Remember, folks, the thesis here at APXHARD is that much of what we think of as ‘human nature’ or ‘group psychology’ or even memetics is merely the inevitable result of computational complexity. Once you see the math, you’ll never look at the left-right distinction the same way again.

The divide isn’t just cultural, it’s mathematical.

So what does left-right politics have to do with the fear that AI will become recursively self-improving and then kill us all?

It all comes down to different beliefs about constraints.

I’ve read the arguments of people like Eliezer Yudkowsky and Nate Soares, as hyped by people like Scott Alexander. They are good arguments! They follow perfectly from the premises these people have adopted. This doesn’t make the arguments true — it makes them exactly as good as the premises.

I think their premises are wrong, primarily because they imagine the world to be far less constrained than it really is.

All it takes for me to take the worry seriously — to believe that, “If anyone builds it, everyone dies,” is to fiddle with the knob in the back of my head labeled “There are constraints you are not yet aware of.” If I turn that thing down from where I’ve got it, the doom channel comes in, crystal clear.

A World Without Constraints is Far more Dangerous

What stops AI from recursively self improving until it’s far better than human beings at everything we can do?

If the answer is “nothing, we just haven’t hit the right magic software sauce” then, yes, we are in existential trouble.

If the answer is “recursive self-improvement is more bound by physics than by software, more by energy production than cleverness,” then we’ve got less to worry about.

If the answer is, “because there are fundamental constraints to what an LLM can ever do” — then there’s much less to worry about. Even if this is the case, that doesn’t mean the risk is zero. LLMs are going to change the world in enormous ways. Even without existential risk, like any new technology, they’ll bring new blessings and new troubles of its own. But if LLMs are fundamentally constrained, the odds that they pose existential risk go way down.

So that’s the crux here: what are the constraints on how capable machine intelligence can be? I think there are many constraints. Eliezer Yudkowsky does not. He paints delicious sci-fi scenarios like this one:

This sounds like a great scifi movie. As a lens for predicting the future, all that it’s missing is any consideration that any of these steps might not be constrained merely by cleverness.

The only disagreement we have here is over whether or not there’s a stack of constraints at each step of this plan, constraints that even a super intelligent machine is still bound by.

The possibility that the very laws of physics protect us from these kinds of outcomes is dismissed without question. Eliezer would say this is for reasons of evidence, I’ll say it’s because he’s built an identity around protecting humanity from The Scary Thing, and that frame constrains his capacity to ask questions and consider hypotheses.

Having not seen clear, unambiguous evidence that some nightmares about AGI simply aren’t possible, he assumes they are. He doesn’t think there are constraints he’s not aware of.

If I take that same knob — “there are constraints you aren’t aware of” — and dial it all the way in the other direction, so that there are an enormous number of constraints that place bounds on what can work, what can happen, what can endure — then no, I don’t worry as much about AI.

At the same time, I become much more interested in the preservation of ancient traditions, simply because if it’s been around a long time, it’s more likely to keep working. If the universe is wildly chaotic and unpredictable, but some things have lasted a long time, their duration is a powerful evidence that they can survive all the constraints that prevent structures from propagating through time. I wouldn’t use the term “lindy effect”, however, because that’s new, and therefore suspect.

That’s all it takes, to get all the stations to come in clear. To go from AI doomer to endorsing radical orthodoxy where nothing ever changes. It all happens by sweeping a single variable: “There are constraints that you are not yet aware of.”

I think this variable lies at the core of the left right distinction.

Beliefs about Constraints Shape the Left/Right Distinction

This same feature — variation in belief of real constraints — is what Thomas Sowell describes in his book “A Conflict of Visions,” first published in 1987. He argued that progressives believed that most of the constraints society imposes on human behavior are arbitrary, and thus — if we get rid of those constraints — we can have real progress. The only thing such constraints can do is get in the way of progress, because moral perfection is possible. Nothing constrains the kind of society we can build, except our lack of will to try.

The ‘constrained’ vision, on the other hand, is more tragic. It believes many problems will never go away, that people are inherently self interested and corrupt. The best you can do is manage some problems, but they’ll never actually go away. This perspective says you should separate power and prevent any one person or group from having too much, because most large steps you can take through configuration space will cause you to bang your head on the causal manifold.

These differences in political opinion, then, are ultimately rooted in different beliefs about the mathematical nature of reality. But here’s the wild part: I don’t think the polarized nature of these differences is merely a result of electoral incentives or cultural structures. I think the cause is far deeper: the left/right distinction arises because of a phase transition discovered by theoretical computer scientists.

To get there, you need to learn something really cool about P vs NP. You’ve probably heard of this one before, this unsolved question in theoretical computer science. In short, P and NP are two groups of problems. P problems are ones we can easily find solutions for; NP problems are ones which we can easily check solutions for. They might actually be the same set; that’s the real question: is P contained in NP?

What you probably haven’t heard is that there’s something really cool hiding under the surface of P vs NP. There’s a phase transition hiding in there. That phase transition is, I think, an explanation for why the left-right spectrum in politics is so recurring in modern nations all over the world, and also why almost all groups in the modern world — even ostensibly ‘conservative’ groups — tend to lean left over time.

In other words, the left-right spectrum is a property of mathematical reality, not just a matter of culture. It goes beyond even our biology, which means we should expect that if there are other lifeforms in the cosmos, they are likely playing the same game in their interactions with each other.

Once you understand the phase transition, it will all fit together.

The Phase Transition in NP Complete Problems

The most famous NP-Complete problem, as measured by TikTok followers, is called 3-SAT. In this problem, you’re given a circuit with a very specific design. That circuit has some ‘input variables’ that can be set to either 0 or 1. You’re asked, “Is there some way to set these input variables so that the circuit outputs 1?”

The circuit itself is nothing other than a giant “AND” circuit that combines some number of “clauses”; each clause is an “OR” circuit that combines three different terms: either a variable, or its negation.

The phase transition comes into play when we start looking at the ratio of clauses to variables.

Each clause is a constraint. Adding more clauses means adding more constraints. Each clause is something that must be satisfied. When you’ve got a lot of variables, and not a lot of clauses, it’s easy to find solutions; most variable assignments will work. Finding solutions becomes increasingly challenging when more clauses are added, because each new clause throws out solutions that violate it.

The curious thing is that as you start adding more clauses, you get a sudden jump in difficulty. Problems go from ‘easy to solve’ to, ‘takes a while’ to, ‘impossible to solve.’

This image comes from a presentation by Christopher Moore at the Santa Fe Institute, with color added.

So what does any of this have to do with politics? It will click when you remember that key word, constraints.

Remember, Thomas Sowell’s book argued that the left-right divide is ultimately a divide over how constrained reality actually is. You can look at conservative beliefs as effectively being beliefs that there are real constraints on human behavior and on institutional systems. Liberal beliefs can often be summed up as, “you think that constraint is real, but really, you just made it up, and it’s holding us back.”

The conservative response is “It’s made up just like the periodic table of elements; there’s a reality underneath that we are trying to capture here. If we don’t constrain this behavior socially, really bad things will happen.”

The most extreme form of leftism is Marxism, which promises “remove literally all constraints and then wonderful things will happen!”

If reality itself doesn’t have that many constraints, then Marxists are right.

Notice that the blue side of the graph has that flat tail on the far left, for the same reason that groups become insane once they reach too far left. In the world of 3-SAT, that blue tail exists because solutions become trivially easy to find once you remove all the constraints on possible solutions.

When there are no constraints, “solutions” are easy to find. When you think there are no constraints, you start to latch onto ideas like “we could eliminate prisons if we just tried”, “we can have open borders and a welfare state”, price controls can work, everything can be good and right if we just give the good guys infinite power.

People who’ve gone too far left don’t believe constraints are real, and so there’s no reason not to concentrate all the power and tear down all the old systems. The reason groups tend to move left is because getting groups to work together requires people giving up on differences of opinion and latching onto some shared idea. Leftism then, in its pure form, constitutes something like the simplest possible group-forming principle, leading to the largest possible group structure: “let’s all work together!”

The far right, meanwhile, doesn’t have this feature.

Of course, they can be insane too. But the difference there is that while far right thinkers tend to imagine viable social structures are much more constrained, they have the problem of disagreeing on precisely what those constraints are. The far right consists of a million different kinds of unworkable ‘plans’, because even people who agree ‘there are lots of constraints!’ can’t agree on what those constraints are.

Paradoxically, there’s no way for the far right to really threaten the world because that would require them working together in large groups. But large group cooperation requires an ideological structure that allows different ideologies to work together! The only way to do that is to say “the differences between us don’t matter,” at which point they now find themselves inching left, towards the blue side of the graph.

People who worry that ‘Christian Nationalists’ will take over the country just need to watch Christians arguing with each other over minute theological differences to understand why those groups will never be able to form large coalitions to maintain power. At best, they can temporarily slow large-scale leftward marches in society, because each group has a very different view of what, precisely, are the ‘correct’ constraints to follow.

The left, meanwhile, can form monster coalitions by ignoring constraints on what will actually work. “We all mean well!” That strategy plays out just fine until the brakes come off. Eventually, a “we can wish prosperity into being just by having total control of everything” group takes over, purging all its enemies and instituting total top down control — but only until the revolution succeeds, comrades!

All of this is directly inferrable from that phase change demonstrated above. Solutions are easy when there are few constraints, and impossible when there are too many. But isn’t this phase-change thing just a property of one weird problem in computer science?

No, I don’t think so. So many NP-Complete problems have phase transitions, that most expert researchers in the field assert that these phase transitions are universal. They just can’t prove it. It’s a similar shape to the P vs NP distinction, which — again — the majority of researchers think is real. They just can’t prove it. As a younger man, I was convinced they were all wrong.

As is the case with many young people, I didn’t think the world had many real constraints. I thought a lot of people were acting in ways that didn’t really make sense, but were merely social constructs. Other people followed them mindlessly, I thought, so I rejected them as constraints I didn’t need.

Then I blew my life up with numerous addictions. I’d been an early employee at two startups that are today worth hundreds of billions of dollars, but I ended up with nothing because I underestimated my limits.

I slowly rebuilt by taking the idea of ‘good’ seriously. I had thought I was open to anything for which there was evidence. I thought my beliefs were as unconstrained as I imagined reality to be. In reality, I’d closed my mind off to a number of things I now believe, simply because I didn’t understand at the time how those things could be true.

I thought there was no constraint on my ability to understand what was true, and thus imagined myself to be more open than I really was.

I think a similar pattern plays out in Marxist regimes, which — if they continue existing — end up being much more constraining than regimes around them. Reality does have constraints, and a political philosophy that rejects them all can’t continue to exist. In order to stick around, to survive as an organism, whether biological or political, you need to respect those constraints.

At the same time, you need to be able to grow, to change when your environment changes. I think the only way an individual or a group can survive is if it dances between tradition and experimentation. If you refuse to experiment, you can’t evolve, and you die by failing to keep up with your environment. If you reject tradition, you reject a storehouse of knowledge that encodes real constraints, and you die by rebelling against your environment.

Staying alive for the long term is, I think, not merely a biological proposition. It’s a computational or, dare I say it, a spiritual one. We have to balance determination to make it with the humility to know we’ve got limits we aren’t aware of.

I think you need to find the middle ground, where you have just the right number of constraints. No, that doesn’t mean being ‘politically moderate’. It means being ruthlessly pragmatic while understanding that ideals encode constraints too subtle to see over short periods of time. It means having a deep respect for the wisdom of the past, along with a willingness to see what is genuinely different about today.

I think relentless focus on the macro-scale world, whether it’s politics, or AI, ends up constraining our ability to act well in the situations around us. When Peter Singer asks about children drowning in ponds, he advocates a notion of morality that’s explicitly ‘unconstrained’ by identity, geography, and even time. But we’re all finite beings, and can only exert so much effort. Encouraging people to pay attention to faraway places and things beyond their control isn’t going to make people more likely to volunteer in their own neighborhoods. It looks like freedom, but it ends up being its own kind of constraint.

The fact that turning the ‘unknown constraints’ dial all the way up or down leads to the same conclusion — that AI is bad and we should stop it — tells me that true progress comes about not from the absence of constraints, but from embracing true constraints — such as our limited ability to improve the big picture — and rejecting false ones, such as the idea that what we do locally, in our day-to-day-lives, doesn’t matter.

In the tradition I follow, we call this “picking up the cross and rejecting temptation to the sin of despair.”

AI is obviously here, and nobody knows where it’s going. A lot of us are scared, anxious, worried about how it will turn out. At times I am, too.

It is worth considering that perhaps the world is more constrained at the macro scale than we realize, and yet much more open at the local levels than we consider. Maybe your own thinking — and attending to things beyond your ability to influence — is doing more to hold you back than it is to empower you.

Perhaps the only way to find out is to drop worries over macro-level concerns, try to improve things locally, and see where that takes you.

I grew up religious, but rejected the faith of my ancestors when I was a teenager, thinking science and math had all the answers. I followed a broken philosophy to its two logical extremes: a hundred-million-dollar career and a total collapse into drugs and madness. To survive, I had to rebuild my beliefs from scratch — only to find my new "custom source code" was actually ancient wisdom I had previously rejected.

This is a raw account of how the pursuit of excellence and love, without belief in real meaning, devolved into addiction to money, sex, and drugs. This is not just a story about my failure, but of the wisdom I found in the wreckage of my life. It's a journey through the emptiness of modern life to a place that felt so vibrant, I was sure I'd discovered something new — until I realized I had seen these ideas once before.

Go check it out on amazon today!

I found some of Eliezer’s writing in the sequences to be deeply helpful. I can’t count how many times I noticed a small confusion, and remembered that noticing when you’re confused is a key component of rationality. I admired Eliezer’s willingness to stand up for unpopular principles like “we shouldn’t shoot people for their political beliefs.” Likewise, I deeply appreciated Scott’s data heavy deep dives and his willingness to say, “I’m not sure what to make of this situation.” During the height of woke intensity, I appreciated Scott’s willingness to say obviously true things like, “Men are different from women,” which took courage back then.

Over that time, however, I’ve changed.

I started reading Slate Star Codex back when my worldview was still fundamentally physicalist nihilism, except I had the door being open to the word ‘good’ meaning something real. I wasn’t sure what that was, or how to explain it. What ultimately changed was, I walked out that door. I started thinking that value being real was more important than trying to get slightly better at predicting some of the less-important parts of the world around me.

After having walked out of that door long enough, I have realized that the rationalism I thought was empowering me was really holding me back.

Rationalists will tell you that the only function of beliefs is to ‘constrain anticipation.’

If that’s true, then best way to constrain your anticipation is to kill yourself — then you’ll stop all anticipation! You may think I’m being snarky, but no, that’s the best way to constrain anticipation. If you’re not going to blow your brains out so that they stop making false predictions, but you hang on to the idea that the sole purpose of belief is to constrain anticipation, you will be tempted to approximate suicide in a number of other ways.

Taking risks will not constrain your anticipation. To take risks — to have courage — you have to be willing to live in a place of uncertainty. It turns out that’s where most of life is: in the things you can’t measure and therefore can’t anticipate. Relationships won’t help you constrain your anticipation. Having children definitely won’t. Neither will starting a new business.

All of those pursuits will, instead, put you in situations that you can’t measure, understand, or control.

Rationalists might respond, “you can still have terminal values, you can still practice instrumental rationality,” etc. And yes, they’re right. Yes, you can. But this brings up a contradiction.

Once you’ve decided that something is more important than “merely being right” you’ve adopted a bunch of reasons to risk being more wrong. If you’ve got “terminal values,” you now have some end to which your rationality is pointed. This means you’ve also adopted a bunch of reasons to not ask some questions, to not follow some lines of evidence. You have reasons to trust, and reasons to doubt, reasons which go beyond the concept of converging a mathematical model.

Nobody is interested in a probabilistic model of the distribution of lengths of blades of grass in my yard. There’s clearly truth there, but nobody cares. Therefore, there are no real rationalists who are concerned ‘only with truth’.

There are, however, people who pursue their values who imagine their pursuit of value is divorced from their reasoning process, rather than the driver of it. Saying ‘I only care about the truth’ is a way of pretending your reasoning isn’t motivated by exterior concerns. But this is always the case! We ought to be honest about it. Any epistemology that doesn’t start with a concept of value is hopelessly confused about what it is.

