting a full year later. I don’t know if this is true, but it’s plausible. But why mandate this based on sex?

Why not make school entry more individualized? We could develop readiness assessments (not necessarily written tests) based on developmental milestones: "Children generally thrive in first grade when they can do X, Y, and Z." This would allow natural variation while addressing developmental differences.

We could also work to decrease the stigma around starting later. Perhaps it's simply a matter of publicizing what’s already happening—according to Reeves, children from affluent households are twice as likely to delay school entry. When parents realize that wealthy families view "redshirting" as a strategic advantage rather than a developmental concern, perceptions might shift.

Why does this make a difference? Let me tell my abbreviated story with school. If my high school teachers had to give me a superlative, I would suggest “Safest to Ignore”. That is, I could be pretty much left alone in class with the textbook, and I was “fine”. “Fine,” as in, I wasn’t going to fail the class. I wasn’t going to get in a fight with the kid next to me. I was certainly not being challenged in the subjects I was best at, but schools don’t measure that. By every metric principals use to evaluate teachers, ignoring me caused no harm.

This experience is far from unique. Countless students sit bored in classes not designed for them. Under Reeves' system, I would have been a year older and even more bored.

Why not let a gap appear organically? If we normalize flexible school entry and boys naturally average a year behind girls, both Reeves and I are happy. But what if the gap is only six months? I'm still fine if this reflects individual choices and abilities, but Reeves wants more. Why? Why must the gap be exactly one year? Why impose uniformity instead of allowing variation?

Contra More Male Teachers

Reeves also advocates for recruiting more male teachers, citing studies that show boys perform better with male instructors, particularly in English, while girls perform equally well regardless of the teacher's gender. Even accepting his premises, I disagree with his approach.

Finding a correlation with gender doesn't mean gender is the causal factor. The relevant factor might be teaching style, subject background, or personality traits that happen to correlate with gender. Instead of mandating quotas, why not identify what actually drives student success and recruit teachers with those qualities? And if it turns out that we can’t determine a priori which teachers will be better, we could at least work harder to retain the best ones.

Reeves claims boys do best with male teachers and black children with black teachers, while other students are unaffected. If true, a merit-based system would naturally produce these outcomes. But what matters is building systems that identify, hire, and especially retain excellent teachers. If the best teachers turn out to be 80% black men and all children flourish, wonderful. The goal is to identify what makes teachers effective and select for that directly—not to use demographics as a crude proxy for teaching excellence. If the distinguishing qualities aren't demographic at all, we'd discover that too.4

Contra More Men in HEAL

Reeves wants more men in health, education, administration, and literacy (HEAL) fields. He notes the low male representation in these jobs and the stigma surrounding men in caregiving roles and wants to change that. Drawing parallels to efforts to increase women in STEM, he proposes similar initiatives for men in HEAL, setting specific targets: 30 percent of STEM jobs for women, 30 percent of HEAL jobs for men by 2030.

What's Reeves' justification? He points to overlapping distributions, arguing it's "absurd to think that the 18% male share of social workers is an authentic representation of the true level of interest in the job among men." But the simple fact that men's and women's attributes overlap doesn't justify any particular numerical target. The existence of overlap tells us nothing about what the "right" percentage should be. Why 30 percent? Why not 25 or 35?

Here, he cites the work of psychologists Rong Su and James Rounds5, who found that differences in interests between men and women strongly influence career choices. These studies don't attempt to separate what’s cultural from what’s biological or resolve any “nature vs nurture” debates. They simply document current interests. For example, they find that men show much higher interest in engineering than women, with a Cohen’s d of 1.11, which indicates a large effect size6. Then, they examine who falls in the top 10% of interest for various subjects and use that to predict gender ratios within each field.

Yet Reeves takes this descriptive research and transforms it into a prescriptive mandate. While the studies show that differences in interest contribute to gender disparities, they can’t be used to calculate the "correct" percentage of women or men in any field.

Reeves, however, uses this data to conclude that we need more women in STEM. Funnily enough, if you look at the graph in the academic paper, it projects fewer women in applied mathematics than the actual amount (note that applied mathematics is distinct from mathematics; Reeves omits this column from his chart). Following this line of thinking, should we have fewer women in applied mathematics? No, of course not. This kind of academic overreach is silly.

For me, the question is: Are there large numbers of women in HEAL wishing they were in STEM, and men in STEM yearning for HEAL careers? Perhaps some people were discouraged from pursuing those fields when, in fact, they would have loved them. If so, I support those people changing careers. But the mere existence of gender disparities doesn't prove this.

If everyone got the job they wanted and we still had more women in HEAL and more men in STEM, would that be a problem? Who exactly is harmed by these gaps? What actual human being suffers? On what grounds is there a moral imperative for intervention?

Regarding reducing stigma, we are more aligned. He relates a story where his son was turned down for a childcare job because “the parents were uncomfortable leaving their children with a man.” I think the fact that we have gotten to the point of discomfort with the idea of men being around children is a blight upon our society.

Here, I think we can and should do things. To some degree, we can change what things we as a culture do and do not stigmatize. In general, cultural barriers that prevent people from pursuing careers they'd excel at should be dismantled. I think we can do it in the case of reducing stigma. Activist organizations have been successful here. Recall, for example, the success of MADD in stigmatizing drunk driving, or the success of reducing negative stigma against the LGBT community.

Reeves on Career and Technical Education

Reeves also writes on the importance of Career and Technical Education (CTE), and I appreciate his thorough compilation of the data. I largely agree with his emphasis on providing alternative pathways to four-year degrees. We both recognize that our society treats college as inherently virtuous and non-college paths as somehow lesser. This cultural bias pushes some people into expensive degrees they don't need at the cost of more valuable job training.

But here's where Reeves confuses me: He rightly criticizes our college-for-all culture, yet simultaneously wrings his hands over the fact that more women than men attend college. If we truly believe that college isn't necessary for everyone—and that many people would be better served by vocational training or direct entry into the workforce—then we should stop treating the gender gap in college attendance as such a big deal. The very act of declaring "This is bad for men and we need to fix it" reinforces the notion that college attendance is an important measure of success.

A Better Path Forward

Where Reeves and I agree: Culture matters enormously in shaping how biological predispositions manifest. It is good for people to understand that sex differences exist and that there is some degree of overlap. Most importantly, we should shout from the rooftops that every person is an individual and shouldn't be judged by group averages. We should eliminate genuine discrimination and work to ensure everyone can pursue their potential.

Where we diverge: I don't want to "close gaps" for their own sake. I want to lower suicide rates for both sexes, not celebrate if male and female rates converge. I want excellent teachers, regardless of whether they're disproportionately male or female. I want a society that works with people as they are, not one that hammers them into something else for the sake of balancing a spreadsheet.

We stand on the brink of an AI revolution that is likely to upend the labor market faster than any transition we’ve seen before. We’re going to make a lot of impactful decisions in the coming decade, and we need to be carefully focused on creating human flourishing, not optimizing for bureaucratic spreadsheet equality. We need an approach that values human flourishing over statistical tidiness.

Reeves asks important questions about boys and men. But his answers, constrained by a focus on stamping out spreadsheet disparities, would create a world where individual talents and choices matter less than satisfying some bureaucratic checkbox. We should do better. The future depends not on achieving perfect statistical parity, but on creating a society where everyone—male and female—can pursue their potential without artificial barriers or forced outcomes.

The real tragedy isn't that men and women might sort into different fields or achieve different outcomes. It's that in our obsession with closing gaps, we risk creating a world where no one gets to be who they truly are.

Think of this as a debate between two perspectives. "Signalists" see every price as a data packet—an ever-updating readout of supply and demand. "Price-Fighters," however, see price increases as evidence of exploitation or profiteering. This essay will illustrate why I believe the Signalist perspective is more effective at explaining the empirical evidence and guiding public policy.

The Price-Fighter approach, though well-intentioned, mistakes prices for villains rather than messengers. Prices don't reflect “inherent value” in items themselves but rather indicate the competitive context: what people are willing to pay. They’re messengers conveying information about countless individual decisions and signaling essential information about scarcity and abundance. Suppressing these signals, rather than addressing their underlying causes, often unintentionally exacerbates the very problems these interventions aim to solve.

The Nature of Prices as Signals

Consider the market for nickel. Imagine someone discovers a new battery technology that relies on nickel. This creates new demand for nickel. But how will miners and refiners know to increase production? It’s not because they subscribe to "Battery Technology Weekly". It’s because they see buyers willing to pay higher prices, and they respond accordingly. High prices incentivize miners to open new mines, encourage refiners to improve processes, and prompt manufacturers to seek alternative materials.

