The brief take:

Demographic data suggest that the mortality rate of axolotls doesn’t increase with age. We can’t train epigenetic clocks to distinguish adult ages. Axolotls have amazing regenerative capacity, even at ‘geriatric’ age. All these factors point to the idea that they don’t age like most animals, just like hydra.

From an entropy perspective on aging, we’d define a non-aging organism as one that has sufficient repair and maintenance capabilities that it can completely resist or reverse any accrual of damage or dysregulation. Because no mechanisms to resist damage (such as DNA mutations) are perfect, non-aging is likely to occur only when there is sufficient regenerative potential for the organism to replace any damaged parts (cells, organs, limbs, etc.). Axolotls do indeed seem to be good candidates for this — they can even regrow damaged parts of the brain! If so, they might be the only known vertebrate to come close to non-aging. But it’s not technically immortality, since they have a definite lifespan.

However, there are a few caveats:

First of all, the demographic data is limited for axolotls. Even though this is published for other salamander species — in which there are no mortality rate increases with age — it will be important to confirm this.

Second, axolotls live very long for their size, yet they have age limits. In captivity, they’re pretty much all dead by age 13, though there is a record of a 21-year-old. Curiously, about a week before they die, they simply become lethargic and pass away – and we still don’t know why. Due to their long lifespan, lack of clear indicators for when an animal is about to die, and the costs of keeping them in captivity, it is not feasible to have large enough populations to really see what their maximum lifespan might be. But if the risk of death doesn’t increase with age, this would still make them a non-aging species, just one with an intrinsic vulnerability that eventually kills them.

What might that vulnerability be?

Lots of interesting questions still to answer!

Subscribe now

The deep dive:

Defining a non-aging species

How do we know if an organism truly doesn’t age?

It’s a hard question and there are generally three types of evidence scientists can bring to bear:

Demographic evidence: Does the risk of death increase in the species?

First, we should ask whether mortality increases with age.

If mortality stays the same across adult ages, it would appear the species isn’t aging — after all, if some physiological process is progressively degrading, this should eventually lead to increased risk of death. However, the challenge with demographic evidence is that it’s hard to follow many species long enough to be sure they wouldn’t age. If only we could just follow them a bit longer.

A 175-year-old tortoise may sound impressive, but maybe evidence of increased mortality only becomes apparent after 200. Same thing for a 400-year-old shark, a 500-year-old clam, and so forth. Just being old doesn’t mean you don’t age.

One nuance we can add is the force of natural selection. If we could follow a million individuals, watch 99% of them die, and see that the risk of death still doesn’t increase among the last 10,000, we could be pretty confident that aging is not emerging even at ages when natural selection has become very weak. Classical life history theory suggests that, if there are trade-offs between survival and reproduction, then aging should kick in within the period when selection is still strong enough to shape those trade-offs. Selection becomes substantially weaker as fewer survive to later ages and less reproduction remains.

There are species like this. Some turtles, sharks, and crocodilians show good evidence of little to no increase in mortality even at ages when the vast majority are already dead. But many researchers suspect that signs of aging might eventually appear if we could follow enough individuals long enough.

No one knows for sure. Hence, demographic evidence is not sufficient on its own to declare that a species doesn’t age, though clear increases in mortality with age almost certainly mean the species does age — with some wiggle room depending on our definition of aging.

Does the biology change in the species?

Second, in order for an organism to age, there has to be some physiological difference between its states at different ages. If, on average, there is absolutely no way to distinguish an 80-year-old from a 20-year-old — if the biology really isn’t changing — then how exactly could we say that aging is occurring?

Again, the challenge is to show that nothing changes. Even if we show that some things don’t change, maybe we simply didn’t measure the right thing, or didn’t follow the animals long enough.

Some things may even change in the opposite direction of what we would expect. If this is the case, then we’d have to ask whether this reflects increasing resilience with age, or compensation for aging in some systems we haven’t yet detected.