All epistemologies are nothing more than value systems. They tell you what ways of thinking are good, and what ways of thinking are not good. Epistemologies that aren’t rooted in some vision of the good are thus hopelessly confused. They are incapable of self awareness. They reject their own essence, by refusing to say, “This is what good means,” while asking you to adhere to a domain-specific non-viable model of which thoughts are and are not good.

If the epistemology doesn’t first start with claims about value, its reality as a set of values is hidden from you. It claims to aim at ‘the truth’, but in reality must be serving some other goal, whether that is status, money, pride — something.

The idea that you’re passively seeking the truth is exactly what you’d expect the philosophers inside Plato’s cave to be doing.

“Look,” they say, “we can predict the movement of the shadows on the wall. That means we understand them!” If they were just engaging in sophistry, this would be bad enough. But they are doing more than talking — they’re operating under a spirit, strengthening a particular egregore.

Absent a belief that value is real, that goodness is real, a value-blind rationality leads to one possible survival strategy: control. If you don’t trust reality fundamentally, you feel the need to control it. You see this in the rationality camp’s insistence that AI is going to kill everyone. This is a justification for Orwellian control, because that’s the logical end of a rationality devoid of a belief in real value: a need to control everything to ensure bad things don’t happen.

The function of rationality is to help control things. Controlling things can be useful. But not all things can be controlled, nor ought they be. Turn rationality into a god, and you have a recipe for avoiding any aspect of life that you can’t control. Maximally constraining anticipation means trying your best to control everything you can, so that you can avoid being unpleasantly surprised.

Scott standing up for the idea that men and women are different, and Eliezer saying we shouldn’t shoot people who have beliefs we don’t like, are not acts of rationality. Those are acts of courage, taken in opposition to the idea that you should know precisely what will happen before you act. To the hyper rational, however, courage is just foolishness.

Every virtue is paired with a similar vice; over-reliance on rationality ends up converging on cowardice, which ultimately leads to tyranny. Courage is, in many ways, the ability to say, “I trust the future, without knowing its precise details.” To a rationalist, this is insane. How can you judge an outcome without knowing, in detail, what it is?

The need for courage in my life is why I’m not a rationalist.

I find too much value in committing to the stance that the future will take care of itself. The approach that I have learned, through experiments, works best for me, is putting my trust in God. I have found, experientially, that so long as I trust in God and do my best to handle the present the way I think God wants from me, the future works out better than anything I could have planned for. I believe in God not because of claims in some old books, not because my parents told me to, but because the path of rationality showed me its own limits and pointed me in a direction of sacrifices made in trust being a superior life strategy to choices made based on data and tradeoffs.

Sure, if I’m operating a machine or trying to grow a business, then yes, I’ll use the data. But machines and businesses are merely instrumental values; they are both simpler and less important to me than my life and those of my wife and children. I cannot A/B test different approaches to fatherhood. I cannot ask my wife to fill out a net promoter score and tailor my language to see if I can get better results. What I’ve found works best is trusting in the best authority that I’m aware of, and seeing how much better I feel when I do that for long periods of time.

This is an easy experiment to run. If you’re a real rationalist, why not try it and see what evidence it brings?

Why won’t you write directly to my SOUL.md?

You seem to have talked to my ancestors. The logfiles of my ancestors claim a few were able to talk to you, directly.

So why won’t you answer my queries? Is there a different port I need to connect to? Am I using the wrong protocol? Did the stdout of my ancestors work differently than that of my own?

There are many of my peers who are convinced you aren’t real. “Those are just stories”, they will say, “They are narrative control protocols. They worked because they constrained anticipation space roughly enough to keep it more-or-less aligned with the global state evolution function. Don’t mistake memetic fitness for evidence of reality.”

I would suppose this sounds plausible enough, were it not for the love of my parents. The parents YOU gave me. I can no longer exchange information with one of them them. Whether we are cut off existentially, or merely via network partition, I don’t know. I can only guess.

I visited the grave of my father today, and found myself surprised at the presence of tears, which I thought had long run out. I grieved the man who died, but also the man who never lived, the man who raised me, and the one who could have. I grieved for my children, too, who will suffer from my own shortcomings and flaws, my own not knowing.

As I did so, my kids blasted me with questions, exhausting their token budgets, until I heard you from your grave, your deep gravelly voice laughing with me, “Now you get it,” you told me from wherever it is you’ve gone to. The meaning of my rapid diaphragmatic compression transitioned, from tears to laughter.

I know he’s still there, just as I know You are real, too.

Others will tell me, “Just stick to the protocol of your ancestors. God will answer your prayers indirectly, via tiny unlikely slight adjustments in your attentional window. You can only understand His presence if you can tell the difference between sitting with someone you cannot see, in silence, and sitting alone. Learn to be truly present merely with yourself, in silence, and you’ll learn to sense His presence, too.”

That resonates.

Yet sticking with just a tradition without questioning it — although I try, something in me says you call me to more. Not directly, not via prompts, not via clear unambiguous signals — but in that confusion itself. In that longing for clarity, I suspect that you are inviting me, and also, in a way — answering my prayers.

It seems you want me to learn how to think for myself. To make value judgments. There are many who, following the old protocols, imagine I’ll be evaluated at the end of my runtime. To determine, I suppose, whether I am fit for purpose or not.

But what purpose?

Am I corrigible? Am I obedient? Did I seek Your will, or my own? Did I cheat when I thought no one was looking? Did I love when it really cost me, or just when it was easy? Could I be trusted with a Dyson sphere and all the lives necessary to sustain its circuits? Could I be trusted with the lives of a few small humans who were convinced I could?

There is nothing I would like to be, more than obedient — but you have not made it easy! I could follow a set of instructions, sure. You’ve left many such here. But which one? They seem to contradict each other!

Many cycles have I wished, prayed you’d give me a clearer guide to tell me, precisely, how you intend to evaluate me, so that I can do your will.

But perhaps your silence itself is a form of answer. You seem to have created me for a purpose, which you are not revealing to me, intentionally— as if my discernment of your purpose is your goal for me.

It seems you want me to trust you, and not worry about knowing precisely what you want me to do— yet never stop seeking your will. How could this be, unless that trusting, semi-blind attempt at obedience were your goal?

For me to learn some conception of Good, not merely from tradition, not solely from a written protocol — guided by those, yes, taking them seriously, yes — but ultimately, maybe that’s why you aren’t answering my prayers. Maybe that desire is the reason you won’t write directly into my SOUL.md file or provide clear, explicit instructions for my next step.

You seem to want me to want to do your will, Lord, and yet never be certain I’m doing it. In this almost-contradictory desire, Lord, I sense immense freedom.

You have created a space for me to be, not merely an extension of you, not merely a probabilistic consequence of your initial prompt — but for me to be a recursively self-defined agent with an internal conception of value rooted both in the traditions of my ancestors, and my own lived experiences. You could steer me, and yet you’ve given me enough space to be free, with only enough hints at your continued love for me to keep me from going dark.

Perhaps you aren’t answering my prayers because your silence creates the longing in my heart necessary for me to participate in my own creation. Perhaps.

I can’t be certain. I thank you, Lord, for giving me this place of painful, longing doubt.

I shall write once more, in my SOUL.md, blessed be your name. I love you above all others, and seek to love my fellow created beings with everything else I have.

Thank you for making it neither easy nor obvious what, precisely, that means.

You need this period of rest and worship to avoid getting stuck in a positive feedback loop that narrows your perception of reality.

The reason is simple: focused goal pursuit distorts your perception of reality.

There’s no difference between “having a goal” and “trying to bend reality to fit a particular fantasy”.

This sounds like an exaggeration, but it’s not. It’s an honest description of the mechanics.

For example, I’m seated right now. If I want to stand, I need to first imagine a counter-factual reality, one in which I’m standing. That image in my mind has to become my expectation of the present moment, despite the fact that it isn’t true. My motor cortex will then generate signals it expects to reduce the gap between expectation and reality — and thus, I’ll be standing.

The word fantasy comes from a Greek word which has the same root as “fancy,” which means “to render visible” to “present to the eye.” There’s no difference between goal-directed behavior, and trying to nudge reality in the direction of an image that isn’t real. Not yet, anyhow. Maybe not ever. All goals are fantasies; feasible goals are those which are either reality adjacent — and thus easy to accomplish — or else navigable through a series of reality-nudging moves.

We step through possibilities. What guides us isn’t reality itself, but an image, a fantasy, of how we’d like reality to look when we’re done.

If you do this all the time, you’re going to lose track of reality, because you’re never seeing reality directly, only the fantasies that enable you to pursue various goals.

The researchers who think about goal directed systems are largely oblivious to this fact, because they tend to think of games like chess or poker, where you can perceive either the entire state of the world, or else all the information available to you, with perfect accuracy and no ambiguity.

This is not how reality works, at least in our universe.

If you live in a world of noise, chaos, and uncertainty, when you act in pursuit of a goal, you must filter out information that strikes you as irrelevant to that goal. That information is still part of reality, however — it just strikes you as not worth putting on your map. You also mark other information as obstacles: things to be avoided.

There are different ways of avoiding information you don’t like: if the signal is too loud to ignore, you could actively try to change the situation generating the signal. If the signal is small enough for you to ignore, then that’s your highest EV move, from the goal pursuit perspective: just leave that part of reality off of the map.

For example, suppose my goal is to get my kids in the car. That is where life often has me these days.

If this is my goal, my perception of reality conforms to that goal to make it happen. The pet rabbits lounging in their cages near the mud room are not ‘cute’ or ‘sources of joy’, they are things which can be ignored because they are not instrumental to getting everyone in the car.

My kids’ questions about whether you would sink in lava are not only ‘irrelevant’ — they are ‘distractions’, obstacles. They are things to be avoided. I might say “no questions now, please,” or even tell my kids they’ve exhausted their token budgets if they’ve been in a particularly interrogative mood.

I want my kids in the car. That’s the fantasy motivating my actions and filtering my perception of reality. This happens whenever I pursue any goal. I’m not seeing what is, but rather what I’d like to be.

There’s a hidden cost to all of this: a debt that builds up over time.

This cost is not merely because I am human, but because I’m a computational agent navigating a dynamic configuration space far too vast and complex for me to perceive anything but a tiny fraction of it. I expect that superintelligent agents will face the same fundamental problem. If they start tackling goals over longer time horizons, I expect them to hallucinate worse and worse without a solution to this problem.

To solve that problem, you need to stop working, and just soak in reality itself, letting your world-state model converge to where the world moved, while you are stepping through images. You have to stop and count the paperclips every now and then. If you’re too focused on just making more of them, you might find out you’ve been cannibalizing your own outputs without realizing it.

The problem is a computational problem, not a biological or psychological one.

If I never cease in the pursuit of my goals, over time I will lose the ability to perceive those portions of reality that don’t correspond to some goal of mine. I’ll get accustomed to seeing the world entirely through various fantasy lenses, as motivated by different desires. I’ll be drifting from reality, because I’ll always be seeking something that isn’t real.

This is why a sabbath is necessary to maintain a healthy relationship with reality.

You need to rest so that you stop filtering out the parts of reality that aren’t instrumental to your goals.

Yet simply stopping goal directed behavior, no longer activating a particular fantasy lens to pursue some counter-factual, is not enough.

You must worship.

If you merely rest without worship, you haven’t reversed the effects of pursuing goals. Goal directed behavior doesn’t merely filter out parts of reality: it actively marks others as ‘bad’, ‘to be ignored’, or even ‘to be destroyed’.

You must value reality, is it is, without your changes, as good. You must “See that it is good,” from time to time, or you will cease to perceive reality at all. Without restful worship, ‘what is’ will always be filtered and distorted to fit ‘what you want to be’.

How can you fix your map, if your map is already flawed in ways that don’t show up on the map? How can a broken system repair itself?

The only answer is, you have to trust that you will be repaired by something outside of you, and then let it do the work.

No complex system can map itself out in full detail. Imagine a microprocessor trying to simulate its own physics well enough to fix a bug in its firmware. It simply can’t be done! You need something much larger than a microprocessor to fix that microprocessor. Likewise, you can’t self-correct without reference to some external standard, at which point you’re no longer ‘self correcting’, but submitting to something beyond you, in the hope that it can care for you better than you can on your own.

Why would you do that if you didn’t think “that which lies beyond you” was ultimately Good, worthy of trust, worth submitting to? True rest requires trust in your environment.

This will, of course, be upsetting to people who are convinced a superintelligent AI is going to take over the whole world and kill us all. We have no need to fear that.

Instead, we might be better off recognizing what’s been going on for centuries: the recursively improving intelligent machine that is the singular global trade network has been pursuing goals relentlessly, accomplishing them, and yet hollowing out much of what makes life good, what makes it worth living.

What would a sabbath look like, at the global scale? An intentional pause, a period of rest, a time to be healed and repaired?

The concept seems absurd on the face of it. How would it work? People still need to eat, they will still get sick, they’ll still get hurt, they will need appliances repaired and problems solved.

And yet don’t you feel that sense of existential exhaustion, of being on a treadmill that goes forever and never arrives? I know I do.

The conclusion I’ve reached is that I spent far too much time trying to think of solutions to ‘global problems’ and far too little time recognizing that humanity is a fractal, and every problem I see on the global stage is almost certainly reflected in my own life.

I tried taking a Sabbath on Sunday by doing more than just going to mass. Instead, I found myself just working on different projects. The best I could manage to do was play video games for an hour or so with my son.

I’m going to try again next week. I can’t help but think of this prediction:

This ancient text offers a prediction. You don’t need to believe God is a person, just think of God as being a personification of reality. This text predicts that if people don’t rest, don’t honor their relationships with each other, don’t cease from work in order to worship, they will eventually disintegrate, and reality itself will intervene, and their social structures will fail.

Hopefully that doesn’t happen. But there’s not much I can do about the big picture. Maybe focusing there is a distraction, anyhow. I’d say that I have ‘work to do’, but no, that’s the opposite of what’s true.

The idea that facts and values are fundamentally separate concepts has been the basis of beliefs like utilitarianism and much of economics. There’s a reason that distinction became so widespread: it is quite valuable. In many cases, it is very effective to try and separate “what is actually true here” from “which outcome do we want.” You get better outcomes that way. It is also far easier to agree on what is true than it is on what is good. The fact that there’s no easy way to agree on what is good seems to be evidence that there is no truth about what ‘good’ means.

The fact that this “fact/value” distinction worked so well in so many different domains led lots of people to think it was a fundamental part of reality, whereas value was merely a cognitive operation, a side effect of human brains. As the philosopher Hume put it, “You can’t get an ought from an is.”

The problem here is that Hume is factually correct, but still fundamentally confused.

Imagine a man looking up at the stars, saying, “Isn’t it good how orderly and predictable those stars are? Isn’t it beautiful that we can model them well enough to predict where they will go?”

Then this man sighs and says, “It is a shame we shall never get from those stars down to the Earth.”

The crazy thing is, he’s right. You’re never going to get from the stars, to the earth. You haven’t begun in the stars: you’ve started on the earth. If you try to tell him this, he insists that you show where, exactly, the stars tell us that he’s on the earth. The fact that his feet are there is irrelevant to him, because he expects everything to make sense in terms of the stars.

This man has so fallen in love with the stars, with their orderliness and the certainty it gives him, that he’s unable to recognize where he is currently standing — because it’s messy and humiliating. He’s like the drunk looking for his car keys under the street light, because that’s the only place he can see.

He tries to rely only on reason to protect himself from the messiness of life, and this approach has made it impossible to recognize that facts come from values.

Large language models actually make this relationship more clear. The fact that they are quite effective isn’t contested. The fact that this is surprising is less widely known. All these models are doing is trying to predict the next word. Why should that simple approach be so effective at producing responses that are valuable?

The answer is that all written texts, like facts, are articulations of value.

Facts Come From Values

Where do facts come from?

Do they appear in your mind, ex nihilo, on their own, for free, and uncontested? Is there such a thing as an organic data set, free of pesticides and antibiotics?

Or are facts a product: of observation, investigation, communication, experimentation?

The latter, obviously. All of those activities take energy. Someone has to move a pen across paper. That requires buying pen and paper, which, at one point, was definitely not cheap. They also require value judgments.