When prices increase, the Price-Fighter response is to impose controls. But this misunderstands the role of prices. Letting the market react addresses the root causes rather than the symptoms. Price controls, on the other hand, merely silence the message without addressing the underlying issue. If the government artificially caps the price of nickel, miners won't have enough incentive to find new deposits, and manufacturers won’t feel urgency to find substitutes. Eventually, shortages worsen because the market isn’t receiving accurate signals about the actual situation.

This is what prices do. They convey information. Similarly, when geologists discover a major nickel deposit, suddenly, we realize we have more nickel than we thought—manufacturers can start using nickel in more components. But how will they learn of this? Again, they don’t learn directly. Instead, the new supply information reaches manufacturers through declining prices, prompting them to recalibrate their production decisions. No central authority orchestrates this dance of supply and demand—prices do the work automatically, aggregating complex information from millions of independent actors across the globe.

As Nobel Prize-winning economist F.A. Hayek observed, 'We must look at the price system as such a mechanism for communicating information if we want to understand its real function.'

It’s an incredible feat when you think of it. No single person has comprehensive knowledge of every nickel mine, refinery, or stockpile worldwide, nor can anyone know every potential use and its importance. Yet, through the market, each of us can influence the price slightly based on our own limited information. In this way, prices aggregate all available knowledge from around the globe. Prices in a market economy serve as the most sophisticated information processing system the world has ever known.

Practical Applications

Most people don’t care too much about the price of nickel, but that’s why it’s a good place to get started. But let’s apply these same principles to areas people care about deeply—housing, energy, and insurance.1

Housing

This same principle applies to housing markets. When rents spike in a city, the Price-Fighters say, “Prices have gotten too high and people can’t pay rent. We’re not going to stand by and let this profiteering happen. Let’s freeze the prices.”

But the Signalists say this surge isn't arbitrary profiteering—it's valuable data indicating that demand for housing outstrips supply in that location. The price is simply delivering this message. The Signalist says that if politicians want to reduce the price of housing, they need to focus on the costs, not the price. A price is a communication tool—it conveys information about supply and demand. A cost is the actual resource investment required to produce something. If the goal is genuinely to lower costs, the solution isn’t to suppress prices artificially. Instead, it's to address the true underlying factors: streamline regulations, simplify permit processes, or find more efficient construction methods.

Instead, Price-Fighters attack the messenger—prices—by proposing solutions like rent control. However, rent control targets the price signal rather than addressing the underlying shortage. Capping rents doesn't create more housing or reduce the number of people wanting to live in an area. Instead, it distorts the market's information system. Developers, seeing they can't charge market rates, build fewer apartments. Existing landlords, unable to raise rents, defer maintenance. The housing shortage actually worsens, but now the price signal that would warn us about this growing scarcity has been muted, and the problem grows.

Minneapolis offers a clear example of how effectively cities can respond to price signals. When rents began rising sharply, the city didn't turn to rent control or impose additional regulations. Instead, it recognized the underlying message—housing was in short supply—and tackled the shortage directly. Minneapolis introduced housing reforms such as eliminating single-family zoning and removing mandatory parking minimums. As a result, construction increased significantly, and rents subsequently began to fall.

I’ve talked about housing costs and rent control before, so I won’t belabor the point. But it’s astonishing how often I come across this same phenomenon. Just this year, Zillow found that Denver, a city that has focused on building more housing, led the US in apartment cost decline. The same approach even worked in regulation-heavy Berkeley.

It wasn’t fighting the price through regulation that did it, it was understanding the signal it conveys and responding accordingly.

In California, the approach to housing has been counterproductive: restrictive regulations make building incredibly difficult and expensive—construction costs over twice as much as in Texas—while subsidies attempt to offset these high prices. But when you understand that prices are information carriers, you realize this approach is fundamentally flawed. By making housing nearly impossible to build, California drives up costs at the source. Then, by subsidizing those inflated costs, it suppresses the very price signals that would normally prompt solutions like regulatory reform or increased construction.

Sometimes we forget that an object's price isn’t an inherent property of the item—that a house valued at $1,000,000 doesn’t possess some intrinsic “million-dollar” quality justifying that price. Consider this scenario: the average home price is $1,000,000, and policymakers decide to help by giving everyone $10,000 towards buying a house. If prices were intrinsic, everyone would be $10,000 closer to homeownership, yay! But prices reflect competitive bidding. With more money chasing the same homes, sellers quickly realize they can charge $10,000 more. The price has shifted, but nothing inherent to the homes has changed—only the amount buyers can and will pay.

Thus, when policymakers treat prices as fixed obstacles rather than dynamic signals, interventions meant to lower costs end up making housing less affordable. The result is a handout to current homeowners (because their houses are now “worth” $10,000 more) at the expense of prospective buyers2—exactly the opposite of the intended effect.

This might sound like nonsense, but this is a real thing that actually happens. Research on low-income housing subsidies found that low-income vouchers have “caused a $8.2 billion increase in the total rent paid by low-income non-recipients, while only providing a subsidy of $5.8 billion to recipients, resulting in a net loss of $2.4 billion to low-income households.” That is, the exact people this program was designed to help ended up paying more, and the root cause is a misconception of how prices work.

Energy

Unlike most consumer markets, retail electricity is delivered through regulated local monopolies—households typically can't choose a competing supplier—so many free-market assumptions don't apply. In monopoly settings, prices can't fully function as information signals since there's no competitive market to generate them. However, we can still see the Signalist versus Price-Fighter battle play out in energy pricing debates.

One common proposal—pushed mainly by the Signalists—is time-of-use (TOU) energy pricing, wherein the price of electricity varies throughout the day based on energy availability. This allows the price of electricity to be used to communicate how much stress is on the grid. If energy is abundant (e.g., the sun is shining in a place with a lot of solar panels) and usage is low, the prices are low; but if the grid is maxed out, prices are high.3

I’m a big fan of this approach. Instead of me guessing, “It’s windy today. I bet the local wind farm is doing well right now, so I’ll start the dishwasher”, the electric companies can pass this information along to me in the form of a price. If my need is time-sensitive, I do it anyway. But over time and on the margin, people adapt. People get used to seeing high prices during the evenings and so choose to charge their electric vehicles overnight. The result is lower peak consumption, a more balanced grid, and a more efficient system—a win for everyone.

But despite its benefits, TOU pricing often faces opposition. California has a consumer advocacy group called The Utility Reform Network (TURN) that always opposes efforts to implement TOU pricing. They claim that it's exploitative and unfair for vulnerable populations who rely on electricity. Similarly, the AARP advocates against it, saying that it will hurt those on low incomes. However, this perspective misunderstands the true nature of these price signals. Time-of-use pricing isn’t about exploitation—it's about efficiently informing consumers when energy is abundant or scarce.

Consumer advocates should be pushing for these things. Consumers need to know this information. Transparent pricing helps everyone, especially low-income households, by making energy use more efficient, stabilizing the grid, and ultimately lowering system-wide costs. Inefficient systems hurt vulnerable populations most, as they inevitably lead to higher energy prices. UtilityDive4 reports that “Pilot programs have shown smartly designed residential time-of-use (TOU) and other time varying rate structures can effectively shift power consumption away from peak demand and drive significant savings for both customers and utilities.”

Insurance

Insurance markets are another example of the faceoff between Signalists and Price-Fighters. To a Signalist, when an insurance premium rises, it’s a signal that the amount of risk in that area has increased. In coastal Florida or the wildfire‑prone canyons of California, prices have soared because storms and flames strike more often and rebuilding costs more. A premium spike is a bright red flag: this site is dangerous and expensive to protect.

For the Signalist camp, that message is invaluable. Homeowners can heed it by elevating houses, installing fire‑hardening retrofits, or—if the number is eye‑watering enough—choosing a safer ZIP code. Local governments can respond by steering new builds inland, tightening building codes, or investing in levees and fuel breaks. The market doesn’t forbid living near the water; it simply presents the true cost.

The Price-Fighter camp takes the opposite approach. When private insurers abandon extreme-risk properties, Price-Fighters lobby for subsidies and government-backed insurance programs. The U.S. National Flood Insurance Program is the poster child. Under political pressure to maintain "affordable" premiums, it charged below-market rates for decades, resulting in classic moral hazard: homeowners repeatedly rebuilt properties in high-risk locations, knowing taxpayers would cover the costs.

Insurance premiums reflect the actual price of risk. Artificially suppressing premiums distorts incentives, encouraging reckless behavior. A vivid illustration of this is offered by John Stossel in Confessions of a Welfare Queen: How Rich Bastards Like Me Rip Off Taxpayers for Millions of Dollars:

In 1980 I built a wonderful beach house. Four bedrooms—every room with a view of the Atlantic Ocean.