This is where certain kinds of biological evidence can become particularly useful. DNA methylation consists of patterns of chemical markings on DNA. Across the genome, there are millions of sites that can carry or lack such a methyl signature.

Many of these patterns change with age, and researchers have had great success building “DNA methylation clocks” or “epigenetic clocks,” that approximate chronological age or mortality risk in humans and many other species.

These clocks work because DNA methylation changes in response to so many aspects of biological state. Almost anything that changes in biology results in changes in methylation. It can also change randomly. Methylation patterns differ across cell types, so bulk sequencing patterns can also shift when the relative mixture of cell types changes.

Because methylation reflects such a vast range of biological processes, it’s hard to imagine any meaningful biological change that wouldn’t leave at least some detectable trace in patterns of DNA methylation. So if DNA methylation doesn’t change systematically across adult ages, that would be pretty striking evidence that at least some of the broad biological changes we associate with aging are absent or strongly suppressed.

Is non-aging plausible in the species?

A third type of evidence comes from whether non-aging is biologically plausible in the first place. If, as we do, you consider aging to be at least partly due to the accumulation of damage — DNA mutations, molecular errors, wear and tear, and dysregulation — then in order to argue that a species doesn’t age, we need some plausible mechanism by which damage or dysfunction doesn’t progressively accumulate.

We know that DNA repair mechanisms miss some mutations, and that perfect DNA repair is apparently impossible.

That makes true non-aging harder to imagine in organisms that retain long-lived cells or tissues that cannot simply be replaced. Because it is so hard to perfectly resist damage, the most plausible way to keep it from accumulating is not perfect prevention and repair, it’s replacement.

An organism that can just swap out damaged components — cells, organs, limbs, etc. — could, in theory, remain perpetually young.

In other words, high regenerative capacity might be key to avoiding aging. At the very least, it opens the door to the possibility.

• It should show perfectly flat mortality — no increase in the risk of dying — until way past the age when most individuals would have died under natural conditions.

• It should show little or no detectable systematic physiological change with adult age.

• We should understand enough about it to have a reasonable mechanism for how damage and/or dysfunction might be prevented, repaired, or replaced, continuously and indefinitely.

  ---

So,What Doesn’t Age Then?

Using the above criteria, we can ask what species seem to fit the bill.

The best known case is the hydra.

It clearly fulfills the first criterion: mortality remains flat. It also does well on the third: hydra continually replace cells and just slough off old ones, allowing constant renewal. To our knowledge, no one has tried to build an epigenetic clock in hydra, but continual renewal would at least suggest such a thing wouldn’t be possible, providing partial support for the second criterion as well.

My goal here isn’t to list every potential non-aging species, but some other strong contenders include planaria, immortal jellyfish, sea urchins, sponges, and some plants that reproduce with ramets — if you consider the genet the individual.

Do axolotls pass the test?

Under this framework, how do axolotls stack up?

^{Ve y Ilumina (Go and Illuminate) by Efrain Sandoval}

Well, first, their mortality rate so far appears flat with age, even though they clearly have a finite lifespan. The data are still limited, but with more animals followed over time, we should eventually be able to evaluate the first criterion much more confidently.

In terms of epigenetics, Maximina has clear data (currently under peer review) showing remarkable stability across adult ages. While you can use epigenetics to distinguish developmental and maturation stages, a striking epigenetic stability emerges once axolotls become adults. We can’t think of any stronger evidence one could reasonably ask for than showing that the epigenetic drift we associate with aging in mammals might be absent or highly suppressed in adult axolotls.

In terms of the third criterion, we also have plausible ways axolotls might avoid or counteract damage accumulation: this is where axolotls shine. They have incredible regenerative potential — they can even regrow their brain! We’re not sure we can conclude that no damage could or does accumulate, but at least this regenerative capacity seems like it might plausibly obviate all damage.