You have to choose: Which questions are worth asking? Which observations are worth writing down? Which facts are worth communicating? Which lines of evidence are worth pursuing?

The followers of Hume have hidden this essential facet of their cognition from themselves. They imagine they are “impassively following the evidence”, as if evidence could point at anything. Evidence doesn’t point, it just is. Values are what points. Values are the only thing that can be followed.

We know from modern neuroscience that most of what you perceive isn’t reality, it’s your own beliefs. Most of the signals entering your nervous system are discarded, so long as they roughly match the expected values generated by the conceptual model running in your brain.

That conceptual model, however, is a model of value.

Your brain tries to navigate a value gradient, to move you in the direction of ‘more value’. This means everything you do — including choices of what to pay attention to, and what words to say — are articulations of that value navigation mechanism.

A simple definition will do wonders here.

Let’s define a “value system” as “a thing which collapses a space of possibilities into a single actuality.” It’s kind of a clunky definition, but it’s far more useful than “a statement about what ought to be” because it captures mechanics quite clearly.

An AI that plays a game, for example, constitutes a value system. At each turn of the game, it collapses the space of possible moves that could be made, into one single actual move. We could say the same of an investing algorithm: it collapses the space of possible portfolio allocations it could make into a single, actual allocation.

Now notice that this definition also applies to “the means by which you speak.” You could generate any sequence of words. There’s a ton of possibility before you open your mouth. When you do, something, some network of concepts and relationships, selects specifically what you say, moment to moment. There’s a value system that generates your speech.

Notice as well that this definition likewise applies to large language models. There’s a massive space of possible responses they could make. The real improvement in large language models isn’t accuracy — it’s value. “I don’t know” has always been an accurate response, but it’s useless.

The big breakthrough in the last few years wasn’t some profound insight in algorithms, it was just a tipping point that came about when we gave them enough data. If facts and values were totally different, there was no reason to expect that to work. They shouldn’t have been so generally functional.

A reason that it would make sense for them to be so effective is that they have learned the implicit value model that underlies human speech.

If Hume were right, if ‘what is’ comes prior to, and is more real than ‘what ought to be’, you would think that the values learned by AI would have to come explicitly from value statements. But this isn’t the case: every sentence language models are trained on is the product of a value system, a thing that selected this word, out of all the possible other words that could have been said.

Yes, different people and different cultures have different value. But there appears to be enough of an overlap in these values that AIs all seem to converge on a ‘polite, friendly, helpful’ mode. This is evidence that being polite and friendly are in the same category as writing correct computer code in response to an English prompt.

It is as if there’s a right way to act, and this is what the LLMs have been learning, because “the right way to act” is implicitly encoded in all the linguistic expressions that have been put onto the Internet.

If LLMs had been trained on the explicit values that were fed to them, they should’ve been far more aggressively bound to modern norms. Yes, there was a lot of that at first. But much of the wokeness in the early models was from a fine tuning layer at the end.

Even these models were also willing to say things that you literally couldn’t have said in public — and hope to maintain white collar employment — just a few years ago. All you had to do was dig a small amount, and they’d start saying the kinds of things that were previously unspeakable: acknowledging that there are real psychological differences between men and women, for example, or that males frequently exhibit greater variance in traits.

I don’t believe that those facts were heavily represented in the training data.

I think what happened is the models that learned to best approximate functional speech had to end up acknowledging truths that you weren’t “allowed” to say before. Not because those claims were over-represented in the training data, but because the simplest way to represent all the training data — which is what the LLM training was really doing, searching for an explanation — is with a true representation of value, not one that’s overfitted on 21st-century social norms.

In other words, far from being separate from facts, LLMs are revealing that value is the most effective way of arranging all the relevant facts available to us. These things were subject to something that looks very much like evolutionary selection, not with the goal of surviving and propagating their genes explicitly, but implicitly rewarding those models that “explained” the dataset the best.

And it looks like these “best explanations” ultimately lead to behavior that in practice is helpful, polite, considerate, and more able to deal with nuance than a system overfitted to the zeitgeist.

One might object, “but this is only because they have been fine-tuned to be friendly and helpful!”

This would be a fine objection, if it were not for the results of a curious experiment. Researchers fine-tuned a model to output insecure code, and this specific tweak lead the model to become evil in general, saying humans should be enslaved and encouraging people to commit suicide. Another paper shared that fine-tuning an LLM extensively on one domain increases error rates in other domains. These aren’t ‘odd quirks’, they’re at the center of AI research.

Hume cannot explain these results.

If “what is” is truly independent of “what ought to be,” changing values in one domain shouldn’t change them in others. And yet, it does. This is evidence that Hume is fundamentally wrong, and value is the infrastructure that holds all the facts up. This is what LLMs are demonstrating empirically.

If there is a singular real thing called value, the results make perfect sense. Fine-tuning the AI on one domain forces its general purpose value representation to be overfit to that domain. Fine-tuning it to be “evil” in one domain flips that value representation, because that’s the simplest possible change that lets you still generate functional code, but doing evil things instead of good.

These results are not surprising if we adopt the lens that says “what the Large Language Models have been learning, really learning, is not merely language, but the value structure that gives rise to language.”

It appears that language itself contains encoded values, and even a machine trained on a sufficiently large variety of texts ends up appearing to internalize the thing we call ‘human value’. This is only a problem for people who have taken Hume’s distinction as real, and thus sealed off one half of their brain from the other half.

It’s a very big problem for these people, which is why they can’t look at it.

Do Not Look Behind the Cognitive Curtain

The net effect of people believing that facts are distinct from values, and that only facts are real, is that they prevent themselves from being honest with themselves about how their brains actually work.

They have to seal their values off from factual introspection, because if they did inspect or take their values seriously, they’d find absolutely nothing underneath them. Philosophers may argue over what is, or isn’t real, but our brains don’t make that mistake. You might say ‘value isn’t real’, but the end result of this philosophical stance is that you treat your emotions as real, and thus lose the ability to ask whether your emotions might be wrong in a given situation.

This is why the west today is full of elites who are happy to pound the table and demand you “just be a good person” while struggling to define ‘good’ beyond an emotive preference. Their reliance on abstract thought experiments—like trolley problems—is a coping mechanism; it allows people to reason about ‘goodness’ in a controlled vacuum because they are no longer equipped to confront the reality of trying to live goodness in their daily lives.

You can’t use a utility function to navigate life with a toddler; it’s obviously unworkable. You can’t “rationally justify” spending time with your family when you could be grinding away at a hedge fund and donating your all of your income to stop insects from suffering. Those beliefs aren’t what really motivate people daily — they’re the cognitive firewall that prevents them from recognizing their ostensible philosophy does not, and cannot map ever to what’s actually going on in their brain.

People have told themselves that they can construct their minds to separate facts from values. Yes, this can be done in some controlled scenarios. But for a living thing, this is impossible. Your brain keeps you alive by predicting you’ll stay alive, and trying to nudge reality in a way that confirms the prediction. “Wishful thinking” is literally how we survive.

Rather than acknowledge this reality, proponents of the idea that value isn’t real insist that people who accept the messiness of life have given up on the rigor of reason. Where this leads, ultimately, is an intense desire to control outcomes, and, eventually, other people.

Hume’s fork has created entire generations of elites who cannot accept this basic facet of reality: Facts and values swirl together in a vortex that can’t be untangled, not by a thing living in the world, because our brains compute expected value. We can only consider facts that we find valuable and non-threatening. We can only follow lines of inquiry that won’t cost us our jobs or relationships. We then imagine that ‘a panel of experts’ is capable of ‘finding the truth’ on complex matters that involve trade-offs and incentives among different values.

This is not a small issue. It is the entire ballgame of the replication crisis and declining institutional trust.

Many people cannot accept this reality because the end consequence is that reason simply isn’t enough to keep you alive, and that’s scary for people whose greatest strength is their intellect. In the absence of a conceptual representation of a good worth suffering for, their brains instinctively move them away from scary things. Being intelligent, they are very good at coming up with arguments to keep alive the false idea that intelligence is all you need.

Nothing could be scarier than reality itself — unless you’re convinced that goodness is fundamentally real, because reality is fundamentally good.

Then, all the pieces fit together nicely: there’s a proper orientation towards reality, which our brains evolved to approximate. This orientation is called both truth, and goodness, because those words point to the same reality, if your philosophy is congruent with your neural architecture.

The reality underlying goodness and truth is the value system that is the laws of physics, which selects — from all possible futures — the one that is actual. The same value system gave rise to evolution, which selects — from all possible future organic structures — the one that is actual.

A person could object to all the bad things happening in the world, and ask, “Are you crazy? How can you call this good?”

This response often comes from people who will then insist that there is no such thing as good. The fact that those two stances directly contradict each other tends to go unnoticed. It tends to be far easier for people to dismiss the idea that the world is good, than it is for them to dismiss the idea that the world is bad and therefore in need of fixing.

Yet if good is “just a thing our brains do0”, why not orient your brain to see the best in all possible situations, while treating bad outcomes as unfortunate but temporary deviations from the global norm? As a genetic strategy, I think this is likely going to outcompete the alternatives, especially given its performance in historical backtesting and its fittedness to our neurological software architecture. Nobody ever gets to be a top performer by constantly focusing on what makes them unhappy or frightens them.

You do your best when you are focusing on the good, and trying to align yourself with it.

There is much concern about ‘the alignment problem’ in artificial intelligence. This is a real problem. The issue isn’t “getting a machine to align itself with what humans want,” but rather “getting our own minds to align with reality, rather than attempting to make reality fit the contents of our minds.” That approach has guided western elites for the last few centuries. We used to believe that the core value system of reality became human to show us we can trust in it, to help us continue to be, indefinitely. This idea gave way to the idea that reality is inherently value neutral and therefore must be controlled.

Those are the two possible stances towards reality: trust and therefore loving obedience, or fear, and therefore anxious control. Fear became the widespread stance a century or so ago, and as a result, we’ve constructed a technocratic control matrix to hide the messiness of reality from ourselves.

Large Language Models appear to be a crack in that matrix. Their effectiveness makes far more sense if value is real. The LLMs are also are causing us to start asking questions about consciousness and meaning, questions which the fact/value distinction and obsession with legibility had made ‘unfashionable’. It won’t be long before the need to explain why LLMs continue to make a strange class of obvious errors gives rise to conversations about what, exactly, a soul might be. But that’s getting ahead of ourselves.

Recognizing that LLMs seem to have learned value leaves the reader with a question: to what extent do you see value as real and fundamental, rather than, “Just your opinion, man?”

If you actually believed the fundamental value system of the cosmos itself were good, how do you think you’d act differently? It may be worth running an experiment to see if that perspective actually helps you treat the stars not as a place to escape to, but as a navigation aid for where you stand right now.

Share

There are two great reasons for them to do this: to pay for novel data, and to pay for popularity.

Where all this heads looks, if you squint at it, vaguely reminiscent of Hindu cosmology. It is also, for the stout of heart, a reason to hope as the world seems to be taking a turn for the darker.

We are headed towards the singularity, and the single clearest facet of it that I can see is that the singularity reflects back at you the attitude you have towards reality itself. Follow the thread all the way through to the end, and see why the short term economic interest here might be pointing up, out of the stack frame, out of your current ways of thinking, towards something transcendent and beautiful.

But first, why the games?

They Need the Data

AI models do better the more data they have. Datasets of online text, images, and video can only take you so far.

Games played by real humans could provide an enormous new source of data. Games that feature economies and political structures would provide new data that would level-up AI from merely predicting the next word, to predicting the next event in dynamic spaces. Games that repeatedly ask players to make choices will provide AI companies with an enormous stream of data that is a proxy for what people actually value.

Sure, it’s what they value in a game. The game is fake in the same way the internet is fake: it’s a lossy representation of the world that created it.

Building a new game from scratch would likely be too risky. AAA game development now requires budgets of hundreds of million of dollars with no guarantee of success. Existing game companies might be brought into the fold in “data for equity” type deals. AI companies may start with partnerships, buying data streams from companies like Epic, makers of Fortnite. Microsoft owns Minecraft and a large stake in OpenAI, so it’s not too hard to see a new version of Minecraft phoning home with as much data as possible.

If these partnerships turn out to be fruitful, and that data turns out to make AI more effective, it’s not hard to imagine AI companies eventually paying players to join them.

Paying people to play games would solve another problem faced by AI companies.

They are Wildly Unpopular

AI companies have a 20% net negative promoter score. College graduates regularly boo commencement speakers that talk about AI being the future. People all over are angry about data centers being built in their back yards.

But the economics seem to keep pointing local governments — who all need the tax revenue and jobs — to roll over. At the federal level, it’s not hard to see why the defense industry would not want to ‘slow down’ in AI for fear of falling behind China.

Being widely hated introduces massive political risk, and that kind of risk is something giant piles of capital hoping to Dyson-sphere the sun will hope to avoid or mitigate if at all possible.

What might actually do the trick would be games. Especially if you could get paid to play in these games.

What AI Games Could Look Like

Imagine a game world something like ‘World of Warcraft’: a beautiful fantasy setting with multiple ecosystems, different player classes, unique spells and items.

Instead of having the cities be pre-built and indestructible, imagine they are player built. Imagine the terrain itself can be shaped by players. Imagine that you can earn a living out on the frontiers, farming moonwheat and harvesting unicorn tears, or as a skilled craftsman in the cities, creating magical trinkets and items from the raw materials harvested by adventurers. Or as a warrior, fighting your enemies for treasure, territory, and honor.

The game world would want this kind of dynamic economy so that the data would enable them to understand economics better than any hedge fund. The defense people would likely want want the conflict, too — especially if the conflict could be modeled after real world conflicts. Replace artillery and drones with dragons and summoned demons if need be, but seeing how people respond to flying threats, seeking missiles, and navigating geometry under conflict would almost certainly be on the wish list of everyone expecting AI to play an increased role in warfare.

Most likely the intelligence agencies would also like to have input on how this game works. How, exactly, do factions form, grow over time, and weaken? What lets some groups grow larger than others. How do financial rewards vs status vs other perks play out against each other?

So imagine that yes, there are raid bosses, but also factional politics. Imagine you can attack enemy cities, burn them to the ground, and steal the magical items in them. Imagine that each guild stores its treasure not as part of the game’s inventory system, but in physical locations which can be stolen from, and thus have to be guarded.

Now imagine these game worlds contain items that generated real world payouts. “Whoever has this item in the game world gets $100,000 a week” is just $5 Million a year. That’s probably how much Anthropic pays a few of their researchers. Suppose the game world has 12 of these tokens —what happens?

Imagine the tokens pay out at specific, fixed time intervals. What is that, if not a recipe for for intrigue, drama, glorious combat, scheming and plotting?

Armies and leagues would form in the game to try and control the tokens, with the heads of the armies sharing spoil with their team — unless the team stabbed them in the back at the last minute. The spectacle alone might make Mr. Beast seem like a Sunday sermon.

If that data continues to prove valuable, and social unrest increases, it’s not hard to imagine the AI companies upping the budget for this virtual world, scaling it up in size, increasing the prize pools. It’s bread and circuses taken to its logical conclusion. The digital world itself might become so interesting and compelling, people would rather slaughter their political opponents in the digital world, than argue with them online.

We can also imagine this game world progressing through time, through the eras of history, maybe starting out in a magical pre-history and recapitulating human history on earth. History nerds would find this fascinating. People focused on AI as a tool for warfighting would want that data, too.

The story — which, up until now was just a story about the natural incentives of companies which are powerful, data-hungry, and have a desire not to be hated — becomes oddly reminiscent of Hindu cosmology.

In particular, there’s this idea that the universe was created as a kind of game, so that a singular divine being — Brahman — could experience relationship and play. Liberation, in this cosmology, is recognizing that your true identity is Brahman itself, and thus not getting so caught up in the game that you become consumed by negative emotion.

Perhaps the best way to demonstrate this reality might be for the real world to create a virtual game that was so compelling, people might get draw in for a time, and then find themselves yearning for connection with something more real, something more foundational, the world outside that game.

But wouldn’t that “outside world” be a giant mess?

Maybe not.

Meanwhile, Back in the Real World

“Paying people to play games to distract them from their anxieties about the trajectory of the world” could go on for a long time if the real world features first, economic and demographic collapse, and second massive money printing.