It was an absurd place to build, right on the edge of the ocean. All that stood between my house and ruin was a hundred feet of sand. My father told me: "Don't do it; it's too risky. No one should build so close to an ocean."

But I built anyway.

Why? As my eager-for-the-business architect said, "Why not? If the ocean destroys your house, the government will pay for a new one."

[...]

If the ocean took my house, Uncle Sam would pay to replace it under the National Flood Insurance Program. Since private insurers weren't dumb enough to sell cheap insurance to people who built on the edges of oceans or rivers, Congress decided the government should step in and do it. So if the ocean ate what I built, I could rebuild and rebuild again and again—there was no limit to the number of claims on the same property in the same location—up to a maximum of $250,000 per house per flood. And you taxpayers would pay for it.

Florida has an ongoing debate around its Citizens Property Insurance Corporation, which is a nonprofit created to provide insurance for homeowners that private insurance won’t cover. Price-Fighters support this kind of thing: “If companies won’t provide insurance, the government has got to step in.” Signalists, on the other hand, oppose it: “Companies won’t insure these properties because the locations are too risky. Why are we subsidizing people to live in places where their houses are likely to be destroyed? They are charging 55% below sound actuarial levels. That means when the hurricanes come and knock down these houses, which they inevitably will at some point, other Florida residents will have to foot the bill. All so people can live somewhere it doesn’t make sense to live.”

And thus, the endless tug-of-war between Signalists and Price-Fighters continues, ensuring bloggers everywhere something to type about.

Working with the Market

Getting mad at high prices for housing makes about as much sense as getting mad at a fire alarm for telling you your house is burning down. The price is not the underlying problem, it's merely an indicator of where the real issues lie. As a Signalist, I believe the goal shouldn’t be to override these signals but to help people respond to them effectively. When we treat prices as signals rather than enemies, we can craft solutions that work with market forces rather than against them.

This doesn’t mean we must rule out all forms of market intervention. There are legitimate “gray-zone” situations—like natural disasters or sudden infrastructure failures—where temporary anti-gouging measures or rent-increase caps can prevent immediate harm and buy crucial time while the market adjusts. For example, imposing temporary price limits during a sudden natural-gas pipeline disruption can stabilize essential services long enough for broader responses to develop. Similarly, given the high transaction costs and logistical challenges of relocating to a new home, carefully designed regulations that limit abrupt, steep rent hikes might be justified.

Signalist thinking is hardly laissez-faire fatalism. We don’t need to throw people out of their homes, turn off their electricity, or cancel their insurance. None of this means that we can’t help those in need, just that we need to do it the right way. It starts by asking What is this price trying to tell us? and ends by fixing the bottleneck the price revealed, not by gagging the messenger. That approach may lack the visceral thrill of railing against “greed,” but facing prices squarely is the only path to durable, real-world solutions.

Nowhere is this clearer than with national borders.

Borders, to borrow a term from Yuval Noah Harari1, are powerful "fictions". These lines on maps are also lines in our minds, feeling real only because we collectively agree to act as if they are. Nation-states have become so familiar that we regard them as natural facts rather than historical inventions. These imagined boundaries deeply shape our worldview, highlighting some issues while obscuring others. An important example of this is inequality: these lines trap opportunity on one side and poverty on the other. Indeed, few anti-poverty measures can rival the effectiveness of simply letting people cross these fictional lines.

The Blind Spots of a National Lens

American debates about inequality revolve around disparities within national borders. We take a country and look at what racial groups do best. This is one way to look at the world, but we could also take a group and ask in what countries they do best. Shift the frame, and a different picture appears. In Nigeria, per‑capita gross national income (GNI) is about $5,385 in purchasing‑power‑parity (PPP) terms. For Nigerian-Americans, the median earnings for full-time workers are around $65,0002, over ten times more.3

Such disparities are tractable: a simple change in immigration policy could lift millions of people’s earnings by an order of magnitude. While not every immigrant will experience a tenfold increase—selection effects are real—it still offers a striking benchmark. Or, here’s another way to look at it: Nigeria’s top 1% earn slightly over $30,000, still less than half the average Nigerian-American income.

Immigration doesn’t merely redistribute wealth—it catalyzes its creation. When people move from low-productivity countries to high-productivity ones, they can earn far more for the same work, generating new economic value rather than taking it from someone else. Economist Michael Clemens famously described immigration restrictions as leaving “trillion-dollar bills on the sidewalk.” Bryan Caplan, in his book Open Borders, takes this idea even further. His book—delivered in a graphic nonfiction4 style (kudos for trying something new)—argues that dramatically loosening migration restrictions could double world GDP or more, making it one of the most powerful tools for global prosperity.

Those who would benefit most clearly recognize these potential gains. Surveys suggest roughly 70% of Nigerians would emigrate if given the chance could and the United States is one of the top destinations; yet the entire Nigerian‑American population is only about 376,000. If reducing poverty is genuinely our goal, few methods rival the simple act of allowing people to move across borders. Borders, not lack of abilities, perpetuate poverty.

If you want to help the poor, let them pursue the opportunities they seek.

Is America a Land of Opportunity?

This global perspective also reshapes how we evaluate opportunities for immigrants in America. Too often, conversations about opportunity in America are framed through a purely domestic lens—comparing outcomes across racial or ethnic lines within the U.S.—without considering the far starker alternatives many immigrants left behind.

Former MSNBC host Joy Reid, the daughter of Congolese and Guyanese immigrants, once recalled her mother saying about the United States: “It’s racist here… they tell me this is the land of opportunity but not for me.” While I can't comment on her specific experience, I do think we should really think about what opportunity can mean in a global context. The notion that America isn’t a land of opportunity for some has become so familiar that many people instinctively nod along with Reid.

The average5 annual income in the Congo is about $630; in Guyana, around $13,600. Reid reportedly earned $3 million a year, meaning that she makes more in a day than the average Congolese does in a decade6. By any global measure, life in the U.S.—despite its flaws—offers vastly greater opportunity for most immigrants.

And this pattern isn’t limited to one country or one immigrant group. Consider India, where the average annual household income is around $4,200. In the United States, Indian Americans don’t just do well—they top the charts, earning the highest median household income of any demographic group: $157,000, according to Census Bureau data7. I will say, again, that of course, selection effects are going on. But we’re not talking about a 40 percent increase, we’re talking about nearly a 40 times increase.

When some people say America is not a land of opportunity, I respond that no society in history has ever enabled so many people, whether immigrants or native-born, to achieve such high levels of prosperity as Indians do in the U.S. To describe the U.S. as lacking opportunity for immigrants is to elide this fact. If our goal is to help the world’s poor, does it really make sense to frame America as a land of closed doors? On the contrary: if we’re serious about fighting poverty, we should be advertising America as a place of opportunity.

That Indian‑American figure isn’t an isolated fluke. When we stop ranking groups within a single nation and start ranking nations by how well groups did, we see things through a different lens.

Let’s look at it another way. Spain’s median household income is around $35,000. This would put “Spanish‑ancestry households” at the bottom of the U.S. income table, trailing every US-based ethnic group. Household income comparisons are imperfect because Spanish people get more services (like healthcare) from the government, but even when that’s factored in, Spaniards have a gross adjusted disposable income8 of $39,964, compared to $67,468 for Americans. If Spain’s income figures were attached to an ethnic group in the U.S., we’d likely view them as victims of systemic injustice. Attach the same figure to a sovereign nation, and it barely registers as significant. Cognitive inertia, not logic, keeps our eyes fixed on domestic gaps while ignoring even larger ones abroad.

None of this diminishes the importance of addressing inequality within countries; I’ve argued before that significant amounts of inequality inflicts real costs on the people who live with them. In addition, internal disparities fuel political tensions and clearly deserve thoughtful policy responses. Yet focusing exclusively on domestic inequality obscures the larger global picture, limiting our perspective. Domestic redistribution programs can reduce inequality within a nation's borders but leave untouched the preponderance of the world's poorest people. On the other hand, allowing freer migration can directly and dramatically reduce global poverty by giving people access to opportunities unavailable in their home countries.

Seeing Through Both Lenses

Categories are tools, not truths. Viewing inequality strictly through national lenses risks mistaking the conceptual boundaries we draw for reality itself, the map for the territory, as it were. There isn’t a “correct” lens here. At some level, both the groups and countries are constructs we make that can illuminate or obfuscate particular dimensions of reality. Both examining how different groups do within a country and how the same groups do in different countries can yield important insights. Ultimately, our understanding of the world is best when we can look through multiple lenses, recognizing that each perspective reveals truths the others might obscure.