So, in theory, axolotls are prime candidates to meet all three criteria for a non-aging vertebrate. But before we declare them immortal, there is a catch…

Then why do they die so young?

Axolotls are not immortal — they do die, even if they don’t age. And here’s where things get really weird…

Not only do they die, they don’t even live that long compared with other salamanders. Many salamanders can live 50-60 years or more, but axolotls max out at 21 years, and most are dead by roughly 10-13. They become lethargic about a week before death, and then are soon found dead.

No one really knows why they die or what kills them — it seems random.

A potentially non-aging species that dies relatively young even under protective conditions is a strange conundrum. Their early deaths obviously aren’t due to predation, starvation, or accidents. Infectious disease in captivity cannot be completely ruled out, but it appears that there is some intrinsic cause.

In other words, maybe axolotls have avoided much of what we call aging — but pay a price in terms of some other vulnerability. Perhaps this other threat strikes randomly at any age and is an intrinsic weakness that is distinct from aging.

In any case, there is at least a decent case to be made that axolotls may be a non-aging species, but if they are, another mystery is awaiting:

What are they dying of and why doesn’t the risk appear to increase with age?

Such cool critters.

References:

Haluza Y, Gruhl B, Wagner A, Kurth T, Tanaka EM, Yun M.H. (2026). Axolotls retain fertility throughout lifespan. BMC Biology, 24(1):52.

Czarkwiani A., Lobo M., Castro L.A.B., Petzold A., Rost F., Maehr R., Yun M.H. (2025) Molecular basis for de novo thymus regeneration in a vertebrate, the axolotl. Science Immunology, 10(114).

Cayuela, H., Olgun, K., Angelini, C., Üzüm, N., Peyronel, O., Miaud, C., ... & Schmidt, B. R. (2019). Slow life-history strategies are associated with negligible actuarial senescence in western Palaearctic salamanders. Proceedings of the Royal Society B: Biological Sciences, 286(1909), 20191498.

Reinke, B. A., Cayuela, H., Janzen, F. J., Lemaître, J. F., Gaillard, J. M., Lawing, A. M., ... & Miller, D. A. (2022). Diverse aging rates in ectothermic tetrapods provide insights for the evolution of aging and longevity. Science, 376(6600), 1459-1466.

In 1996 in Italy, Carlo Petrini founded the “slow food” movement: a response to fast food culture, asking us to slow down, cook from scratch, and savor what we eat, all while respecting the workers and the environment involved in production. Fast food was convenient, quick, and cheap. It was efficient. But Petrini recognized that these monetizable virtues are not the only ones.

Today, science has developed its own version of fast food: “fast science.” We rush to publish results before we are scooped by a competitor, or before our funding runs out while we finish our Ph.D. We are judged by the number of publications and grants we produce, often more than by the quality of the work itself.

In this hypercompetitive environment, slow and steady does not always win the race.

Like fast food, there are advantages to fast science. The pressure makes people churn out results quickly. Standardization in the system enforces certain types of codified rigor. It makes procedures transparent and perhaps fair in certain ways. This system produces a lot of science!

But producing more science is not the same as producing better science. I covered some of the broader costs in my post on whether the system of science is killing science. Here, I want to focus on one particular problem: what happens when we prioritize speed, competition, and standardized approaches over understanding what our results are actually telling us.

Today’s brief take is simple: under time pressure, we don’t always double- and triple-check things the way we should.

We can miss errors, but even more importantly, we can miss the real, interesting story in our findings.

There is simply no substitute for trying an analysis in many different ways, probing little anomalies and giving ourselves enough time to change our minds.

This idea intersects with others I’ve written about: the other harms of too much competition, and the problems with rigid ideas about falsification and hypothesis testing that close us off to slow, careful exploration. It’s part of a broader argument I would make that science works as a system when there is a low level of competition and when validation of ideas, hypotheses, and paradigms can unfold slowly over decades. It works when individual researchers can follow their curiosity and passion with minimal constraints from external systems. So let’s all take a breath, slow down, resist the system, and do careful work!