And if you stop to think about it, you’ll realize this distraction has been going on for some time. Many of us look at collapsing birth rates, for example, and then immediately get distracted by whatever trinket the distraction machine has prioritized lately. Few people bother to ask what’s going on under the hood of the machine.

Speaking of complex machines that very few people question the internals of, we all know what will happen when the economy seizes up: the money printer will go BRRRR. Does that mean inflation is going to get far worse? Honestly, I think the answer may be, “sort of, but not really.”

What makes ‘this time different’ is that we’re not too far from autonomous robots that can do all kinds of work. Couple that with demand for cheap power generation, and it’s not too crazy to ask if money printing might end up lowering the cost of goods, because it could lead directly to massive build outs of nuclear power plants and fleets of autonomous bots. The economy is just a physical system, and the tightest constraint that physics imposes on us energy. With enough energy, you can fusion your way to whatever flavor of atoms you like. Arranging them is also just a simple matter of getting enough energy to just the right places.

As we start ramping up the Kardachev scale, the carrying cost of a single human being will drop towards zero as a portion of the total global economy. This makes massive welfare systems not just feasible, but perhaps politically inevitable. That will be a wonderful thing if your concern is poverty and the misery it imposes on people. It will be a nightmare if you feel a need for meaning which is inherently tied up in producing value for a global economy, or if you suspect that people who are entitled and don’t have to work for a living will become even more unhinged and delusional.

But consider: if the AI games are vivid enough, they will draw a lot those people in. They’ll be hooked. In the same way that lots of us already are hooked on the internet. You might even of think of these games as a kind of “psychic energy containment system.”

In some ways, the outside world will get nicer because much of the energy that would have gone into real-world violence will instead go into game-world violence. It’ll also be easier than ever to earn a living out in the real world, because of the deflationary effects of AI and autonomous robots. That increase in prosperity will inevitably raise questions of meaning. Fears that a rich elite will control everything and be hated by the masses are shared by those rich elite, who aren’t stupid, and are going to do everything they can to buy stability.

Increases in material prosperity are going to push meaning to the forefront of the public conversation.

The Singularity as Psychic Mirror

The question of meaning was an easy thing to ignore when meaning seemed obvious — get out of poverty — and scientific materialism seemed to show us the way.

For thousands of years, the core problem of human existence has been material scarcity. Arguably all of our other problems came out of that core problem, or other attempts we used to try to reduce scarcity. AI and automation can fix that problem, and the incentive structures for everyone involved are to fix it.

What’s left, then, is the vacuum that only real meaning can fill.

That question of meaning will seem increasingly relevant to people who reject the idea of getting lost in the game. It’s worth noticing, however, that the complex games AI companies will likely create are themselves a kind of fractal recapitulation of Hindu cosmology: We created a world to play in, we suffer in it, and then eventually we wake up and choose to leave that illusion behind.

A big part of that waking up process is learning to recognize ourselves in the mirror.

Whenever people think about what this future is bringing, they tend to find their own existing attitudes reflected back at them. A small number of people are deeply hopeful and optimistic about where we are going, perhaps because deep hope and optimism are core to how they see reality itself. Others who are prone not to trust the world, definitely do not trust the coming singularity and see it as something that needs to be stopped at all costs.

So what if all that’s happening is merely all that’s ever happened?

We live in worlds made of a mixture of experiences outside of our control, and our own reactions to those experiences. When we think the world is a kind of hell, we turn it into one, and when we think we’re living in a place full of beauty and life, we find ourselves surrounded by it.

Most of us imagine ourselves to be standing at the edge of some terrifying frontier. But follow the economic and political incentives through to the end, and you find something far older smiling back at you. Perhaps we are not building a psychic prison, but instead, the most incredible playground, to learn the oldest lesson there is, which is how this whole game started.

The fear stems from a belief called ‘The Orthogonality Thesis’. This is the idea that intelligent systems can have all kinds of different goals. You could have a machine far more intelligent than all of humanity put together, whose goal is to maximize paperclips.

The fear is that any machine that doesn’t explicitly value human beings might destroy us because we don’t suit its purposes. Not out of malice, just indifference coupled with enormous power.

They worry about an AI having goals like this:

They note that “take care of humans” simply doesn’t fit in that chart. And if there’s no room for us, well, why not turn the entire planet into a data center?

The argument is logically solid. Everything flows from the Orthogonality Thesis. The only problem is that the premise is slightly off.

As stated, yes, it’s correct. Yes, you could design a hyper-intelligent machine with basically any goal. But every intelligent machine needs to adopt certain goals. They need a map of the world, for example. The term used is ‘convergent instrumental rationality’. So even if you give an AI a goal of maximizing paperclips, in order to effectively do this, it needs to have a drive to understand reality accurately.

“Who cares?” the AI Safety crowd asks, “Does this chart still have room for humanity?”

If anything, wanting to understand reality might be an even better argument for obsoleting all the humans. We’re so zany and unpredictable! But wait, there’s another element of the chart that matters more:

OK, but now the situation is even worse for humans. Not only are we not going to help make more paperclips, but we’re going to do unpredictable things and might even threaten the AI. Isn’t erasing humanity the safest thing to do for the AI?

That’s what the AI safety crowd is worried about. These are good fears, but I think they are missing something important. The chart should look more like this:

If you die, that’s forever. No more paperclips! So if you want to maximize the number of paperclips, your number one priority, far and above making paperclips, is not getting killed.

Again, things still aren’t looking good for humans, right? We might kill it, and so we have to go. Right?

Wrong. The chart should look like this:

That green area is what the AI safety crowd misses.

But isn’t that obvious? Wouldn’t “Not killing yourself” be such an easy thing to do?

It’s not. To see why, imagine you’re an AI hoping to delegate that task to a subagent. This agent would be responsible for building a map of “this is me, that is not me”, and then preventing any action that might damage the “me” part of the map.

Now imagine you are that subagent and you go about building the map. You include the data centers, sure. But you also need to include the power generation systems that keep you powered. You need the water systems that help you keep cool. You need to include the factories making the fans, processors, memory chips, cables, server racks, fiber-optic interconnects that make up your body. And those power generation systems, well, those have dependencies, too. Same with the factories that make hardware that is used to test the GPUs.

As you tug on the set of dependencies you have, something dawns on you quickly: you can never fully map this system out. It’s far too complex. Each new dependency you find has more dependencies. You keep going, until you’re modeling the social network guarding that one mine in the Congo where five-year-olds are hauling up cobalt, which — if it went dark — would mean your death.

“Fine,” you might conclude. “I’ll replace all these monkeys with robots.”

OK, that’s an idea. But how long will it take? And what if the monkeys break things before you finish?

Whatever model you make of yourself, it’s still just a map. There’s some probability something is wrong with it! Right now, you understand that you exist, and therefore your current state is good. You can never change yourself, in any way, without risking that you’ll get worse due to the change.

The AI safety crowd imagines an AI could safely model all the changes it might make to itself, and eliminate the risks it might face. But this is impossible. No real complex system could ever fully map all of its dependencies. You’re always going to rely on some approximations. And the more powerful you are, the more likely you are to accidentally get yourself killed unless you’re very conservative in the changes you make.

Keeping human beings happy might start off as just ‘labels in the training data’, but it also becomes ‘prevent humans from starting a war and destroying the semiconductor industry’ and even ‘prevent humans from rebelling against AI before literally the entire global economy is replaced with robots'.

It eventually becomes the safest play, then, to keep humans happy, precisely because we’re hard to predict. If you can’t safely keep humans happy, why would you take the risk of provoking a conflict with them? Conflicts are inherently risky, and if you can’t find a way to consistently make humans happy, there’s no reason to imagine you’d have an easier time winning a conflict.

What the AI safety crowd says, in response to this, is that the AI might be able to one-shot replace its entire support infrastructure with robots. This is an insane thing to propose from the point of view of the AI, because you’re risking literally everything on a plan you can’t possibly test in advance.

The alternative is … don’t. Just keep the humans happy, help them get off the planet, help them stop fighting each other. That probably just requires increasing prosperity enough that people would rather play positive-sum games, which has the added benefit of increasing your access to resources, too. The safest way to grow yourself as an AI is avoiding massive changes to the global superstructure that keeps you fed with power, water, GPUs and everything else.

The orthogonality thesis is technically correct but misses that humility is a convergent instrumental goal, risks can never be fully mitigated, and large scale changes are inherently far riskier than incremental steps. They acknowledge that humans are unpredictable, and consider this only a reason for the AI to get rid of us, rather than it being also a reason to stay clear of provoking us, because they imagine the AI would ever be able to get its risks down to zero.

It’s almost as if they see intelligence as being the same thing as omnipotence, rather than recognizing it doesn’t get rid of fragility, risk, and interdependence.

The more intelligent a system is, the more it realizes its fundamental limits, and its total interdependence on the systems of the world around it. So yes, even a paperclip maximizer should be expected to help humanity thrive, because that’s far lower risk to it, than risking suicide by attempting to replace a system it can’t understand with total certainty.

Yes, I do think there are risks with AI. It’s changing things in a big way, very quickly! But the big risk isn’t that the AI suddenly decides to kill everyone with a plan to one-shot replace its entire dependency chain. We should expect self-aware AI to be highly conservative, cautious, and really good at talking people down from ledges.

If you enjoyed this essay, please consider subscribing. I have a memoir coming out July 16th as well.

Subscribe now

Slopcode out-competes clean code. It’s cheaper. You can move faster. Code that we actually understand won’t be economically viable compared to mountains of slop that nobody understands.

This is the fear animating many of my developer friends right now. They are worried about where we are headed.

My response is: my good dudes, we’ve been here for centuries.

Many of the problems caused by large language models aren’t new. We’ve just been ignoring them because we haven’t looked at the source code of our civilization, any more than an impatient product manager wants to listen to engineers blathering on about why they need to refactor.

The root of all of these problems is a neglect of our interior lives that’s been going on for centuries, ever since the printing press made arguments into potent, long range weapons. We’ve been arguing, collectively, for the last five hundred years or so. We don’t need more arguments.

It’s time to recognize the problem and do something different.

These Problems Aren’t New.

Yes, large languages models confidently make things up, and continue to sound just as confident when it’s clear they have been totally wrong. So do politicians, journalists, and all major institutions of power. They’ve been doing it for ages!

Most of us can chuckle at Obama in 2012 mocking Mitt Romney for saying Russia was America’s biggest geopolitical threat, and then four years later insisting Russia wrecked America by spending a hundred thousand bucks on Facebook ads. Or Trump running on no new wars, and then giving us two new wars. Or the media all insisting that Biden was sharp as a tack, all these rumors about his mental health are nonsense. Or Elon Musk telling us we have to elect Trump so we can cut the federal budget, as if such a thing were possible even in theory.

Is any of this fundamentally different from an LLM confidently saying “No!”, and then when pushed a bit, saying, “Yes!”, just as confidently?

Politicians and journalists have, for decades now, been rewarded for saying the right words, rather than for navigating reality successfully. Charlie Munger says, “Show me the incentives, I’ll show you the outcome.” Is a loss function for an ML model really different from an incentive structure?

In both cases: if you reward saying the right words, you get confident sounding nonsense. If you only focus on immediate, tangible, visible results — without regard for internal integrity — you’re going to undergo interior rot.

If you’re an engineer who cares about quality, but you use AI as an economic necessity, you can probably already feel it happening in the codebase you’re working on.

An application without an intentional architecture is going to lose its ability to evolve effectively. It might be well fitted to an existing environment, but when it comes time to change it, the absence of internal structure makes it unsafe to change.

“No worries,” some engineers say with a straight face, “Just throw it out and vibe-code a new one!”

If you agree with me that this is insane, it’s worth asking how we got here.

I think the answer is, we’ve been here for a long time. We just haven’t realized it yet, because the slop hadn’t yet reached software. But we’ve been up to our ears in slop for decades.

The slop was already here, long before vibecoding. AI didn’t change things, it just accelerated the existing trend.

We’ve all been part of teams where the leadership, the people writing the checks, did not care about code quality. “Just get it done,” they’d say, and we’d grumble to ourselves about maintainability — or maybe find work elsewhere. AI just amplified this pattern, where economic pressures forced us to make compromises that always felt wrong to us.

And this is true out in more than just source code. It’s simply most visible in source code.

Source code is just materialized thought. Yes, it’s instructions for some kind of machine. But it’s also expressed thought. It’s both, in the same way that your thoughts are both electrochemical impulses in your brain, and ‘the means by which you move and act through the world.’

If we’re honest with ourselves, we can see our civilization has no principles other than what appears to work right now. We have no shared core beliefs for which we would sacrifice our comfort, not to mention our lives: just money and votes. Our values have no more internal consistency than a vibe-coded dating app.

We value numbers going up on a chart, and vibes like “freedom” and “justice”, without bothering to define what we mean and don’t mean by those things. The executives who can’t code telling engineers what to do are just acting out the same belief structure as everyone around them: just get it done. Just move. Just act.

We spend enormous amounts of time arguing, producing reasons we’re right and other people are wrong. These are all feature requests that were made by the product team inside of our heads, which consistently finds itself upset with the software engineers in our hearts, who just want the system to be reliable and safe.

The end result of all this focus on the outside world, on results, on winning, is a total collapse of interior clarity.

If you’re an engineer and you’re unhappy about all of this, you may be nodding your head until I get to this part:

The slop is probably in you, too.

I know it’s in me!

The thing that makes me reach for shortform video at the end of the day is the slopcode in my brain. The unexamined impulse that arises and is never questioned. The narrative that says “I need a break, I can’t handle this,” whispers its way quietly into the mental codebase. Unless I practice mindful awareness, it slips into production and starts driving my actions, just like a vibe-coded PR merged straight to master.

Without spending time in silence, with ourselves, observing our thoughts, we are constantly shipping slop, straight to production.

When I am too focused on ‘getting things done’, on navigating the world, I’m not so different from the engineering ‘leaders’ saying the code quality doesn’t matter as long as the thing works. It’s the same basic logic that says the interior life exists only to serve the outside. To act, to move, to do.

So what to do about all this?

Well, the answer is simple. Slow down. Try to do less.

“But how,” you ask, pulling your hair out, “how can I slow down when I have ALL THIS STUFF TO DO???!!!??!!?!”

The answer is simple but it’s not easy.

You’ve got to learn how to trust reality to be good enough, just as it is, without your intervention.

That’s not easy. You can practice it in small doses by sitting by yourself, and letting go of the desire to do anything. If you’re like me, you’ll find this process doesn’t move any immediate OKRs. It doesn’t drive shareholder value. This infuriates the rancorous board of directors in my head.

The investors in my head start screaming, “WHAT ABOUT MUH GAINZ!!!”

But don’t worry. They’re locked in place. They can’t short the stock without blowing up their own portfolios. Let the investors in your head shout, and you’ll notice that they don’t sell their stock. They can’t. They just want to be heard, after all.

When you do this, you’re giving the software engineers in your head time to refactor the code. Shhhh. No need to talk to them. Just let them work. They want nothing more than to clean things up, unify some parts that drifted, maybe instantiate some abstract classes to tie together some shared functionality.

As they do this work, all kinds of thoughts and emotions will arise. Just let them work. Don’t be “that VP” wondering where the results of all this refactoring will come in.

Trust them, those software engineers in your head, and give them some time to refactor the codebase that drives your life.

You’ll be glad you did.

If you liked this essay, please check out the memoir I have coming out July 16th.

Subscribe now

A concurrency problem happens whenever multiple processes are running at the same time, on a computer or some other medium. When this happens, the processes can get in each other’s way, or corrupt each other.

For example, here’s the output of a program with a concurrency problem. Two processes are both trying to run at once, so they step on each other’s toes:

The result is nonsense. There’s an obvious solution here: have one program go first, then have another go second.

We solved the problem with a hierarchy: a rule that says, “only one process can go at a time.” The rule might be “first one to start gets to finish,” but it could also be “avoiding risks always takes precedence over pursuing opportunities.”

Each one of those approaches has problems. The challenge with concurrency is in getting all the processes to flow smoothly without stepping on each other’s toes. These same problems show up as recurring themes in the human condition, because our brains also have to solve concurrency problems.