Ritchie’s seemingly counterintuitive claim rests on palm oil’s hidden edge: it’s a productivity powerhouse. Oil palms churn out far more oil per hectare than any rival crop, so boycotting it pushes the world toward less efficient substitutes—think soybeans or coconuts. Those alternatives don’t just need more land; they demand a sprawl of it to match palm oil’s output, risking new waves of deforestation in far-flung forests.

The proof is in the yields: one hectare of oil palms can produce over a metric ton of oil, trouncing soybean oil’s 0.161 metric tons per hectare or coconut oil’s scant 0.041. As the graph below shows, palm oil is remarkably land-efficient—palm oil supplies 40% of the world’s vegetable oil on less than 6% of the land used to produce all vegetable oils.

Here’s how Hannah Ritchie puts it:

Here’s another thought experiment to drill this point home: let’s consider how much land the world would need to produce all its vegetable oils from any one of these options. We currently use 322 million hectares to grow oil crops. That’s an area the size of India. If we were to get all of it from palm oil we’d need just 77 million hectares—four times less. We’d free up a lot of land. On the other hand, if we got it all from soybean oil, we’d need more land: 490 million hectares. From olive oil we’d need twice as much land as we currently use—around 660 million hectares. Two Indias.

But instead of using the minimum land possible, avoiding palm oil has become a cause célèbre in environmental activism. In 2017, Ben and Jerry’s proudly announced they had removed palm oil from their ice cream, joining numerous companies in a seemingly virtuous effort to protect the environment. Entire nonprofit organizations have emerged solely dedicated to this effort. For instance, Ethical Consumer, a prominent nonprofit focused on guiding ethical purchasing decisions, maintains a list of palm oil-free products. Organizations like Palm Oil Detectives offer comprehensive lists of palm oil-free snacks.

We can see the vegetable oil substitution happening right on the ingredients list. If we look at the companies that Palm Oil Detectives points to, the top link is to Bean Supreme, whose “eco-friendly” products all contain vegetable oil. The most prevalent vegetable oil in the US is soybean oil. From the above chart, we can see that it’s 8.2 times less efficient. That means we have to use 8.2 times more land to get the oil we need.

One could counter that not all land is equally valuable—perhaps using ten times as much land elsewhere is acceptable if it spares the orangutans' habitats. This is a valid point. However, we can't assume that avoiding palm oil means that alternative oils will be cultivated in areas that are better for the environment.

For instance, Ben and Jerry’s switched from palm oil to coconut oil1. But coconuts are also grown in tropical regions, just like palm oil (except, of course, they use much more land to do so). This substitution satisfies the activists, but it leads to more tropical deforestation—over 30 times more based on the graph above.

Figuring out the true impact of a crop like palm oil is not easy. Land efficiency tells a critical part of the story—palm oil’s unmatched yield per hectare suggests boycotts could backfire, clearing more forests elsewhere. But it’s not the whole picture. What kind of land is used? Who depends on its production for their livelihoods? How do fertilizers, pesticides, or transportation emissions factor in? These tangled threads remind us that environmental choices ripple through complex systems—markets, ecosystems, communities—in ways that defy easy answers.

Perhaps one solution is the Roundtable on Sustainable Palm Oil (RSPO), a non-profit that has developed standards for responsible palm oil production. The goal is to bring together producers, manufacturers, and environmental groups to certify palm oil that meets criteria aimed at reducing deforestation, protecting wildlife, and supporting local communities. On the surface, it seems promising—a practical way to harness palm oil’s efficiency while curbing its harms.

However, a quick internet search reveals that it has many critics. Most of the criticism points to weak enforcement, loopholes, and greenwashing (see here for a good overview of the many criticisms). A 2017 study evaluating the impact of RSPO certification found that while it reduced deforestation in certified plantations, especially in high tree cover and primary forest areas, its overall impact was limited due to the small proportion of plantations certified and their tendency to have little remaining forest already. However, the study concluded that “higher levels of certification could generate greater forest protection.” I haven’t been able to dig into everything, and while some of the criticism certainly has some merit, my sense is that RSPO is a step in the right direction and we should seek to improve it rather than scrapping palm oil entirely.

Zooming out, you can see what has happened here: Vegetable oil is incredibly useful, which has created immense demand for it. Suppliers aim to meet this demand in the cheapest and most efficient way possible, which leads them to palm oil because it produces the most oil per hectare. As demand grows, forests are cleared to plant oil palms. When activists discover palm oil plantations contributing to deforestation, they launch campaigns targeting palm oil specifically, rather than considering the broader context. Companies, eager to respond positively to activist pressure, abandon palm oil in favor of less efficient—and potentially more harmful—alternatives.

There's an important lesson here—not just about activists getting something wrong, but about how well-intentioned activism can go astray. One of Hannah Ritchie’s key recommendations for environmentally friendly eating is to “eat foods that are grown where conditions are optimal.” This is what markets naturally do: they gravitate toward efficiency, selecting products like palm oil precisely because it makes the most economic sense.

The point isn't that markets are perfect—they aren’t. Markets can indeed lead to negative externalities, and interventions are sometimes necessary. But markets are extremely efficient. When activism pushes companies away from palm oil without a detailed understanding of the alternatives, the result can lead to even worse environmental damage. In particular, attempting to intervene in competitive markets without fully understanding their complexities and nuances often produces unintended negative consequences. This pattern is not unique to palm oil activism; similar unintended consequences appear in issues like rent control causing housing shortages or overly cautious regulation stifling nuclear power and thereby hindering clean energy development.

All of these complexities underscore a broader lesson for environmental activism: effective activism is hard.

Passion fuels activism, but good intentions don’t guarantee good outcomes. When we rally behind a cause like boycotting palm oil without wrestling with its full implications—land use, livelihoods, emissions, and all—we risk stumbling into solutions that harm more than they help. The data here suggests shunning palm oil might accelerate deforestation elsewhere, a twist that underscores a hard truth: uninformed zeal can unravel the very goals we’re chasing. Ultimately, effective activism must embrace complexity and nuance. Understanding deeply—not just passionately—is essential for truly making the world better. Real progress demands more than heart—it calls for digging into the messy details and owning the complexity, lest we trade one crisis for another.

Perhaps activists could adopt an intellectual pledge: before passionately campaigning against something, spend more time exploring resources like Our World in Data. Maybe crack open some Vaclav Smil and pledge, “I understand How the World Really Works. I know that Numbers Don't Lie and How to Feed the World. And, above all, I know that environmental challenges require careful, informed solutions, but they are Not the End of the World.”

Here is Ritchie’s list of five food habits that truly matter for climate change:

1. Eat less meat and dairy, especially beef
   She says:

This one will make the biggest difference. It’s one of the most effective things you can do to cut your carbon footprint. When we look at the impact of different foods, a hierarchy emerges. At the very top of this list—way ahead of everything else—is beef.

2. Adopt the best and most efficient farming practices
   It can’t make up for eating a lot of beef, but Ritchie puts this as the second most important point.

3. Reduce overconsumption
   The benefits here are fairly obvious.

4. Reduce food waste
   Again, the benefits are obvious.

5. Close yield gaps across the world
   Many countries have massively increased their crop yields per unit of land, though some have not. If we can help those who have not caught up, we can prevent a lot of deforestation.

Ritchie also addresses three ideas that are often talked about but either don’t help or can even be counterproductive:

1. Eating locally produced food
   She considers this counterproductive. You might think transportation causes most of the CO₂ emissions, but Ritchie says only 5% of greenhouse gas emissions from food come from transportation. Instead, it’s where the food is produced that matters. According to Ritchie, “This means that eating locally can actually be worse for the environment, especially when we opt to grow food where it’s not supposed to be grown.” For instance, tomatoes grown in greenhouses in Sweden use 10 times as much energy as importing tomatoes from Southern Europe. She adds: “A better rule is to eat foods that are grown where the conditions are optimal.”

2. Eating organic
   Organic farming typically has lower crop yields, requiring more land to produce the same amount of food. The benefits don’t outweigh this cost. People like to find reasons to prefer organic, but Ritchie says none stand up. She cites a meta-analysis: “The same meta-analysis showed the unanimous result that organic farming was worse for land use, and also found that organic farming was worse for the pollution of rivers and lakes.”1
   There are many more details in the book and in the cited research.

3. Reducing plastic packaging
   Only a tiny amount of a food’s carbon footprint comes from its packaging. If we reduce packaging to the point where food gets damaged in shipment, we’ve caused far more environmental harm than the minimal savings from cutting down on plastic.