Maximilien Franck: Too slow for modern science

A few years ago, I had a postdoc in the lab, Maximilien Franck. His CV wasn’t stellar when he applied, but he seemed thoughtful and curious so I decided to take a chance on him.

He was working on comparing profiles of inflammatory markers across populations. The project progressed in fits and starts. Maximilien often wanted to go back and double check things that to me seemed settled. I was sometimes impatient, and I started to see why he didn’t have that many publications: he was too slow. But he made good points and kept suggesting analyses that turned out to matter.

While the results of his project were interesting from the beginning, our interpretation of them, and thus our conclusions, shifted over time as we continued to analyze the results.

After one year, we could have published something.

After two years, we tentatively thought we had our story. We thought we understood what the data were saying, and we started to write up the findings. But even then, we went back and checked some things. We realized we didn’t trust some results and needed to remove one of the datasets. We replaced it with another, and this changed everything.

Before this substitution, we thought each population was very different from the others. But afterward, we realized that the two industrialized populations, the old one from Italy and the new one from Singapore, were actually pretty similar in their patterns.

Instead of our story being “Inflammation patterns with age are highly unstable across populations,” our story became “Inflammation patterns with age differ markedly between industrialized and non-industrialized populations.”

When slow science meets fast publishing

Now it was time to turn the work into a paper. I had been in contact with the editors at Nature Aging, and they had expressed interest, so I asked Maximilien to draft a 2500-word manuscript we could submit to Nature Aging.

Nature Aging is the highest-profile journal in the aging field, and it’s hard to publish there. You have to submit something at least close to the format they’re looking for, or you don’t have a chance.

Maximilien came to me with a 9000-word draft, far longer than the journal would ever publish. It went deep into the evolutionary origins of inflammation in an attempt to properly explain our surprising findings. I skimmed it, and I knew there was good material there, but also that it was too far off target to merit a serious revision.

So I sat down with him, and I asked, “Maximilien, did you even read the instructions for Nature Aging? Did you read any of their articles?”

“Yes,” he replied, “but I didn’t like them.”

This was classic Maximilien.

He was stubborn. He did things his way, without compromise. He felt the big journals focused too much on quick, short papers that didn’t take the time to consider a problem from all angles. It was so frustrating to work with him!

But here’s the thing: he was (and is) right. We’ve become so obsessed with speed, and with fancy figures that show thousands of analyses, that we’ve lost sight of what real science means.

It means taking the time to make sure you get the right answer.

It means correcting yourself when you make a mistake.

It means considering a problem from all angles, and then communicating this depth to your readers.

If we’re too impatient and competitive to do this, what business do we even have in this field?

After talking with Maximilien, we decided to pursue two complementary manuscripts. Together with another lab member, I reshaped the work into the shorter format required by Nature Aging, while Maximilien developed his broader synthesis and deeper take for a review journal, Aging Research Reviews. It ballooned to around 15,000 words! Both papers were solid in the end, and both were published mid-2025, complementing each other and getting a lot of media attention.

I learned a lot from Maximilien, both about the value of slowing down and about science itself. His interpretation in the second paper was something I wouldn’t have reached on my own. He’s currently working as a clinical research coordinator, rather than pursuing the traditional academic path of professorship. Fifty years ago, in a less hurried system, someone like him would have been perfect as a professor, but today’s cut-throat academic science system often rewards people who can move quickly, compromise, and learn how to play the game. We should all realize that that is to our shame, not his.

The Anomaly we almost could have ignored

Maximilien’s example is not the only one (though it may be the most striking), and this is not only a story about one unusually stubborn postdoc. This spring and summer, I’ve been working to finish a paper up with Stella Koo, an excellent MPH student. She’s been testing whether various metrics of aging and dysregulation are associated with health outcomes similarly across different populations around the world.