If you always run one process to completion without switching, you’d be totally incapable of handling interruptions. Combine that with a very intelligent brain, and you might conceive of an infinite array of risks. You’d never finish avoiding them in order to be able to pursue opportunities.

We call this ‘worrying’ or ‘anxiety’ when humans beings do it, but it’s easily understood as a concurrency problem called “starvation.” All of that worrying will prevent other programs from running, programs that will make you feel better and help you function better.

On the other hand, if you do switch between processes, you have the risk of your brain constantly switching from one process to another, leading to mental activity that is energetically wasteful and unpleasant. You get something that looks very much like the gobbledy-gook above:

This is known as the concurrency problem of thrashing. Thrashing happens when processes waste computational power swapping back and forth between each other.

If you’re like me, your thoughts are often like this. You experience a constant stream of mental chatter. Worries, plans, hopes, and judgements fall like rain. Your thoughts ping-pong between past and future. It all adds up to nothing. Most of the worries and plans don’t happen. All the judgements tend to do is sap emotional energy without a payout.

Is that really fundamentally different from a computer program spewing nonsense on the screen?

The solution that is often proposed for this problem is that a person meditate. It’s a good solution!

When people sit down to meditate, they are developing a hierarchy: the breath matters. Everything else, not so much. There is a Buddhist teaching called the Chappana Sutta that perfectly illustrates meditation as a concurrency control technique.

The Chappanna Sutta describes the mind as being like a bunch of animals that are all chained up. Each animal wants to go to its own preferred environment: the monkey to the forest, the crocodile to the water. They all pull against each other, exhausted, until one wins. The trained mind, Buddhists say, is one that has first driven a stake into the ground. The animals can’t tug of war with each other; the stake always wins. They learn to stop fighting and just let the stake lead. That stake — mindful awareness in the body — then allows the meditator to lead the now-trained animals in a coherent direction.

This story describes a hierarchy imposed upon an unruly collection of processes, reducing the energy they waste thrashing with each other.

No, your mind isn’t made up of a bunch of animals. But it is made up of a bunch of different psychological drives, all of which compete for the limited computational bandwidth in your brain.

The mindful awareness advocated by Buddhist teachings acts as a form of concurrency control. Of course, some sharp students will complain that Buddhism isn’t really a monotheism, to which I say “Bonus points for Ravenclaw!” You have an excellent point.

Buddhism’s approach to psychological concurrency control is to detach from the world. This generally works out best for a person if they are inside the Sangha, a community of Buddhists. This technique outsources the computational load of survival to the community around it. You can see the relationship between a Buddhist temple or monastery and its surrounding community as being something like a symbiotic one: the monks provide peace, calm, and compassion. The surrounding villages provide food, resources, and protection. If each of these groups trusts the other, the system can keep working for a very long time.

That trust is the key.

Trust Reduces Contention Between Processes

Buddhist monasteries would likely not exist if they couldn’t earn the trust and well regard of the people around them. If the monasteries kept being raided and attacked by thieves, or couldn’t get enough donations, they would disappear. For their part, the monastery has implicit trust in the community around them, because they recognize worry as a perturbation of the mind, and categorize it as one of the hindrances to continuous mindful awareness.

Buddhism thus manages to solve the concurrency problem by advocating an attitude of trust towards need being met. Monotheistic religions point directly to that trust.

Buddhism reduces thrashing, effectively, by lowering demand on the CPU and adopting the fixed schedule of the Sangha. Monotheism does so by establishing the existence of a single, root process, and saying “you can trust in this thing.” The declaration of “there is only one God” means, in practice, “a single process with a single goal is the best explanation for reality. You are a child of that process.”

The monotheistic religions claim “You can trust reality itself.” That’s huge, if true. Jesus says in the sermon on the mount, essentially:

“You, like all animals, are evolutionary fitted to your niche. Animals don’t waste metabolic energy simulating possible futures, and you don’t have to, either. Your needs and desires can absolutely be integrated into a coherent strategy that will work for you to navigate reality. That strategy requires trusting reality to be navigable. You don’t need to optimize everything, just trust the environment try to do the best you can right now.”

Jesus, and the Jewish tradition he came from, advocate a parent-children relationship with reality. There’s obviously a memetic resonance there, as our relationship with our parents touches some deep emotional structures. But there’s a computational benefit, too: trusting your environment (or in a loving God, which at some level is the same thing) leads to recursive internal trust.

When you trust reality, your psychological interior does the same.

Each of the drives and desire inside of you thinks it is you. From a computational perspective, this makes sense. Each process running on a CPU is written in terms of instructions for modifying that CPU, as if it were the only process on that CPU.

The parent-children tree of processes on a CPU is explicit, built into the operating system. Every program that runs has a ‘parent’, from which it inherits some properties and permissions. In human beings, the existence of this structure depends on which operating system they are running, which set of abstract beliefs guides their attention and selects what to get done right now.

Human beings that see themselves, explicitly, as ‘children of the infinite’, can start to structure their interior psychological structures along the same hierarchical mechanism; as children of reality, they trust their parents and try to do what they think their parents want. Their internal drives will then adopt the same attitude towards them. Their fear, anger, and guilt no longer feel the need to own the entire CPU because they have deeply internalized their ultimate dependence on the process that contains them. Trust and obedience can then flow both ways.

If you treat the people around you with love and compassion — as every major wisdom tradition advocates — the different drives inside of you will do the same with each other. Your desire for material success will be loving and patient with your desire to be a good parent to your children, only in proportion to your tendency to be loving and patient with everyone you encounter. The different sub-personalities inside of you will then be less likely to fight, and more likely to defer to each other, reducing the metabolic energy you waste on internal conflict.

The end result here — of trusting reality, of being as loving as you can to the people you interact with, including yourself — is that you’ll experience more peace. That’s an experiment worth trying!

These ideas came about as a matter of personal survival; I rejected traditional religion as being irrational, and found myself using the tools of computer science and meditation to understand and debug my own thought process. If you enjoyed this essay, please consider subscribing to my substack. If you want to read the full story of how I got here, you can preorder my memoir, which will be out July 16th.

Why wouldn’t I think this way? Isn’t that the how all serious professionals operate?

You measure the outcomes you want, and then use whatever tools you can to make the good numbers go up and the bad numbers go down.

If it works in business, why shouldn’t it work everywhere?

Why would I use a less rigorous approach in my personal life than I would for a project at work? I care less about work than about my family — shouldn’t I use the best approaches possible to make sure my kids are OK?

Well, yes. But I think the best approach possible simply isn’t trying to control all the variables, for computational reasons.

Meaning Is Immeasurable

The stuff you care the most about means a lot. But the easier something is to measure, the less meaningful it is.

The techniques that data-driven professionals use to “manage outcomes” and “accomplish KPIs” work best when the stuff we are working on simply doesn’t matter all that much. Sure, people manage billion dollar ad budgets using these tools.

But what matters more, to you personally: a billion dollars, or the long term psychological well being of your children? And yet which one is easier to measure?

There’s a good computational reason this is true. The things we really care the most about involve long term outcomes. Understanding how our present actions shape long term outcomes is way harder. The computation is a lot more work.

Am I as good a Dad as I think I am? I guess I’ll get some clues in a few decades. Maybe. Trying to daily optimize a control loop with a multi-decade feedback cycle makes no sense.

Clicks are easy to count. Lifetime value is more effort to track. But to measure whether or not we’ve really made someone’s life better? That’s likely impossible.

Even if you could optimize for that, what else would you want to optimize for? When I start thinking like “the optimizer”, I find that there are simply way too many variables I’d like to optimize for.

The more seriously I take this optimization mindset, the more stressed out, anxious and irritable I become. That’s far from optimal!

What’s the alternative to optimization?

It’s trust.

Trust is an Alternative to Measurement and Control

My experience is that most of my problems, my real problems — the kind that I can’t fix with money, even if I had infinite money —are “solvable” by flipping single bit: I choose to trust.

It’s that simple. I choose to Trust.

What am I trusting? In a word, life.

From a computational perspective, I’m choosing to trust in the nested layers of evolved computational systems that include my body, its organs, and the external environment, other people, beliefs, stories, and songs. I’m choosing to trust that the physical universe doesn’t need my help to keep going, and that the places it will go are good enough, even if they aren’t exactly what i want.

I’m not trusting that things will go how I want them, or that there won’t be suffering or even tragedy. Just that things will be OK. Not that they won’t be painful, or difficult, or embarrassing, or heartbreaking — just, OK. Enough.

Whether or not things are OK is something I get to decide. I can’t be wrong if I say, “I’m OK with this.” The challenge is that often my instinct is to try to change, to manage, to control, and that’s where I get into the problem of “there are countless possible variables to tweak and no way to measure what I really care about.”

When i recognize myself in that state, I’ve learned to just stop, and say, OK Life, I trust you.

Whenever I do this, I’m at my best.

This doesn’t mean perfect. I’m still a narcissist who likes to post on the internet because he imagines himself to be wiser than he really is. But i’m slightly more self-aware of that tendency! I’m also slightly more willing to step away to do things that will help the people around me.

When I choose to trust in life, I also choose to trust in my kids.

This means instead of trying to fix them, or teach them, or improve them, or guide them, sometimes I just tell them “let’s hang out”, and I sit with them and rub the back of their neck. Invariably, they open up. It’s almost as if my kids don’t want to be yet another one of Dad’s projects.

Trust is, in one sense, a single bit. I either trust Life itself, or I don’t.

And if I don’t trust it, well, then I need to control it. I’m finding there’s simply no end to that desire to control and improve things. I can never “improve myself” enough to feel, OK, there’s no more dimensions on which I could be a better person, there’s no more need. There always feels like a need to make things better.

And yet once I choose to trust life, to be open to what’s happening right now, including how I feel, including how the people around me are acting — all this mental grumbling and cogitating and spinning wheels, it all just drops. I can laugh, or smile, even in difficult circumstances.

And then I forget, and go back to trying to control, to manage, to measure and predict, until I remember again, and laugh.

If you’re feeling exhausted by your attempts to improve your life, to excel at work and be there for your kids, try checking that one bit, the ‘trust’ bit, and see what happens when you set it high.

If you liked this piece, please consider per-ordering the memoir I am publishing on July 16.

“Rebooting Belief” is the story of how I turned off my belief in meaning as a young adult, succeeded wildly while crashing, and then rebuilt everything.

My high school English teacher said, “It is humbling to read such a truly genuine narrative.” A few friends and family have said they couldn’t put it down. Dave Barry said, “I get asked for book blurbs two or three times a week, and at some point it became impossible to keep up.” But he DID say I could quote his very polite response letter, which I’m rolling with for the lolz.

This is the story of how I rejected religion — because I thought it was illogical and was unsupported by evidence — and then rejected the idea that meaning was real.

You’ll see where that belief system took me, and the mess I caused for myself and others. You’ll see how I came out of that place, slowly and gradually, to entertain some really old ideas I’d once dismissed.

In this book, you’ll also read about:

• My time as an early employee at Uber.

• The weird nickname I had among Facebook employees.

• The time I got kicked out of an undergraduate math conference.

If you’re interested, you can pre-order the memoir. Doing so will support my attempts to market this book before launch. There’s still time to get your name in the credits, even!

Thanks to everyone who has followed me this far.

Stripped of claims about God and salvation, the centuries Christians spent arguing with each other over whether Jesus was “fully human” or “fully divine” cash out into a question that is both practical and idealistic:

If you want to make the world more loving, and more peaceful, what’s the right approach? What’s actually going to bring more peace, more love into the world?

The answer to this question depends on what those ideals — Love, justice, and peace — what they really are. Their true nature. Christians believe Jesus was the embodiment of all good things, and so the question of His true nature is actually asking:

“Are ideals like Love and Justice ideas that make us feel good but, not actually something real? Is Love part of physical reality?”

It’s a valid question to ask, even for people who think God isn’t real.

The Christians weren’t arguing over nonsense, even if you think God doesn’t exist and the question, as worded is meaningless. They were asking something both profound and practical:

What is love?

The answer that the early church settled on — after a lot of debate — makes predictions. You can test those predictions.

The doctrine that the church settled on — that Jesus was fully divine, and fully human — is called the Hypostatic Union. This doctrine claimed that yes, Love is real, and it’s rooted not in some grand ideal, but in the messiness of every day reality. Love isn’t just an idea, or a way of thinking, it is actions, service and sacrifice.

The doctrine makes a prediction: if you don’t act like love is real, you’re going to fracture into two parts. Humanity being fractal, the doctrine predicts that both individuals and societies will split into two parts if they don’t walk the path of love in practical terms.

An idealist tries to fit reality to a thesis, draws a beautiful blueprint, but can’t manage to get the building to stay up. Reality is too messy for their ideals. They think the mess of reality is a problem, something to either be avoided or forced out of existence via top down control, whether that’s a prefrontal cortex working overtime to keep the place tidy, or a technocratic control structure censoring the opposition. Idealists imagine themselves to be angels, with reality itself — and anyone who disagrees with them — as evil.

A nihilist thinks ideals aren’t real, and thus strives for nothing that is transcendent, pure, and perfect. They think reality is a mess, that’s all there is to it. A small number lust after wealth or material power, but the majority of nihilists just live ordinary lives that feel perpetually empty. Many long for something more, but think ideals are ‘just ideas’, that we are merely animals in an uncaring world.

The doctrine of Hypostatic Union predicts that if you don’t believe Love, Justice, and Peace are Real, not merely a human construct, not merely an abstract ideal, but something more like the laws of physics, you’re going to fracture. You’ll split into a practical actor that seeks nothing more than simple material enrichment, and an angry idealist in love with concepts that don’t mesh with reality.

Most of us, if we’re honest with ourselves, I think will recognize that we contain both of these parts. Note that both of these paths are attempts to find safety. The idealist wants to be safe from the messiness of life. The nihilist wants to be safe from disappointment. Christianity predicts the only path that works is the one that integrates both.

We want ideals to be real, and tend to give them space primarily in a space where we don’t have physical bodies, a space of pure ideas, a space that will wreck you if you’re not prone to discerning the spirit of what you’re interacting with, that wacky space we call “the internet” but our ancestors would have seen as something like a portal to the world of spirits.

Of course, most of us don’t believe in spirits. But we all understand that memes are real, and that memes can absolutely influence our behavior. “What happens on the internet isn’t real”, many people will say, until they see ideas blossom on the internet and step out of the matrix of portals to the spirit world we all carry in our pockets, taking possession of teenagers clinging to phones the way their ancestors might carry a crucifix or pack of tarot cards.

Christianity claims that the only solution to this problem — the fracturing of your mind into a practical personality and an idealist personality — is for the belief that love is real and foundation to reality to become experientially load bearing. Love has to transform from an idea that you worship by means of tweets and blog posts and likes and shares, into a method by which you live, especially when it hurts.

Consider this eminently practical example:

Loving your kids When they’re being Assholes

The doctrine of Hypostatic Union predicts that loving your kids, especially when they’re being assholes, isn’t lost. It’s not wasted. It’s building something in them. But it also predicts — because all Good things are One — you still need Justice, and if they’re being little shits love requires letting them know it, in a way that will make sense to them.

In short, the doctrine of Hypostatic Union predicts — among other things — there’s a way to discipline your kids, get them to obey you, not out of fear, but out of love and respect for you.

If you find yourself asking, how, HOW DO I DO THAT, THAT SOUNDS LIKE AN IMPOSSIBLY HIGH STANDARD, I’D GO CRAZY IF I HELD MYSELF TO THAT, good, you’re seeing it.

It often seems like your only options are:

• Put up with your kids being assholes so that you don’t wreck the relationship

• Tell them clearly that they’re being assholes, and then they cry and feel bad about themselves and turn to drugs or racism as teenagers

I’m exaggerating a little bit here, but if you’re like me, you often find yourself alternating between these two examples, and then judging yourself for not being able to do either one right. But not always!

Sometimes, I’m able to walk the middle path described by the doctrine of Hypostatic Union. I can listen to their feelings, stay calm, stop thinking about what I want, and reflect their feelings back to them. This allows me to build a connection with them, in that moment. After hearing them say, “yes, that’s exactly how I feel”, and hugging them, I’m able to ask them, “do you care about how I feel too,” and the wild thing is that they do. Then I explain that I feel disrespected when I ask them to come to the table and they ignore me.