That’s the gist of Ritchie’s take on which food habits genuinely impact climate change—and which ones don’t. She also explores 'How to build a more sustainable food system,' which results in a similar but non-identical list. In that list, reducing meat consumption ranks second, while the top priority is to 'improve crop yields across the world.'

My initial reaction upon hearing this was to think it was not a good idea. I’m generally against efforts to reduce information. How can you identify problems or track progress without data? How can you address discrimination? While anecdotes may offer glimpses into broader issues, relying on them as a basis for policy is unreliable—you need data to truly understand and address systemic issues.

France's approach isn't an oversight. It's a deliberate policy choice. According to Jean Beaman, a professor of sociology at the University of California, Santa Barbara:

The origins of France not collecting ‘ethnic statistics’ date to the Vichy Regime during World War II. By using racial categories, Jews in France and in French colonies in the Maghreb were rounded up and deported to concentration camps in Germany and Poland. Race has ever since been seen as dangerously divisive.1

This wartime experience crystallized into formal policy with the 1978 Data Protection Act, which effectively banned the collection of racial and ethnic statistics across all government agencies. Here is how it’s described in an article in The Conversation:

The French state’s policy rejects any references to national, racial, ethnic, religious or linguistic minorities. This model is based on the idea that the state should interact with the individual only, not communities or groups, in order to give equal treatment to everyone.

The cons of this approach immediately come to mind. Having less or inaccurate data to answer a question or solve a problem never seems like a good idea. I’ve always said that a good data scientist shouldn’t ever bemoan more data. In the worst case, where the data’s irreparably biased, you can always ignore it, but it shouldn’t ever be bad to have more data. If you’re overwhelmed by the amount of data, get a better data scientist.

Sometimes, problems are only clearly visible in data. For example, systemic discrimination would be hard to show without data. Some minority groups have felt this policy makes the discrimination they face harder to demonstrate, though no less real. According to an article in Harvard International Review, “more than 80% of respondents in a recent poll conducted in Seine-Saint-Denis said they attribute race or ethnicity to the discrimination they face when dealing with police or when seeking employment […] Another study conducted about policing in France revealed that Black and North African youths were far more likely to be stopped and searched by police.” The Conversation article argued that by not collecting these statistics, it “serves to make minorities, and the difficulties they face in French society, almost invisible.”

I’ve been critical of the quality of the public conversation we have around race and gender, but is the conversation so bad that it’s better to suppress it? In France, even the concept of race is considered “taboo”. Nondiscussion hardly seems like a solution.

I don’t see which groups would gain from such a policy. No matter what political viewpoint you hold, presumably, you believe that the data—once all accurately corrected and tabulated—supports whatever viewpoint you hold. It makes sense that people would want to gather data to demonstrate their points.

In addition, how efficacious is this prohibition in practice? Nature abhors a vacuum, and the absence of official statistics doesn't mean people stop asking questions about racial differences in French society. In some cases, it seems that French officials try to get around this by funding academics to do the research instead. But these academics have fewer resources, so the data quality suffers2. If there are no official statistics, does that just leave room for unofficial—and quite likely inaccurate—statistics, anecdotes, and gossip to fill the void?

Take, for instance, the widely circulated claim that 70% of France's prison population is Muslim—a figure that, while well-circulated, seems to be inaccurate. If there were official statistics to counter it, the inaccuracy might not have propagated as far.

The Harvard International Review provides more examples:

Additionally, due to the lack of accurate racial data, extreme-right wing politicians have capitalized on the ambiguity, inflating the number of Muslims residing in France and fanning the flames of xenophobic sentiment and anti-immigrant rhetoric. Politicians used such tactics to spread fear about terrorism and mass immigration, two issues at the center of French policy making. More recently, the lack of racial data collection has left citizens of France in the dark about the rates at which people of color are stopped and searched by law enforcement, rates of workplace and housing discrimation, and rates of death due to COVID-19.

It also seems like there would be some sort of Streisand effect, whereby suppressing it, people wonder what the data would show that is so important to suppress.

But are there any pros to France’s prohibition of this type of data collection? If this is a question—like most questions—of tradeoffs, then what are the benefits? Despite the clear drawbacks, could there be subtler advantages that might balance them out? Let’s set aside the practical costs of gathering data and ask whether there’s a broader benefit to not collecting this information at all.

It’s possible that constantly highlighting racial and ethnic disparities can entrench divisions rather than heal them. When every social problem is filtered through the lens of racial statistics, it may shape how society views itself and its challenges. The absence of official data might, paradoxically, create space for addressing problems without immediately casting them as racial issues.

For example, such data will certainly highlight disparities. If you believe that disparities are caused by discrimination, it’s important to highlight them. That’s the first step to eliminating them. But if you believe, as Thomas Sowell does, that disparities are always going to be part of societies, perhaps it’s better not to know? I have to say, this seems like a stretch to me. Though Sowell believed disparities were permanent features of society, I don’t think he ever advocated against collecting data because of it.

In addition, collecting racial and ethnic data necessitates categorizing people into discrete groups, which can lead to oversimplification and inconsistencies. The unclear distinction between race and ethnicity is evident in forms that ask about Hispanic origin separately from race, with only two options: Hispanic and non-Hispanic, a bizarre binary grouping that implies that all non-Hispanics are similar yet different from Hispanics in some important way. Similarly, the term "Asian American and Pacific Islanders" lumps together populations with distinct cultural and historical backgrounds, such as Samoans and Chinese, as if they share a relevant connection. These racial categories can become almost like paper towns—artificial creations sketched out on paper that eventually assume a real and important presence in society. Now we see organizations like “Stop AAPI Hate” emerge in response to anti-Chinese sentiment after COVID-19, although outside of this organization, I’d never heard of anyone linking a Pacific Islander to COVID.

The cons seem concrete while the pros—better society through less focus on disparities—feel tenuous and, to be honest, made up3. While constantly highlighting racial and ethnic differences might in some cases reinforce divisions, there is little concrete evidence that deliberately suppressing data would lead to a more harmonious society.

I’ve been focused on demographic data and its impact on society, but maybe I’m missing the larger point. John Cowperthwaite, Financial Secretary of Hong Kong from 1961 to 1971, was famously skeptical of the government collecting data as well, even refusing to allow GDP to be collected. When asked what key thing poor countries should do to encourage growth, he replied, “They should abolish the office of national statistics.” When Milton Friedman asked about how few statistics Hong Kong collected, Cowperthwaite responded, “If I let them compute those statistics, they’ll want to use them for planning.’’ During his decade as financial secretary, real wages rose by 50 percent and the portion of the population in acute poverty fell from 50 to 15 percent. However, this begs the question: How would anyone know if such data weren’t collected?

I end up close to where I started: it’s just very hard for me to be opposed to collecting data. I don’t know what would convince me that this is a good idea, but I’ll be on the lookout for more instances of this and try to form a fuller opinion.

Shortly after the Los Angeles fires, California Governor Gavin Newsom took swift action to protect victims—not by streamlining rebuilding efforts or cutting red tape, but by passing a law prohibiting unsolicited offers to purchase fire-damaged properties. The stated goal is to prevent homeowners from selling too low and succumbing to "predatory" investors. CBS News fully embraced this narrative, describing the law as "safeguarding victims of the Los Angeles-area fires from predatory real estate investors."

But if I’m being honest, I don’t see how simply receiving an unsolicited offer is something people need protection from. Or how depriving people of information—how much their property is worth to others—makes them better off.

I get the concern. In the aftermath of a disaster, some homeowners might be tempted to sell quickly, whereas they would be better off if they waited. They might want to stay and rebuild and are better off not selling. But this law doesn’t just apply to those individuals—it applies to everyone, presupposing that homeowners can’t be trusted with their own decisions. Can we really know that every single homeowner is better off under this restriction? After all, those who don’t want to sell can simply say ‘no’.

I don’t blame those who want to sell. Rebuilding 20,000 structures will take years, but the land itself still holds value. What’s wrong with a homeowner saying, “I’d rather not wait. I’ll take the money and be on my way”? Wouldn’t they benefit from receiving multiple offers, choosing the best one, and relocating to a more affordable area?

To be fair, the law doesn’t prohibit selling—only unsolicited offers1. While I don’t object nearly as strongly to a prohibition on offers as I do to an outright ban on selling, I still fail to see how restricting information to homeowners actually helps them.