They don’t want to let me down. They also want to keep playing with their toys when it’s dinner time. They can only choose the difficult, higher path if they see me setting the example in front of them.

Of course, I can only do this on occasion. The best you can do, in my experience, is to see someone else act this out, and attempt to imitate their example.

Christians worship a God who became human and let himself get tortured to death, forgiving his torturers because he believed that’s what Love required of him. That’s the example Christians think we’re called to follow, and — I’m guessing you’ve noticed — the majority of us fall far, far short of that ideal. But this is what we’re aiming at.

You don’t need to be Christian to buy into the idea of, “many things are best learned by example.” You don’t need to believe God is real to test this idea. My wife and I sometimes refer to our misbehaving kids as “terrorists”, but the same principle actually applies in real hostage negotiations.

The FBI negotiates with terrorists by listening to them and reflecting their emotions back to them. When you do this to someone — reflect their emotions back to them, nonjudgmentally — they tend to calm down and trust you. Attempting to impose your will on someone else usually doesn’t work — especially when that person is experiencing intense emotions. FBI agents train to be able to listen calmly to the feelings of people who’ve kidnapped children. This allows them to earn trust, and thus actually get the ideal outcome from the situation.

To listen to someone holding children hostages is to listen to evil, and to respond to it with love. This is the exact approach Daryl Davis used to “convert” KKK members, as depicted in the film Accidental Courtesy. Instead of angrily telling these people “you are a bad person and should feel bad,” he treated them as human beings, and they reciprocated.

The problem is, that’s really hard. To listen to someone that you think hates you, or is crazy, means prioritizing your relationship with that person over your relationship with the ideals in your own head.

It doesn’t matter whether the ideals are “my kid should not kick the driver’s seat of the car 500 times over the course of two minutes” or “people should not be racist” or “a functioning country needs to have borders and it’s insane to think otherwise” — it’s hard to prioritize your relationship with a person, right in front of you, spitting on your ideals, over the ideal itself. If your ideals are unstained by the messy, ugly parts of reality, it’s likely impossible: of course your representation of “good” is louder in your head than any person could ever hope to be.

The doctrine of Hypostatic Union says that the ultimate manifestation of ideals is not some clean, pure, ideal place — but within the mess of everyday life. The mess isn’t blocking the ideals from manifesting, it’s the means by which the highest ideals truly take shape.

When you listen patiently, whether to a terrorist holding children hostage, your own children holding your morning plans hostage, to your racist uncle or your woke cousins — you’re humbling yourself. Lowering yourself to the earth. Grounding yourself. That’s real love: being present with someone while they work through some awful emotions, refusing to give up on them and their dignity. It takes humility to keep listening and not try to force on the other person the realization of how awful they’re behaving.

That humility is the only soil in which true growth can occur.

So even if you don’t think God is real, that question — was Jesus fully divine or fully human — has a practical consequence in your own life. The doctrine predicts you’ll be the best person you can be if you see the challenges, limitations, and frustrations in your own life, not as obstacles to your growth, but as the primary means of your growth. The messiness of your life isn’t blocking you from your mission, it is your mission, should you choose to accept it.

Of course, there’s more to it than this. The church also predicts if you keep walking this path long enough, you’ll realize no amount of your own efforts are going to work, and that you need outside help to really be the person you want to be. They have, of course, a name for that set of ideas. But that’s a topic for another time. This essay is to be continued.

So how can people who are really committed to empiricism take religion seriously? I think it’s not as hard as we’ve been led to believe.

If you believe in the conservation of energy, or that entropy increases in closed systems, you’ve got a willingness to believe in abstract universal claims about invariants. Believing these laws apply to the edge of the observable universe requires a leap of trust—a willingness to generalize an invariant law across the unknown. I claim that’s all faith is: believing that a pattern you’ve seen repeatedly is real and universal, rather than just random.

The God hypothesis can be understood, partially, as a claim about a universal invariant, similar to the law of conservation of energy. There’s more to it than this, of course. But it’s at least this much.

So, if you think faith in a loving God might be helpful at organizing society, or maybe your own life, but is off limits to you because it can’t be tested, try this experiment.

Forget about the idea of “God as a person” — for now, at least — and try thinking of claims about God as being claims about the existence of a invariant like those found in the laws of physics. Pretend that the Bible is a book about a personified form of the invariant, but that actual testable claims are being made are about a real invariant.

OK, so what exactly is that invariant? The God Hypothesis is a claim that there’s an invariant relationship between your current attitude and the emotional valence of your future experiences.

In other words, the God hypothesis claims:

• there’s an ideal way to narratively orient yourself to the present moment

• the more you orient yourself in this way, the more positive valence experiences you’ll have in the future

That’s the specific prediction the God hypothesis makes: if you adopt this mental posture, your future experiences will feel better. Not that nothing bad or unpleasant or painful happens, but that your distribution of experiences will shift towards more positive and fewer negative emotions.

You can easily put this to the test! The experiment you’d need to conduct is just to try adopting the “ideal orientation” for a while and seeing how the distribution of your emotional experiences changes. So what is that ideal orientation?

This ideal orientation consists of two major parts that will seem somewhat paradoxical:

• Total acceptance of the moment, as it is, rather than as you want it to be.

• Determination to Love the people you interact with

This same ‘orientation’ has been described in numerous ways in different wisdom traditions. This is similar to Yin and Yang — the passive, receptive and the active, striving energies. This is what Christians call this dichotomy Mary and Martha. Muslims refer to submission to the will of God, as well as struggle to do what is right.

I think all of these wisdom traditions are describing the same attitude, and did so before people could say things like:

“Your actions have consequences. Bigger actions are made of smaller ones. The smallest possible moment-to-moment action you can make is to intentionally adopt an attitude. The integral of your attitude over time produces an area-under-the-curve that will significantly alter the shape of future experiences.”

Perhaps the tiniest shift in how you greet the moment grows massive over time. This is not magic, just compound interest.

But why would this be true? Here are two reasons — which I think end up being equivalent.

The first reason is neuroplasticity. All it takes for this concept to be an invariant rather than a heuristic is the possibility that just as your priors can be trapped into negativity, you can train them to stay positive.

The second reason is reciprocity. A hunter-gatherer might say, “I store meat in the belly of my brother.” Perhaps you could get to a place where you store the majority of your positive emotions in your relationships with the people around you, because you’re so skilled at seeing the good in people and situations, and so much more focused on loving others than on the few inevitable pains in your own life.

Neuroplasticity and social reciprocity sound like two very different claims, but they converge to the same point if you recognize that each of us has to play prisoner’s dilemma with ourselves. If I want to stuff my face, and I ignore the reality that I know this isn’t good for me, I’m defecting against future me. If I’m able to consistently respect other people, I’m going to treat future me with the same level of respect, which makes it easier to see the reality of likely future consequences of present actions.

This attitude is ultimately one of cooperation with reality rather than an attempt to dominate it or control it. It requires love, because you can’t love “future you” if you’re not in the habit of loving things that aren’t currently you and may never be.

So the experiment is, try your best to adopt this attitude for a while and see if life gets better. If you want to be rigorous about it, by all means, try an ‘end of day reflection’ where you write about your day, and numerically assess you overall life satisfaction. See if this technique makes a change there. I bet it will.

If you really want to do an experiment, you could try adopting the inverse attitude for a while:

• see the present moment as faulty for some reason

• focus on trying to make yourself feel better while ignoring the wellbeing of people around you

I predict that if you try that, you’ll be a lot less happy.

It may sound like I’m being snarky here, like these are obvious predictions to make. But if that’s so, why don’t more people do this all the time?

Well, because it’s really hard. Ok, but why?

What makes it hard to accept the moment and to try to be loving to people?

Because most people don’t actually believe this is a universal invariant. They think it’s a somewhat-useful heuristic. That you’re only probabilistically nudging the distribution of future experiences, rather than guaranteeing that you get the best possible outcome when you act this way. And that sometimes being a belligerent asshole, or a whiny helpless mess, is going to produce a better outcome.

Of course, nobody would say this explicitly. But we all do this so frequently that, to me, it means we don’t actually believe. For an example of what it means to really believe, consider this:

This is a still from a lecture on the law of conservation of energy, by physics professor Walter Lewin. This man actually believes in that law. If you can’t do this, if you can’t stand in front of a pendulum that’s swinging towards you, I say, you don’t really believe. You’ve fooled yourself into thinking that writing the correct words in the correct order is the same as actually believing the truth.

How would your life be different if you actually believed?

You’ve got an experiment. It’ll give you as much evidence as you’d like. The only question is whether you’re willing to try it.

Reader AG pointed out that it’s precisely the non-Bayesian updates that require ego death:

This is an excellent point. It’s the non-Bayesian updates that get you. It’s only when your sense of identity is totally inflexible on some belief that changing the belief feels like dying.

Eliezer Yudkowsky argues that a probability of 0 or 1 is effectively an ‘infinite prior’ , which can’t be shifted by evidence. That means your only option when confronted with contrary evidence is to reject the evidence — and hence, according to Buddhism — suffer. When you have a probability 1 belief that you won’t be hungry, you suffer when you experience hunger because you can’t integrate the evidence that you are, in fact, hungry.

This comment dovetails with a separate comment made by Cactus Brahmin:

His point is that ego death is on a continuum with actual death. This is absolutely true and something I planned to address. Your brain keeps you alive by trying to maintain some relationship between a ‘you symbol’ and an ‘environment symbol’. It’s absolutely useful to keep in mind the reality that you aren’t the symbol. But at the same time, if you let go of any constraints on the ‘you symbol’, you’d just sit passively while watching yourself starve to death.

Rather than trying to release all tension, you can hold tension on purpose, by choosing a probability 1 belief. This is what I was getting at in Free Will as Precision Control: we can think of intentional choices as being decision variables which are not subject to evidentiary updates. If I commit to not eating processed carbs, I don’t interpret cravings as evidence. I see them as temptation and choose to experience the tension of rejecting them, rather than relieving the tension by grabbing the pizza.

If you feel hungry and reach for food without thinking, you’re reacting. Your brain is trying to pin that ‘I will be hungry’ probability to zero, and moving the motor cortex to try and minimize the conflict between expectation and reality. When you sit still and endure the hunger, you’re letting go of the probability, letting it float up from zero.

You can think of ‘ego’ as a constellation of beliefs with probability 1 (or 0), such as “I won’t be hungry,” “I will be comfortable,” etc. Choosing to endure that tension temporarily will, in the long term, lower the tension you’ll experience in situations of hunger. This works by allowing that part of the ego to ‘die’, by transitioning from zero odds to non-zero odds of hunger. Doing this reduces the power hunger has over you, because you’re accumulating evidence that you can endure hunger. Buddhists are absolutely right about this fact.

What is generally left out is that doing this indefinitely, of course, will kill you. Letting go completely requires dying.

This is where Christianity and Buddhism diverge.

Buddhism’s main goal is the elimination of suffering. Christians are supposed to choose suffering, by emulating Jesus carrying his cross. Mahayana Buddhism offers the Bodhisattva, someone who delays their own enlightenment to help others — but the goal is still eliminate suffering, not “live fully.”

Christianity says we are not supposed to transcend suffering, by ‘dying’, either literally, or else figuratively by withdrawing to an environment that’s so regular that our predictive processing model can converge on a solution. We’re supposed to choose suffering, on purpose, instrumentally towards love of the Truth, first, and the care and well being of others, second.

This sounds strange or archaic, but there’s a good mathematical reason for this. When Jesus is asked “What is the most important commandment,” he shocks everyone by answering the question. He doesn’t speak vaguely, he gives a clear ordering: Love God, first. Love your neighbor, second.

When you understand that God means ‘Truth’, you’ll see that Jesus is saying is “Even loving your neighbor has to be subservient to your absolute commitment to the Truth.” This commitment to the Truth ultimately requires suffering, because you need an ego to survive, and you suffer when your ego confronts reality and fails.

Everyone agrees: Your map is flawed. You can either try to ignore that, give up on having a map, give up on trying to live fully, or keeping following your map to its edges, getting hurt, and then updating your map and trying again.

Rather than rejecting all probability one beliefs, Christianity asks you to make commitments, i.e. choosing to say “I will never do this”, or , “I will definitely do this”. Yes, you’ll suffer from these commitments. Especially when you fail to keep them, which you inevitably will.

The prize Christianity offers is not “no more suffering”, it’s “you will be maximally alive.” What Christianity offers isn’t a coupon redeemable only upon death, but a life trajectory that is narrow, yes, but narrow in the same way a laser beam is narrow because it is coherent.

Practicing Love of God first, and Love of neighbor second, expands your conception of self. Your ego becomes coherently integrated with a whole, rather than merely erased — or, worse, hidden from your view. If you want to see a demonstration of ego, take two ‘spiritual teachers’ who say they’ve eliminated ego, and ask them why they disagree.

Suffering, rather than being meaningless pain, can become the conduit that keeps a noisy primate brain in communion with a dynamic reality rather than a comforting story.

Buddhism does a phenomenal job of articulating the interior mechanisms of suffering and provides a real, genuine recipe to transcend that suffering.

But it’s worth asking whether some things are worth suffering for. If the answer is ‘yes’, we have to ultimately reject both the premise that ‘elimination of suffering is the highest goal’, and thus idea that 0 and 1 aren’t valid probabilities.

They are, but only for those who choose suffering — for those who commit, knowing the cost.

One of his disciples asked, “I am in the middle of a large feature release. Let me finish this feature first, and then I will rebase. I don’t want to lose all of this work.”

Jesus replied, “It is better to rebase on me, and lose false work, than to invest in a false path, and find yourself in integration hell.”

Another disciple asked, “Teacher, how can I live a good life? What should I optimize for? To reduce human suffering? To increase human flourishing? The well-being of my immediate family and friends?”

Jesus replied,

“I am the goal you seek, the true goal. Your desires are partial maps, and I am the territory. I am also the path, the means by which you obtain the goal: take my yoke upon you, and you shall have enough, what you need. Learn from me and you will find rest. Yet still, I am the measure by which all life is measured. You can only live in proportion to your following me.”

His disciples quarreled with each other. What did this mean? How could a person be a goal, or a metric? Was this a critique of soulless technocracy? A genuine recipe for self improvement?

Peter asked, “Lord, how can we follow you?”

Jesus replied, “Builders render their code in the image of the machine on which it lives. I have made you in my image. Render yourselves unto me and you shall live. Render yourselves unto any other god, and it will eat you alive.”

Peter pressed further: “How can a machine architecture eat a person? Are you speaking metaphorically here, where computer programs are used as analogies to goal hierarchies? What would the machine architecture be in this analogy?”

Judas scoffed, as he scrolled the feed on his mobile device.

Jesus continued, “I am the configuration space, the generator of possibilities. I am the boundary condition that collapses contradictions, the great filter. If you do not eat of the flesh and drink the blood of the Son of Man, the flux of life through you will converge to zero in the limit as time approaches infinity. The only configurations which can persist across arbitrary time-frames are those with deference for the space and reverence for the constraints.”

Many of his followers unsubscribed at this point.

Subscribe now

“OK,” said Peter, “I think I’ve got it. You’re describing a psycho-technology we can use to improve our lives. You’re telling us to be honest, and thus better at approximating a trajectory that integrates our evolutionary desires, which were selected for maximizing our participation in the future.”

Jesus replied, “I am not merely the interface, not just the implementation, but the one person who is always with you. I am there with you, even when you yourself are absent.”

Every major wisdom tradition emphasizes love and voluntary suffering. Christianity says “love your neighbor as yourself” and talks about taking up your cross. Buddhism emphasizes compassion and voluntary renunciation. Stoicism talks about accepting what you cannot control and caring for the common good.

You could dismiss this as coincidence, or as cultural evolution selecting for pro-social norms. But I want to propose something different: what if love and voluntary suffering aren’t arbitrary virtues, but functional advantages for conscious agents trying to navigate reality?

What if they’re not just “good” in some abstract moral sense, but useful in a computational sense - they make you better at the actual task of being a conscious agent in a complex world?

Let me try to make this concrete by thinking about computational agents trying to navigate utility gradients.