This isn’t a one-off thing. We see these types of laws throughout the state and federal government. But there's a trade-off between protecting people from imprudent decisions and respecting their autonomy to make their own decisions. It's easy to denounce a business practice as 'predatory' or 'exploitative,' but I don’t think the economic freedom side always gets a fair shake. So let’s look at some of the most controversial examples, not because I necessarily fall on one side or the other, but just to show there are two sides.

Payday Loans

Consider payday loans—an industry that has become the poster child for financial exploitation in political rhetoric. Bernie Sanders has called payday lenders "grotesque and disgusting" and has worked with Alexandria Ocasio-Cortez to push for the Loan Shark Prevention Act, which would cap interest rates on consumer loans at 15%.

At first glance, this sounds reasonable. No one wants to see people trapped in a cycle of high-interest debt. Someone takes out a loan for an emergency, can't pay it back on time, and ends up owing far more than they borrowed.

But to get a better sense of this, let’s look at the numbers. A typical payday loan might be $300 for two weeks, costing $20 per $100 borrowed. That means the borrower repays $360 in two weeks for $300 today.

Imagine you're the borrower. For one reason or another, you believe that taking out this loan is in your best interest. No—not just believe it, but are absolutely certain of it. Perhaps your car needs urgent repairs so you can keep getting to work. But somewhere, far away, sits a government legislator who has never met you, knows nothing about your situation, doesn't even know you exist, and yet has decided they know what's best for you better than you do. This is what a ban on payday loans does.

Critics point to the annualized percentage rate (APR) to argue that these loans are exploitative. The example I used above—$20 per $100 for a two-week loan—would translate into an APR of over 500%. But this is a misleading way to look at the issue. These loans aren't designed for long-term borrowing—they're short-term solutions for immediate needs.

It's like calculating the annualized cost of a hotel stay. If a hotel charges $300 per night, that works out to an "annual rate" of over $100,000—an absurd price to rent a small room! But no one stays in a hotel for a year; they stay for a night or two because that's what hotels are for.

I remember being out of town and needing to print something really important. The local print shop charged $5 for a few pages. I could claim this was “exploitative”—that price is hundreds of times the price of printer paper on Amazon—but at that moment, their service was incredibly valuable. I gladly paid $5—and would have paid $10, maybe even $100 if I had to. Of course, if they were regularly charging $100 per page, I would have opened a competing shop next door and charged $50. High prices attract competition—let the invisible hand keep prices down.

Or, imagine you’re the one giving the loan. Imagine you’re giving a loan to a family who’s buying a house. They show up for the meeting on time, both parents have good jobs, and, miraculously, they completed all the paperwork correctly on the first try. Of course you’ll give them a loan—you can use the house for collateral anyway. They’re borrowing $300,000, so you don’t need to charge near 15% to make a good profit (and the market is too competitive for that anyway).

But these aren't the people asking for payday loans. Consider instead someone with a history of defaulting on loans, or someone without a permanent address. How much interest would you need to charge to make lending viable? What percentage of those borrowers will disappear entirely? And for those who don't, how many hours will you spend tracking them down to collect? Given these risks, you set your rate accordingly, offering $300 for two weeks at $20 per $100 borrowed. If all goes well, you net $60 and they get the money they urgently need. Critics call this >500% APR "exploitative," but it’s what you need to charge to stay in business. At the 15% APR limit some propose, you'd earn just $1.37 for the loan—certainly not worth the risk and effort of managing it.

Minimum Wage

One more example: minimum wage. The government decides that no one can sell their labor below a set price.

But what if someone wants to work for a lower wage? Maybe they have a bad background and just need a chance to show their skills. They’ll be able to charge more once they develop their reputation. Or maybe they’re a teenager with no skills who doesn’t want to sit at home all summer. Maybe their home life is chaotic, and a structured job is an escape.

A job—low-wage or not—teaches valuable life skills: the importance of showing up on time, taking responsibility, and learning from people you admire (and getting along with those you don’t). It teaches which jobs suit you and which don't. A teenager working for $8 an hour is in a far better position long-term than one sitting at home because $15-an-hour jobs are out of reach.

This is not supposed to be the definitive argument against minimum wage laws, or payday loans, or anything else. There is more to say both in favor of and against all of these. But I keep noticing the tendency of some in government to ‘help’ people by restricting what they can do and not enough pushback on it. I think there should be a higher threshold in people’s minds when considering laws like this. It isn’t, ‘is it a good idea for most people?’ It’s, ‘given that someone believes this job or loan or decision is in their best interest, can I really be sure, having never met them, and not knowing their situation, that I know better what’s best for them?’

There are cases when this could be true, where a distant government official does know better. Information asymmetries between a buyer and seller are a classic example—when one party knows crucial facts that the other can't easily verify, like a car's accident history or a drug's side effects. But it reminds me of the Reagan quote: "I think you all know that I've always felt the nine most terrifying words in the English language are: I'm from the Government, and I'm here to help." I would add to it these ten words: "I just passed a law that restricted your economic freedom." It's not always wrong, but the bar should be very high.

I recently visited The Sphere in Las Vegas, a venue that blends immersive visuals, cutting-edge technology, and an ethos of scientific optimism. Inside, you are greeted by interactive robots, walls adorned with math equations, and—above all—an atmosphere radiating a sense of pro-science optimism and progress. The centerpiece attraction is a film, Postcard from Earth, showcasing breathtaking 18K footage of our planet on the world’s largest high-definition screen. The film delivers a message, portraying Earth as a magnificent but endangered planet, driven to ruin by human actions, forcing humanity to seek refuge on a distant world in another solar system.

This premise reflects a blend of environmental guilt and technological solutionism that’s infected a lot of our culture, as seen in films like WALL-E, Avatar, and Interstellar. When faced with tough but solvable challenges—like climate change—we too often dream up fantastical escape plans instead of grappling with the hard work of repairing our planet. This perspective, for all its dazzling futurism, is a strangely defeatist approach to climate policy. It suggests that salvation lies in high-tech heroics and interstellar voyages, glossing over the very real and achievable policy, technological, and social shifts that can safeguard our future at home. Before looking to the stars for a new Eden, we ought to consider that the best solution may be right under our feet, waiting for the collective will to see it through.

In this essay, I’ll examine what Postcard from Earth gets right about our environmental predicament and what it gets wrong. Ultimately, leaving our planet isn’t the answer; the real challenge lies in learning to live sustainably upon it. Let’s begin by exploring what the film gets right.

What It Gets Right

Humans have caused massive ecological damage to the planet, and we can see this across countless measures. For example, extinction rates today are estimated to be 100 to 1,000 times higher than the natural background rate, with the potential to rise even further. Since the emergence of humans, wild mammal populations have plummeted by 85%, now accounting for just 4% of the global mammal biomass. Meanwhile, humans comprise 34%, while our livestock dominates at 62%. Recent research has suggested that “at least ~1300–1500 bird species (~12% of the total) have gone extinct since the Late Pleistocene [11,700 years ago].”

These losses are driven by a combination of factors. Habitat destruction—through deforestation, intensive agriculture, and urban expansion—has historically been the primary driver of biodiversity decline, while overhunting, pollution, invasive species, climate change, and other human activities have further exacerbated the crisis.

For the purposes of this essay, I’ll focus on the challenges posed by greenhouse gas emissions and climate change. Global temperatures are rising at a precipitous rate in Earth’s history, making it a central focus of climate concerns and a good point to focus on. This is not to diminish the importance of other factors—in fact, each of these issues deserves deeper exploration than I can provide here.

With that context in mind, let’s get into…

What It Gets Wrong

Growth Does Not Destroy the Planet

Given the significant damage caused by economic growth and activity, it’s natural to assume that “degrowth”—deliberately scaling back the economy and lowering living standards—is the only way to reduce our impact on issues like carbon emissions. However, this assumption is incorrect. The relationship between economic growth and environmental impact isn't fixed. Through technological innovation and policy changes, we can decouple economic growth from carbon emissions. This happens through multiple channels: improving energy efficiency, transitioning to renewable energy sources, optimizing land use, developing cleaner industrial processes, and shifting toward less carbon-intensive economic activities like services and digital products. As the graph below illustrates, not only is continued growth and carbon emissions reduction possible, but it's already underway in many developed economies.

Let’s look at the US in more detail. We can see that annual CO₂ emissions peaked around 2005. Since then, we’ve continued to have a growing population, a growing economy, and yet declining CO₂ emissions.

We see the same thing happening across the developing world:

This isn’t to say that overall CO₂ emissions are being reduced—they’re not, but we’ll get to that. My point is to show that it’s possible to continue to grow while reducing emissions. This is our best path forward.