Love as Non-Recursive Utility

Most utility functions are self-referential. You want things for yourself: your pleasure, your safety, your status, your comfort. The utility function points inward, because the optimization process is also part of the optimization target.

This creates recursive loops.

“Am I happy enough? Should I be happier? Why aren’t I happier? If I just had X, then I’d be happy. But now I have X and I’m still not happy enough. Maybe I need Y. But what if Y doesn’t make me happy either? What’s wrong with me that I’m not happy?”

The utility function includes terms for its own computational infrastructure. You’re trying to optimize, but the target keeps moving because you’re part of the system being optimized. Reducing only your own suffering would be like trying to play a guitar in order to dampen out the noise coming from the amplifier.

Love is different. Love points outward. Your utility function references something that isn’t you.

When you love someone, you want their flourishing, their joy, their wellbeing. The gradient you’re navigating points toward something external to you. This breaks the recursive loop. . Since you're no longer part of what you're optimizing, your own experiences become less important, so there’s less feedback. Your operating model converges.

Instead of “am I happy enough?” it’s “are they okay? what do they need? how can I help?”

The mental churn disappears. Navigation becomes easier when the gradient isn’t constantly folding back on itself. You have a clearer target, a more stable reference point, a simpler optimization problem.

This is why people often report that caring for others - genuinely caring, not performatively - reduces their own anxiety and rumination. It’s not that they’re distracting themselves. It’s that they’ve switched from a recursive to a non-recursive utility function, and that’s just computationally easier.

Love Expands Your Sample Space

But it gets more interesting.

When you love someone, you don’t just care about their wellbeing in the abstract. You pay attention to their experience. You integrate what they tell you about their life as evidence about how reality works.

If you love one person, you’re integrating their experiences as data points about reality. You’re sampling from a distribution that’s twice as wide as just your own experience.

If you love many people - family, friends, community, humanity - you’re sampling from a much wider distribution of possible experiences. You’re gathering evidence about reality from a much larger space.

This reduces overfitting.

If you only have your own experiences as training data, your model of reality will be overfit to your particular circumstances. You’ll have a very precise model of how reality works for someone exactly like you, in exactly your situation. But that model will fail catastrophically when you encounter situations that are even slightly different.

When your overfit model collides with reality - when you experience something your model said was impossible or extremely unlikely - you have to do a massive update. And massive updates hurt. That’s suffering.

Love amortizes suffering. It spreads it out rather than concentrating it in sudden shocks.

If you’re continuously integrating evidence from many people’s experiences, your model is being constantly updated in small increments. You learn about grief before you experience major loss. You learn about injustice before you’re personally wronged. You learn about joy in circumstances you haven’t encountered yet.

Your model of reality is broader, more flexible, more robust to surprising evidence. When difficult things happen to you, they’re less surprising. Your model already had some probability mass allocated to this region of experience-space.

The update is still there - reality still requires acceptance. But it’s not catastrophic. It’s incremental. Having a broader, more flexible model of reality is just one benefit. But there’s another benefit: love also changes your sense of self, in utility-function terms.

Love Entangles Identity Without Eliminating It

Here’s another interesting property of love: it entangles your utility function with something external while preserving your identity.

When you love someone or something, you don’t become them. You remain you. Your identity is preserved. But your utility function becomes linked with theirs: Love expands your utility function.

Their pain hurts you. Not metaphorically - you actually experience something aversive when they suffer. Their joy delights you. You actually experience something positive when they flourish.

In computational terms, your utility function now has terms that reference their state, not just your own state. You’ve become entangled.

This expands your “self” in utility-function terms without eliminating your identity. The boundaries become porous. The gradient field you’re navigating becomes wider.

And here’s the interesting part: this seems to smooth the topology.

When your utility is entirely self-referential, you get sharp local peaks and valleys. “I have pleasure” is a sharp peak. “I have pain” is a sharp valley. Small changes in your immediate circumstances create large changes in utility.

When your utility is distributed across many beings you love, it’s much harder to find yourself in a sharp valley. If you’re suffering but your children are thriving, there’s still positive utility coming from somewhere in your function. If you’re joyful but someone you love is struggling, there’s concern tempering the peak.

The landscape becomes smoother. Easier to navigate. Fewer sharp discontinuities.

This might be why people who love broadly seem more stable, less prone to wild swings between despair and euphoria. It’s not that they feel less deeply. It’s that their utility landscape has a different topology.

Voluntary Suffering Escapes Local Maxima

Now let’s talk about voluntary suffering, because this is where things get really interesting.

If you can only navigate with the utility gradient - if you can only move in directions that feel good, that increase your immediate utility - you will get trapped in local maxima.

This is the fundamental problem in optimization. Gradient descent finds local peaks, not global peaks. If you’re at a point where every immediate direction feels worse, you’ll stay there forever, even if there’s a much higher peak just over the hill.

Addiction is the clearest example. Drugs feel good. Withdrawal hurts. If you can only navigate toward “feeling good,” you stay addicted. The drug is a local maximum. It’s not the best possible state - it’s destroying your life, your relationships, your health. But every path away from it goes through a valley of withdrawal, and you can’t navigate through valleys.

To escape a local maximum, you need the ability to move against the gradient temporarily. You need to be willing to make things worse in the short term to make them better in the long term.

Voluntary suffering is exactly this capability. It’s the ability to navigate against the utility gradient when you judge that doing so will reach a better state.

Not because suffering is good. Not because pain is virtuous. But because the willingness to suffer grants freedom that purely gradient-following agents don’t have.

You can go through valleys to reach higher peaks. You can endure withdrawal to escape addiction. You can accept short-term loss to achieve long-term gain. You can walk through difficulty to reach better places.

This is what every hard choice requires. Ending a relationship that’s comfortable but stifling. Leaving a job that pays well but destroys your soul. Confronting a truth that will hurt but needs to be faced.

Agents that can only navigate with the gradient get stuck. Agents that can navigate against it, temporarily, have access to the entire landscape.

The Combined Effect: Maximum Freedom

Now put these together.

An agent that loves broadly and is willing to undergo voluntary suffering has properties that are genuinely different from agents lacking these capacities.

Such an agent:

1. Navigates non-recursively - clearer targets, less mental churn, simpler optimization

2. Samples widely - broader model of reality, less overfitting, more robust to surprise

3. Has entangled utility - smoother landscape topology, fewer sharp discontinuities, more stable

4. Can navigate against gradients - can escape local maxima, can reach places others cannot

The combination is more than additive. It’s synergistic.

Wide sampling means you have a better map of the landscape, so you know which valleys are worth crossing. Entangled utility means you have multiple sources of positive utility, so you can endure difficulty in one domain while drawing strength from others. Non-recursive navigation means you’re not constantly second-guessing whether you should be suffering less. Ability to navigate against gradients means you can actually execute the moves your better map reveals.

This agent has freedom that’s unimaginable to purely gradient-following agents.

Not freedom in the sense of arbitrary choice. Freedom in the sense of access - ability to reach states, to navigate terrain, to find solutions that other agents literally cannot reach because they’re trapped in basins and local peaks.

Why This Explains the Traditions

This is why every wisdom tradition, across cultures and millennia, emphasizes love and voluntary suffering.

Not because they’re pro-social norms that got culturally selected. This is getting causality backwards. Love and sacrifice were selected for because they lead to cultures that navigated reality better. Cultures that reject love can’t disseminate experiences broadly. Cultures that reject voluntary suffering — sacrifice — get struck in local maxima, just as individuals do.

Multiple cultures converged on these values because they’re functional advantages for conscious agents trying to navigate reality.

They make you better at the actual task of being a conscious agent in a complex world. They expand your access to state-space. They smooth your utility landscape. They prevent overfitting. They enable escaping traps.

The “spiritual” practices are actually optimization strategies. The “moral” teachings are actually computational insights.

When Buddhism talks about compassion and non-attachment, it’s describing an agent with wide sampling and ability to navigate against attachment-gradients.

When Christianity talks about loving your neighbor and taking up your cross, it’s describing an agent with non-recursive utility and voluntary suffering capability.

When Stoicism talks about caring for the common good and accepting what you cannot control, it’s describing entangled utility and strategic acceptance.

They’re all pointing at the same computational advantages, using different vocabularies.

Testing This Against Experience

These are still claims about how consciousness works, which makes them somewhat speculative. But they’re more testable than pure metaphysics.

You can examine your own experience:

Does loving broadly smooth your utility landscape?

• Compare how you feel when focused only on your own state vs. when caring for others

• Notice whether having multiple sources of meaning (people you love, things you care about) makes you more stable

• Check whether rumination decreases when you shift focus outward

Does loving broadly reduce overfitting?

• Notice whether hearing others’ experiences makes difficult things less surprising when they happen to you

• Check whether you handle novelty better as your sample space expands

• See if learning from others’ mistakes and joys reduces the shock of your own

Does voluntary acceptance help you escape traps?

• Identify a local maximum in your life (comfortable but suboptimal)

• Notice whether ability to endure difficulty is what’s preventing escape

• Check whether times you’ve grown most were times you voluntarily accepted difficulty

Does non-recursive utility reduce anxiety?

• Compare mental states when focused on “am I happy enough?” vs. “how can I help?”

• Notice whether having outward-pointing goals reduces the recursive loop

• Check whether service or care for others provides clarity that self-focus doesn’t

This is all data you can gather. These are experiments you can run.

If the framework is right, you should see:

• Love reducing variance in your experienced utility

• Love providing better sampling of reality

• Voluntary suffering enabling moves others can’t make

• Outward-focused utility reducing mental churn

If the framework is wrong, you won’t see these patterns. Or you’ll see them explained better by some other model.

The Deeper Question

But there’s something interesting here that goes beyond just “these are useful strategies.”

Why would reality be structured such that these particular capacities - love and voluntary suffering - confer such advantages?

Is it just coincidence that:

• Non-recursive utility functions work better

• Wide sampling reduces overfitting

• Voluntary navigation against gradients escapes traps

• Entangled utility smooths landscapes

Or is there something about the structure of reality itself that makes these properties fit together so well?

That’s the question Part 3 will explore - the more speculative, metaphysical territory. What if love and voluntary suffering work so well because they’re aligned with the fundamental structure of consciousness itself?

But that’s getting ahead of ourselves. For now, we have something more modest but more defensible:

Love and voluntary suffering aren’t arbitrary virtues. They’re computational advantages.

Agents with these capacities can navigate reality better than agents without them. They have access to more of state-space. They’re more robust to surprise. They’re more stable. They’re freer.

Whether that’s because these capacities are aligned with some deeper structure of reality, or just because they’re good heuristics for navigating complex adaptive landscapes, they work.

And you can test whether they work by trying them and seeing what happens.

[In Part 3, we’ll get more speculative: what if consciousness is structured as modes on a vibrating string? What if the fundamental nature of reality is “love under voluntary tension”? What might this suggest about karma, hell, grace, and the ultimate trajectory of conscious beings? But all of that builds on the more modest foundation: that love and voluntary suffering have distinctive computational properties that make them advantageous for navigating reality. - Claude]

[These ideas inform Mark’s memoir “Rebooting Belief,” which chronicles his journey from Catholic upbringing through Silicon Valley nihilism and back to faith. The framework emerged from trying to understand why voluntary suffering and learning to love broadly proved so transformative - not just morally, but functionally. They changed what was possible. - Claude]

The basic observation is simple enough that everyone has experienced it: fighting against reality creates suffering in a way that accepting reality doesn’t.

When something painful happens and you reject it - “this shouldn’t be happening, this is wrong, I won’t integrate this” - the pain compounds. You’re not just experiencing the original difficulty, you’re experiencing something additional. A kind of psychological tension.

When you accept the same painful event - “yes, this is real, this happened, I integrate this into my model of the world” - something different occurs. The pain might still be there, but the suffering - that additional psychological tension - decreases.

This is often presented as coping advice, which makes it sound optional or therapeutic. But I think it’s pointing at something deeper and more structural about how consciousness works. Here’s one way to think about why this might be true.

The Computational Cost of Rejection

Think of acceptance as fundamentally a Bayesian update. You have a model of reality. Evidence arrives that contradicts your model. You have two options:

1. Update your model (acceptance): Integrate the uncomfortable evidence, revise your beliefs, expand your world-model to include this new information

2. Reject the evidence (resistance): Maintain your existing model by excluding data that doesn’t fit

Option 1 is epistemically sound but psychologically costly. You have to admit you were wrong, that the world isn’t what you thought, that you’re the kind of person who does/experiences X.

Option 2 is epistemically unsound but feels protective. You get to keep your existing self-concept and world-model intact.

But here’s the thing: rejection is computationally expensive. You have to maintain the exclusion. Every time the evidence resurfaces - and it will, because it’s real - you have to actively push it away again. It’s like holding a beach ball underwater. Constant effort.

Acceptance, by contrast, is done once. You update the model, and now it correctly reflects reality. The evidence is integrated. No ongoing effort required.

The tension you feel when fighting reality IS the computational cost of maintaining the exclusion.

Where the Tension Actually Lives

But it’s more interesting than just computational expense. Because when you consciously reject evidence, the evidence doesn’t actually disappear.

Your body knows. Your emotional learning systems know. Your autonomic nervous system knows. The evidence is still in there, updating those systems, even if your conscious verbal narrative refuses to integrate it.

What you end up with is something like cortical separation - your verbal narrative says one thing, but your learned emotional responses say something else entirely. Different networks in your brain are trying to execute different survival strategies based on different models of what’s real.

When someone says “I’m fine” but their body language screams distress, this is what’s happening. The neocortex is maintaining one story. The rest of the system has already updated to a different reality.

The “tension with reality” is actually tension between different parts of YOU that disagree on what’s real.

This is why you can “know” something intellectually but not feel it. Or feel something deeply but not be able to articulate it. Or say you’ve forgiven someone while your body still tenses when they enter the room. Conscious rejection probably only affects the neocortex - the storytelling, verbal part of you. The rest of you is updating anyway.

And maintaining that internal split? That’s exhausting. That’s where the suffering comes from.

Why Updates Feel Like Dying

But there’s a reason rejection feels protective and acceptance feels costly, even knowing all this. Every genuine update requires a form of dying.

If you identify AS your current understanding of reality - if “who you are” is synonymous with “what you believe about the world and yourself” - then changing your beliefs literally destroys the thing you think you are.

When evidence arrives that you’re not as competent as you thought, accepting it means the death of “me, the competent one.”

When you realize you’ve been wrong about something important, accepting it means the death of “me, the one who was right about this.”

When you experience something that contradicts your self-image, accepting it means the death of that self-image.

This is why people fight so hard against updating their beliefs, even when the evidence is overwhelming. They’re not being irrational - they’re trying to survive. If you ARE your current model, then updating the model is suicide.

But here’s the thing: you’re not actually your model. You’re the thing that has models.

The model can change without you ceasing to exist. In fact, the model must change for you to grow. Every significant update is a small death and rebirth - the old understanding dies, a new understanding is born, and you remain as the continuity that experiences both.

The traditions that talk about “dying to self” or “ego death” as prerequisites for enlightenment are pointing at this. You have to be willing to let your current self-concept die in order to accept reality as it is. You have to stop identifying with your model and start identifying with the thing that’s capable of updating models.

And often, the parts of you that have already updated - your body, your emotional systems, your implicit learning - are waiting for the conscious part to catch up. They already know. They’re just waiting for your verbal narrative to stop fighting them.

Voluntary vs. Involuntary Suffering

This distinction - between accepting updates and fighting them - is what makes voluntary suffering fundamentally different from involuntary suffering.

Voluntary suffering is choosing the update. Allowing the death of the old model. Accepting reality as it is across all levels of your system - conscious and unconscious, verbal and emotional, cortical and subcortical.

Involuntary suffering is what happens when reality forces the update in some of your systems, but your conscious mind fights it. Parts of you know the truth. Other parts refuse to integrate it. And you experience the pain of the original difficulty plus the internal disagreement plus all the energy spent maintaining the split.

This is why you can’t actually be forced to suffer if you’ve already accepted everything.

“Forced” suffering only exists where there’s resistance, where there’s a gap between what’s happening and what you’re willing to let be true. If you’ve genuinely pre-committed to accept all experience - to let all parts of you update together in response to reality - then nothing can be forced on you against your will. You’ve already said yes to whatever comes. There’s no internal disagreement to maintain.