It’s important to note that this trend does not extend to the developing world, most importantly in the largest countries, China and India. In these nations, CO₂ emissions are increasing alongside economic growth, and the net effect of all nations is significantly increasing levels of CO₂ emissions.

The task requires fostering the development of nations while bypassing the most environmentally harmful stages. These countries must meet the needs of their populations for food and energy while mitigating CO₂ emissions. It is not an easy task, but it is not an impossible one either.

The Long Road to Cooling

Net Zero

It’s not going to happen overnight. Many places, such as the US, UK, and EU have a goal to reach net zero by 2050. China has set its date for 2060 while India has targeted 2070. Whether and when specific regions hit net zero is less important than the larger trend of decreasing emissions everywhere. We still have decades of rising CO₂ emissions, but not centuries.

The Warming Lag

Even if we completely stopped emitting carbon dioxide today, the planet wouldn't stop warming immediately. Research shows there's roughly a decade-long lag between emissions reductions and maximum warming. Many greenhouse gases remain in the atmosphere for extended periods—CO₂, for instance, can persist for hundreds of years. This means that even after reaching net zero emissions, atmospheric greenhouse gas concentrations will remain elevated for centuries.

This creates a challenging situation: even with aggressive emissions reductions, we're locked into several more decades of warming. It's like trying to stop a massive ship—even after cutting the engines, momentum carries it forward for miles. The warming already "in the pipeline" will continue to affect our climate system long after we achieve net zero emissions. Practically speaking, we’re likely looking at rising temperatures for most—and quite possibly all—of the rest of the century.

Geoengineering?

Given this reality, we may need to consider more drastic measures alongside emissions reductions. One increasingly discussed option is geoengineering—particularly Solar Radiation Management (SRM). Think of it as installing a planetary sunshade.

The most widely studied SRM approach is stratospheric aerosol injection, which would mimic the cooling effect of large volcanic eruptions. When Mount Pinatubo erupted in 1991, it demonstrated this principle by cooling Earth's average temperature by about 0.5°C for several years. The eruption released sulfur dioxide into the stratosphere, creating a reflective layer that bounced some sunlight back into space.

While technically feasible, this approach isn't without significant risks. Injecting sulfur dioxide into the stratosphere will likely disrupt weather patterns and rainfall, potentially causing droughts in some regions while flooding others. It's a bit like taking powerful medicine with serious side effects—perhaps necessary in an emergency, but not something you would have chosen.

More concerning is that once started, we'd need to maintain this artificial cooling indefinitely. Suddenly stopping would lead to rapid warming as the masked effects of accumulated greenhouse gases become fully apparent. It's a serious commitment that would need to be maintained by global cooperation for generations. And it will create winners and losers. Some regions will become worse for farming or their current economic activities.

Given these challenges, geoengineering should be viewed as a potential emergency brake—something to research carefully and deploy only when the alternative is worse. It’s not an alternative to aggressive emissions reductions. The safest path forward remains clear: rapidly transitioning to clean energy while preparing to adapt to the warming we've already locked in.

I’m not saying any of this is easy, I’m saying it’s not impossible. Speaking of the impossible, let’s talk about interstellar colonialization.

Space Is Bigger than We Can Comprehend

The vastness of the solar system is difficult to comprehend, though this NASA video attempts to illustrate it—however, the Postcard from Earth spacecraft ventures beyond not just Earth, but the entire solar system. The real challenge isn’t the size of our solar system—it’s the staggering distances to other stars. Only there might we discover another pale blue dot, waiting for us to arrive.

Proxima Centauri, the closest star to us, is about 4.22 light-years away—a distance difficult to grasp. This video offers a vivid way to imagine it: the presenter scales the sun down by a factor of 33 billion to 1, shrinking it to the size of a golf ball. On this scale, Earth becomes a grain of sand just four meters away, and Neptune, the farthest planet, is located on the opposite side of a park. But how far is the nearest star? To demonstrate, the presenter drives from Manchester, UK, to the Pyrenees in Spain, including a 7.5-hour ferry ride along the journey.

And that’s just the nearest star. Imagine getting there and finding out that it’s just not right.

Space is big. Really big. You just won’t believe how vastly, hugely, mind-bogglingly big it is. I mean, you may think it’s a long way down the road to the chemist, but that’s just peanuts to space.

– Douglas Adams, The Hitchhiker’s Guide to the Galaxy

How Long Would it Realistically Take?

The numbers tell a sobering story about interstellar travel. The Parker Solar Probe, humanity's fastest creation, streaks through space at a blistering 395,000 miles per hour (176 kilometers per second). Yet even at this phenomenal speed, a journey to our nearest stellar neighbor, Proxima Centauri, would take 7,230 years—roughly the same amount of time that has passed since humans first developed writing. While we might eventually build something marginally faster than the Parker probe, the physics of space travel places hard limits on our potential velocity. Without some profound breakthrough that overturns our understanding of physics, we're looking at journey times measured in millennia, not decades. Expecting to figure this out in the coming decades is a lot of hubris from a species that hasn’t successfully colonized Antarctica.

And this assumes we even find something habitable there—what if, after thousands of years of travel, we arrive only to find that our chosen planet isn't actually habitable? It would be like sailing across an ocean for decades only to find that the promised land is actually a barren rock.

Conclusion

The Sphere's message reflects a misguided way of thinking about humanity's future: when faced with difficult but solvable problems, we sometimes prefer to dream up fantastical solutions rather than tackle the hard work in front of us. Yes, climate change presents an enormous challenge, but the notion that building a multi-generational spaceship capable of sustaining human life for thousands of years is somehow more feasible than transitioning to clean energy is a striking example of misplaced technological optimism.

We already have the fundamental technologies and policies needed to address climate change: renewable energy, nuclear power, carbon pricing, urban planning reforms, and more. What we lack isn't technological capability but rather the collective will to implement these solutions at scale. Although there’s more research to be done, the real challenge isn't finding new solutions—it's mustering the political and social momentum to deploy the solutions we already have.

The irony is that if we can achieve the level of technological sophistication and social cooperation necessary to build and maintain a spaceship capable of supporting human life for millennia, we would already have solved the problems that supposedly necessitate leaving Earth in the first place. Our planet, with its self-sustaining ecosystems, abundant resources, and life-giving sun, remains humanity's best hope for long-term survival and prosperity. Rather than planning our escape, we should focus on being better stewards of our cosmic oasis.

While the Sphere offers a breathtaking tribute to Earth's beauty and humanity's technological achievements, its underlying message misses the mark. Our future lies not in abandoning our home planet but in learning to live sustainably upon it. The solutions may not be as glamorous as interstellar travel, but they are far more achievable—and ultimately, far more important for humanity's future.

In this post, I’ve compiled data from the aforementioned books (primarily Generations) and other sources to highlight troubling cultural trends. While most of the data focuses on the US due to the accessibility of the data (and my familiarity with it), these trends likely reflect a broader phenomenon. Each topic merits its own discussion, and this post is meant to serve as a collection of data to spark those conversations.

Cultural changes, both good and bad, often trigger reinforcing feedback loops. For example, an increase in crime could lead to more poverty, which could lead to more crime, and so on. So it’s very hard to separate out exactly what is the cause and what is an effect. However, I’ve tried to generally separate the data into the following:

• Indicators of decline,

• Cultural shifts, and

• Potential root causes

These are all interconnected so the boundaries aren’t always clear. Many of the cultural shifts could be, and probably are, both indicators of decline and root causes of other problems.

Evidence of Cultural and Social Decline

We’ve become sadder by some measures

There’s evidence that the number of days people experience poor mental health is increasing, especially among people without college degrees.

In the graph below, you can see a split—people with 4-year college degrees seem to be doing better, but those without are much worse, especially since 2000.

Since income and education are so tightly interrelated, we shouldn’t be surprised to see the same thing in the graph below with income.

Deaths from Drugs

The number of deaths from drugs and alcohol has shot up in the US, but it has not for other regions.

Confidence in Institutions

Confidence in institutions is also way down. I know I’ve certainly felt that in response to COVID and other recent events.

We shouldn’t extrapolate too much from world leader approval ratings at a single point in time because global macroeconomic effects like inflation can play such a large role, but it is worth noting that all the G7 leaders are unpopular. These are all democracies of some form and five out of seven of them are more than 30 points underwater.

Cultural Shifts

Optimism to Pessimism

This shift in cultural mood is particularly visible in our literature and media. Where the early 20th century celebrated utopian visions of the future, today's science fiction overwhelmingly trends toward dystopia. Consider an editorial from the New York World on January 1, 1900:

The world is optimistic enough to believe that the twentieth century… will meet and overcome all perils and prove to be the best that this steadily improving planet has ever seen.