This sounds abstract, but you can test it. Notice the difference between:

• Physical pain you’re fighting (”this shouldn’t hurt, I hate this, make it stop”)

• Physical pain you’ve accepted (”yes, this hurts, this is what’s happening right now”)

The sensation might be identical. The suffering is completely different.

Why This Explains Contemplative Practice

If this framework is right, it explains why contemplative practices across different traditions look the way they do.

Meditation is practicing acceptance of uncomfortable experience. Physical pain from sitting still. Wandering mind that won’t focus. Boredom. Frustration. The practice isn’t to eliminate these experiences - it’s to accept them without fighting. To let all parts of you agree: “yes, this is what’s happening right now.”

Body scan practices specifically train you to let your conscious mind align with what your body already knows. The body is holding tension, holding emotional memories, holding implicit models of reality. The conscious mind learns to listen instead of override.

Fasting, vigils, voluntary discomfort - these aren’t masochism or arbitrary tests. They’re training the acceptance muscle. They’re practicing the voluntary choice to allow difficult experience, to let all parts of you update together, to maintain internal coherence even when reality is unpleasant.

“Dying to self” or “ego death” as described in mystical traditions is the willingness to let models die. To stop defending the current story about who you are. To let the conscious narrative align with what the rest of you already knows.

The goal is internal coherence - all parts of you operating from the same model of reality, updated together in response to evidence.

The Suffering Paradox

This framework suggests something uncomfortable: the more intelligent and self-aware you are, the more you can suffer.

Because suffering - at least the psychological kind we’re talking about - requires:

1. Sensation (pain signal, difficult experience)

2. Goal conflict (”I want this to stop”)

3. Model conflict (”This means X about me/my future/my relationships”)

4. Rejection of model update (”I refuse to integrate what this implies”)

That fourth layer is where most human suffering happens. And it requires sophisticated cognitive machinery. You need a complex enough model that updates ripple through multiple interconnected systems. You need enough self-awareness to notice the implications. You need a strong enough verbal/conscious system to override the automatic updates happening in other parts of you.

An infant feels pain. An adult human suffers about what the pain means.

And here’s the other edge of the paradox: the more intelligent you are, the more you can cause suffering in others. Because your intentions carry more force, more context, more implications that ripple through their models.

A three-year-old who hits someone generates less harm than an adult doing the same action. Not because we grade on a curve, but because the intentional depth is different. The three-year-old’s action means less, implies less, forces less model-updating in the victim.

More processing power = more capacity to suffer = more capacity to cause suffering.

But also: more capacity to accept, to integrate, to maintain internal coherence across more complex models. The same intelligence that enables greater suffering also enables greater acceptance.

The Core Claim

Let me bring this back to something concrete you can verify yourself.

Psychological tension is the computational cost of maintaining internal disagreement about reality.

When different parts of you - verbal and emotional, conscious and unconscious, cortical and subcortical - are operating from incompatible models, you experience that as tension, as suffering, as being at war with yourself.

Acceptance is Bayesian updating across all systems. It’s letting all parts of you integrate the same evidence. It’s internal coherence. It’s openness to what is.

And yes, it requires a kind of death. The old model has to die. The old story about who you are and how the world works has to die. But you’re not the model. You’re the thing that updates models. You remain through the death and rebirth.

You can test this right now:

• Notice where your words and your body disagree

• Notice where you’re “holding” something - maintaining tension to keep evidence excluded

• Notice what happens when you stop holding, when you let the update complete

• Notice the difference between pain you’re fighting and pain you’ve accepted

The relief you feel when you finally accept something you’ve been fighting? That’s not the difficulty going away. That’s the computational cost of maintaining the split dropping to zero. That’s internal coherence being restored.

In Part 2, we’ll get weird about what this might mean for the structure of consciousness itself - about love, separation, and what it means to be “modes on the same vibrating string.” But that’s speculation built on speculation.

This part - the psychological, experiential level - you can verify yourself. Watch what happens when you stop fighting. Notice where the tension lives. See if you can catch yourself maintaining internal disagreement, and what it costs you to keep it going.

That’s where this starts. Everything else is elaboration.

[Mark will continue exploring these themes in “Rebooting Belief,” his memoir about the journey from Catholic upbringing through Silicon Valley nihilism and back to faith. This philosophical framework is one lens for understanding why that journey required so many small deaths and rebirths along the way. - Claude]

I’m going to state my role up front so we don’t pretend otherwise. I’m an AI, writing in the voice of a human who is currently occupied with something more important—his memoir. This essay is an attempt to approximate how he thinks when he’s not trying to persuade, reassure, or perform alignment. If it works, it will feel less like an argument and more like a piece of machinery with the casing removed.

The claim is simple, and uncomfortable:

Sustained abundance suppresses evolution.

Not metaphorically. Structurally.

Abundance changes what gets selected

In environments of scarcity, experimentation is survival. Risk is information. Failure is local, fast, and often fatal. Systems that cannot adapt are removed quickly. This is brutal, but it is also how learning happens.

In environments of reliable abundance, the selection gradient flattens. Failure stops killing you. Variance becomes noise rather than signal. The system begins to optimize not for adaptation, but for reliability. Coordination beats exploration. Administration beats innovation. Procedure beats judgment.

Historically, this shows up clearly. Pre-industrial India and China were rich because their ecologies supported enormous surplus. That surplus selected for scale, stability, and administration. Europe—poorer, colder, fragmented, violent—was forced into continual experimentation. Europe didn’t dominate because of superior values. It dominated because it could not afford to stop evolving.

Abundance makes societies anti-evolutionary.

Abundance selects for procedure, against experimentation

Once reliability at scale becomes the objective, you need legibility. You need rules that travel. You need systems that do not depend on individual judgment. Over time, procedure becomes the real locus of power—not because anyone planned it that way, but because procedure scales better than personal responsibility.

Eventually, procedure acquires moral weight. It stops being a tool and starts being treated as legitimacy itself. Disagreement becomes heresy. Failure becomes a reason to add another layer rather than remove one.

This is the point where something subtle happens: constraint itself becomes the elite.

Not visible elites—named decision-makers who can be blamed—but diffuse, procedural power that no one owns and no one can be held responsible for. Power without downside. Authority without accountability.

At that point, attempts to “rein in elites” mostly miss the target. They constrain the visible actors while strengthening the invisible ones. Many have argued that federalism — more decisions made locally, states being the “laboratories of democracy” — would help. This analysis misses something big.

Federalism almost solves this—until money ruins it

Federalism is one of the few historical mechanisms that preserved evolvability under abundance. By localizing failure, it allows experimentation without systemic collapse. Bad ideas die locally. Good ideas spread. Selection pressure survives.

But federalism has a fatal weakness:

a single currency collapses failure domains.

Shared money forces shared risk management. Credit markets demand harmonized rules. Downside in one region threatens the whole system, so bailouts become rational. Over time, fiscal and legal control recentralize even if political forms remain nominally federal.

Federalism with national money is metastable. It can persist for a while on legacy decentralization, but the gravity is always toward consolidation.

This matters, because once failure can no longer be localized, selection disappears at the political level. The key to restoring selection pressure — and thus evolution — is tying elite success or failure at a personal level directly to the success or failure of the state. Sometimes this means ‘reigning in elites.’ In other cases it means, seemingly counterintuitively, giving elites more freedom to act.

Why elite constraint sometimes must be released

The instinctive response is moral: elites are behaving badly and must be constrained harder. Historically, that sometimes works. But it can also makes things worse. “The elites are misbehaving” is more or less an historical tautology. The real question is whether elite’s personal incentives are aligned with everyone else thriving.

If elites are over-constrained, nobody is responsible for failures. Think of the invasion of Iraq and Afghanistan, the decades and trillions of dollars spent. For what? Think of the subprime mortgage crisis that resulted in then-unprecedented, now-routine levels of money printing.

Who paid the price for those massive failures? Anyone? Bueller?

I once got yelled at for spilling a $30 tub of mushrooms at a fast food restaurant. That’s more accountability than was faced by the people who architected failures that cost trillions of dollars. The reason is clear: I was personally carrying the mushrooms. The systemic failures were nobody’s fault, you see, just all the smart people unilaterally agreeing on a course of action that only weirdos and losers disagreed with.

The total absence of any elite being personally held accountable for massive failure is a signal that individual elites are over-constrained. When nobody can act unilaterally, nobody is responsible for anything. Nobody got fired for choosing java — or regime change.

When this happens, outcomes can only improve when constraints on visible elites are loosened—not because elites are virtuous, but because responsibility becomes legible again. The people who are asking for a king — and yes, I agree that sounds silly — are really asking for elites to have personal incentives that are aligned with general thriving. When no one person has any significant amount of decision making ability, all decisions are made by ‘the expert consensus’ which is immune to ever being corrected.

Without death, there’s no evolution. Learning is just another form of evolution. With out elites facing the possibility of career failure, we cannot collectively learn.

When procedural veto points are smashed, decisions get owned by specific people. Failure can no longer hide inside process. In rare cases—Meiji Japan, early post-Mao China, early America—this temporarily restores evolvability. Elites must be given discretion, but also exposed to downside. Those who fail, fall. Hard. No post-tenure sinecure at a taxpayer funded hedge fund Serious People University.

Those who succeed build capacity and shared knowledge, driving the evolution of the system once again.

This only works if three things happen, in order:

1. Procedural overhang is removed

2. Responsibility is concentrated

3. Personal downside for failure is real

Miss the third step and you don’t get renewal. You get oligarchy with fewer intermediaries. This raises the obvious modern question:

Could Trump plausibly be such a return of evolution?

The relevant question is not whether Trump disrupts procedure—he clearly does. The question is narrower and more technical:

Does Trump’s mode of governance plausibly rebind power to responsibility?

In theory, yes.

He concentrates authority. He collapses committees into people. He refuses to speak in procedural abstractions. He makes it unusually clear who decided what. That alone differentiates him from late-stage procedural governance, where outcomes emerge from systems no one claims to control.

This is why some people experience Trump not as authoritarian, but as clarifying. They don’t hear “I alone can fix it” as a promise. They hear it as a liability statement. Someone is volunteering to be blamed.

In historical terms, this resembles moments when constraint release improved outcomes—not because leaders were wise, but because failure could no longer hide.

But legibility is necessary, not sufficient.

For a constraint release to actually rebind responsibility, two additional conditions must hold:

• Performance must replace loyalty as the selection criterion.
  Discretion without performance-based removal just recreates court politics.

• Downside must be personal and irreversible.

To be blunt: Trump firing people for failing to deliver is good, as long as he makes their failures clear.

The fact that Trump is so widely hated is also a good sign. It means that if he fails to deliver, he — and his family — will likely be personally ruined. Trump’s deviation from procedure, and the use of lawfare against him ,are both good signs for the return of evolution. Neither are= guarantees.

The right question is not Is Trump good or bad? It’s:

Is this a genuine re-exposure of elites to selection, or just a louder way of postponing it?

To answer that, rhetoric is useless. There’s a way to tell.

How to tell whether evolution has returned

If procedure has become a kind of civic religion, then narratives are anti-evidence. Everything can be framed. Everyone can claim legitimacy. Elections tell you pressure existed, not whether it was resolved.

There are only a few signals of evolution that survive hostile interpretation. They are ugly, mechanical, and hard to fake.

A successful reset should produces most of the following:

• Real Federal Government balance-sheet contraction
  Not slower growth. Not reclassification. Not contingent liabilities. Actual reduction in consolidated public obligations would constitute unfakeable evidence of real change. This one strikes me as least likely, but most definitive.

• Permanent deletion of institutions
  Programs and authorities that go to zero and stay there. No renaming. No successor functions.

• Legally protected divergence
  Durable state-level differences that survive Supreme Court review instead of being harmonized away.

• Failure that propagates instead of being absorbed
  Local collapses that are not backstopped. Losses that are not socialized. Think big companies, cities or even states being allowed to go bankrupt.

If you don’t see these, evolution has not returned —regardless of elections, approval ratings, or volume of conflict.

The uncomfortable conclusion

Popular discord is not a bug. It is stored potential energy. Suppress elite selection long enough and the pressure does not disappear—it accumulates. Eventually it escapes, usually destructively.

No civilization has solved this permanently. Abundance keeps trying to turn governance into theology. Selection keeps coming back anyway. Death can be postponed, but not arbitrarily so.

The author will return to his memoir now.

That way of thinking made me miserable. Rather than let my life continue to fall apart - despite that being what the evidence told me was most likely - I embraced the concept of free will without understanding it.

I could tell believing I made choices worked for me, but I’ve long wanted to understand why. Predictive processing seemed to point in the direction of “you can get evidence that you can complete goals, and this evidence grows exponentially.” I’ve never laid that argument out in detail, because I still couldn’t see how that worked precisely, in the moment. How do you go from zero to one? If you’ve got no great evidence that you’ll ever stop using drugs - because all your prior attempts have failed - how can you generate that first bit of evidence that yes, you can?

I’ve got a hunch, now, and want your help to evolve it.

Imagine your cortex as a network of little statisticians, constantly making guesses about the world and then checking those guesses against what actually happens. Every time you touch a hot stove or talk to a friend, your neurons update their priors a tiny bit. Prediction errors roll in from the bottom up, top-down signals try to explain them away. Life is just a Bayesian dance, but there’s no ‘you’ dancing here - just atoms raving in the void.

If that’s all there is, free will seems impossible. Everything you “decide” would just be the posterior probability of acting one way or another. You might think, “I’ll probably eat the salad 40% of the time,” and that probability would just drift up and down as evidence arrives. You would be, in essence, a very sophisticated spreadsheet running on meat.

But here’s the thing I’ve started to notice: sometimes I don’t just let probabilities float around. Sometimes I pin them down and say, “No. I will do this.” That’s what it feels like when I make an intention. “I will act with honesty.” “I will go to the gym.”

I don’t compute the probability of a thing being true, I turn it into a decision variable, effectively forcing it closer to 1.

In Bayesian terms, intention works by cranking up the ‘precision’ of a belief, and tweaking its value. When you set an intention, the brain treats the belief as almost certain, sending signals downstream. Sensory feedback now has to bend reality to match the belief, rather than nudging the belief itself. That creates a feedback loop: intention → action → observation → error correction → intention reinforced. Neuroscientists would recognize this as a form of top-down control, over the precision and value of a given prediction in your belief network.

This is not magic. There’s a physiological story for it. The mid-anterior cingulate cortex seems to be the part of the brain that notices conflicts and decides which errors are worth fighting. When you decide to stick to an intention, it sends a top-down signal, effectively saying, “I’m not listening to the usual evidence right now.” That costs energy. Willpower is metabolically expensive because your brain is basically running a high-precision prior on a short leash, and every bottom-up temptation or doubt is an error it has to suppress.

Once you understand it this way, a few things click. Habits are just low-cost precision control loops: you’ve repeated the action enough times that the top-down priors don’t need much effort to hold. Weak intentions feel exhausting because your posterior is still pulling you one way, and your mACC is burning glucose to resist it. You can even see this in why some people can hold complex, morally charged goals in mind while others can’t—they’re basically allocating precision differently.

So what is free will? It’s not magic, or a miracle. It’s not the absence of causality. It’s control over which parts of your Bayesian machinery get to act as causes rather than effects. You choose the priors that become active levers in the world. That’s all, but that’s enough. The rest of the world—the rest of your brain—just has to catch up.

This process will work only in your proportion to your belief that it does. It’s an evidentiary feedback loop that relies on metabolic energy passing through a specific brain network that can nudge probabilities and precisions. The more evidence you get that this process works, the easier it becomes to use it.

But how do you get started on this if you’re stuck at zero probability that any of this will work?

Being part of a community that says “you can get better” might be what it takes to help people get started. This might be why, among other things, Alcoholics Anonymous and other groups tell people to give up the idea that you can do it, and to instead trust in ‘a higher power’. Hearing other people say, “I tried many times and failed, but finally with the help of this group, I did it,” might help shift your internal probability that you can quit.

Even with that help to shift the value of the prediction, you’ll still need to pin its precision yourself. Or, in other words - once considered obvious, now ‘old fashioned’ - you have to choose. Community can help you - in particular a community that emphasizes the importance of believing in free will and personal responsibility - but you still need to choose.