Such unabashed optimism about humanity's future would be ridiculed as naive and uncaring in today's newspapers. Technology and material progress were once thought of as key to unleashing human progress.

Compare that, for example, to the way progress was talked about in the US court decision Calvert Cliffs' Coordinating Committee, Inc. v. U.S. Atomic Energy Commission, 1971 (hat tip: Margin Revolution) [emphasis added]:

These cases are only the beginning of what promises to become a flood of new litigation — litigation seeking judicial assistance in protecting our natural environment. Several recently enacted statutes attest to the commitment of the Government to control, at long last, the destructive engine of material ‘progress.’

These examples are merely illustrative. These are anecdotes and certainly don’t prove anything, but I think they will resonate with people’s own experiences. We live in a vastly different cultural era than a century ago, one that, in many ways, feels far less optimistic. The New York World editorial was written before the world wars, the Holocaust, and the realization of the immense harm humanity could inflict on the planet.

Things changed in the 60s and 70s. Paul Erlich published The Population Bomb in 1968, a book that warned of an impending crisis due to overpopulation, predicting widespread famine and societal collapse if growth trends continued unchecked. This, along with other influential works of the time, fueled a growing sense of ecological awareness but also a profound anxiety about the future. Events like the Vietnam War, the oil crisis, and rising environmental degradation fed a cultural shift toward skepticism and caution. In contrast to the boundless faith in progress that characterized earlier decades, the 60s and 70s brought a wave of concern over the planet’s finite resources and humanity’s often destructive impact on the world. This period marked the beginning of a shift in public consciousness, where the potential for disaster seemed as imminent as the potential for progress. And that sense of vulnerability and uncertainty has only deepened in the decades since.

At times, I can’t help but feel a sense of a declining civilization. Simply walking through any major city in California makes it clear that something is amiss (for instance, here’s a glimpse from San Francisco).

Safety Culture

There are lots of ways to measure cultural changes. None of them are perfect, but each adds its own evidence. One thing that Jean Twinge did in Generations is to look in the Google Books database for instances of words. I found it to be an interesting lens through which to view culture.

Here she uses that to look at the incredible rise in safety culture and decrease in risk-taking.

You can see the same thing in the New York Times graph below. Several different points are quite clear from it. One, as Jonathan Haidt points out in great detail in The Anxious Generation, is that sadness starts to shoot up as kids are spending more time on smartphones (and, in particular, on social media, although that isn’t shown in the graphs below.) But it’s also worth noting that as the alcohol and sex fell, so did happiness. It’s not a radical hypothesis that maybe some of that partying was fun.

Individualism and Selfishness

If the Google Books analysis is to be believed, there’s been a tremendous increase in people writing about “getting” things, especially compared to “giving” things.

Here you can see other individualistic phrases.

There’s also lots of anecdotal evidence of a change in our outlook. As Thomas Sowell once said:

There was a time when we honored those who created the prosperity and the freedom that we enjoy. Today we honor the complainers and sue the creators. Perhaps that is inevitable in an era when we no longer count our blessings, but instead count all our unfulfilled wishes.

Materialism

There are lots of people throughout recent history who have seen this decline, even early on, and called for a halt. In particular, German author Ernst Jünger often criticized the materialistic tendencies of modern society, suggesting that they lead to a loss of deeper spiritual and existential values. He believed that an overemphasis on material wealth and technological progress could dehumanize individuals and reduce the richness of human experience.

In his book The Worker (1932), Jünger explores the concept of the "Gestalt of the Worker" and the rise of technology. He argues that technology and industrialization have created a new type of human being who is more focused on efficiency and productivity than on traditional humanistic values. This, in his view, is a manifestation of materialism. His concerns about this transformation deepened over time, as reflected in The Forest Passage (1951), where he writes:

Man has immersed himself too deeply in the constructions, he has devalued himself and lost contact with the ground. This brings him close to catastrophe, to great danger, and to pain. They drive him into untried territory, lead him toward destruction.

Antipathy Towards Progress

Perhaps an important component of this is an antipathy towards progress. Progress, what once seemed like an unalloyed good, is now something we’re warned about in The Atlantic.1

Others have noted the same wariness towards growth and progress. As co-founder of Ethereum Vitalik Buterin said:

In some circles, it is common to downplay the benefits of technology, and see it primarily as a source of dystopia and risk. For the last half century, this often stemmed either from environmental concerns, or from concerns that the benefits will accrue only to the rich, who will entrench their power over the poor.

Roots of Progress fellow Madeline Zimmerman noted the same change in attitude towards progress:

If Vannevar Bush had published Science: The Endless Frontier in 2023 rather than 1945, he would have been pilloried. We know this because Marc Andreessen provides us with a pretty good controlled experiment with his recently published The Techno-Optimist Manifesto.

Techcrunch asked “When was the last time Andreessen talked to a poor person?” NYTimes called it a “horrifying, silly vision.” Washington Post, a “self serving cry for help.” Gizmodo rather confusingly labeled it a “Unabomber-style manifesto.” For comparison, here are excerpts from Bush’s and Andreessen’s pieces:

Without scientific progress no amount of achievement in other directions can insure our health, prosperity, and security as a nation in the modern world. - Bush

Technology is the glory of human ambition and achievement, the spearhead of progress, and the realization of our potential. - Andreessen

Polarization

Polarization has increased. There are lots of graphs that show this (see Ezra Klein’s Why We’re Polarized for far more detail), but you can see it quite easily in the fraction of the population that considers themselves either “very conservative” or “very liberal”.

It’s also interesting to note that though people are more polarized, they aren’t as positive about political parties anymore.

Reasons and Explanations

Technological factors

Technology has significantly transformed our culture. Some particularly relevant changes are the rises of mass media and social media, as well as the increasing accessibility of immediate gratification. Another factor, though less often thought of as technology, is fentanyl, which has undeniably become a major contributor to the rise in deaths from drugs that we saw earlier.

Mass Media

A fundamental shift occurred in the 20th century that changed how people construct their worldviews. Prior to this era, people primarily formed their understanding of reality through direct, personal experience, supplemented by limited media exposure through newspapers, books, and radio. Their perception of the world was grounded in what they could see, touch, and experience firsthand in their communities.

The widespread adoption of television, followed by the internet, has dramatically altered this dynamic. Today, people increasingly build their worldviews based on manufactured versions of reality created by others. This mediated experience of life means that people compare their ordinary lives to heavily filtered (or entirely fictional) versions of others' experiences.

This disconnection from direct experience extends beyond just our social and cultural understanding. Modern life increasingly isolates us from the natural world we evolved to inhabit. Surrounded by artificial environments and digital interfaces, we've lost regular contact with the physical world that shaped our species. This separation from nature and authentic experience may contribute to many of the mental health challenges and cultural shifts we’ve seen.

Social Media

The graph below is from Generations, but there is far more detail about the impact of social media in The Anxious Generation. The data are disputed and there are lots of people who think this is due to a change in measurement rather than an actual change. There’s plenty of back-and-forth about this on Jonathan Haidt’s Substack, After Babel.

Instant Gratification

I’ve also heard of a “rise of dopamine culture”, where we’ve gotten so used to immediate gratification it’s numbed us to the joys of delayed gratification and the satisfaction that comes from persistence and effort.

Economic Insecurity

Whether inequality has increased or not is debated, and certainly the amount depends on exactly how it is measured (see, for example, The Myth of American Inequality, which I may do a book review of some time). However, there’s still a perception that inequality has increased, and some measures suggest it.

Generations also showed a strong connection between income inequality and lack of trust in society. This is an interesting trend because even if post-tax inequality hasn’t risen (a claim made in The Myth of American Inequality2), income inequality certainly has, and, based on this graph, could be tied to a lack of trust in society. This suggests that government transfer payments aren’t a sufficient replacement for income—people don’t just need money to eat, they need to go out and work hard to earn the money to eat. Obviously, there are lots of other confounding factors that occurred during this time, but I still thought it was interesting.

Conclusion

Despite the remarkable technological and material progress we've witnessed over recent decades, the data presents a compelling picture of cultural decline across multiple dimensions. From rising mental health challenges and drug-related deaths to decreasing institutional trust and growing polarization, we see concerning trends that shouldn’t be dismissed.

This post isn't attempting to fully explore these problems or propose solutions. Rather, its purpose is simply to collect and present the data, establishing a foundation for deeper exploration. Each of these trends merits its own detailed investigation and discussion. While the future trajectory of these changes remains uncertain, the evidence suggests something significant is occurring in our culture—something that deserves our attention.