Again, just as we were finalizing the paper, we found a small, anomalous result. It was only a detail in one of the appendices. It didn’t change our conclusions. But it was weird: among the Tsimane, an Indigenous population from the Bolivian Amazon, having a higher biological age score seemed to predict a lower risk of mortality. That is the opposite of what a biological-age score is supposed to do.

So we investigated further. We didn’t rush it. And the further we probed, the more we needed to keep probing to get to the bottom of it. It took many months of back and forth. Kamaryn, my close collaborator and an excellent biostatistician, was deeply involved. We reran the analysis in many different ways, changed assumptions and checked whether our results would hold.

In the end, we found that the result wasn’t stable: it depended on some analytical decisions, so we can’t claim that this inverse association between biological age and mortality is a robust result. However, the opposite clearly wasn’t true either: higher biological age did not clearly predict greater mortality risk among the Tsimane. Something funny is certainly going on!

There’s a lot of nuance in that conclusion, and it will get a good, important paragraph in the Discussion when we write it up. It may not change the paper’s headline conclusion. But it changes what we think the results mean.

For both Stella and Maximilien, there was a natural arc to the process. First, we had to go deep into the analyses and let them lead us where they needed to. Second, we had to stay open-minded and curious, ready to change our conclusions. And then at some point it was done — not because we were rushing, just because we could feel that we had our answer. The data had divulged their secrets, like a fish that the fisherman finally reels in after it gives up the fight. When it happens, you know.

Why scientists can’t simply slow down

The problem of rushing and being impatient is one that we can all try to improve a bit. We can double-check our results, follow up loose ends, and try to slow down. These are individual efforts that every scientist should make.

But this cannot only be an individual responsibility. The system makes slow science difficult, and sometimes profoundly risky. If Kamaryn and I hadn’t adapted the inflammaging work to Nature Aging’s format, it would never have been published in a high-profile journal, and then the world would not have paid as much attention to it — even though it reveals striking and important findings. Anyone who doesn’t bend to the system is out of the system, and can’t make any difference anyway.

The challenge is, there are so many choke points. From early in their careers, students are evaluated at each stage based on their records: how many publications, and in which journals. If I care about my students, I need to help them get their publications out, otherwise they won’t have a promising career in front of them.

I also need funding for my research (and, because I’m in the NIH system, to actually pay my salary). That funding is so competitive — success rates are around 10-15% in most years — that I too need those publications to boost my CV.

Top journals have become accustomed to demanding submissions that are more and more massive. They generally include many multi-panel, complex figures that are often beside the point. It’s not slow science to create these huge monstrosities. It is simply additional pressure on researchers trying to get published, an effort to give a semblance of quality through volume and complexity of analyses.

We are all part of this system, and we can’t individually stand against it, whether or not we’d like to. It’s too large. Our individual efforts to do better will mostly be rounding errors. The best we can do is to push for change, in lots of ways large and small.

What would make slow science possible

1. Push for fewer rules and structures. Even when well meaning, funding agencies, universities, and departments have created a system within which we compete, and which constrains us in many unintended ways. Any rules we can make less rigid will help decrease competition back to healthy levels.

2. Push for new funding models. The current ways most countries allocate grants create hypercompetitive environments. There are other ways this could be done!

3. Push top journals to simplify. I often refuse to review papers that are too massive. If top journals start more generally to have trouble finding reviewers for certain papers, maybe their criteria will change. It is up to us to send that message.

4. Make the individual effort to do slow science when you can. After you get tenure, or on side projects without a deadline. When no student is dependent on it for their next fellowship.

5. Take a chance on students outside the system. I took a chance on Maximilien. Let’s not just reward those who follow the rules, and let’s keep as many contrarians in the system as we can.

6. Create a cost to blindly following the system. I review many grants and research articles that have followed the formula well. They’ve done nothing particularly wrong, but they lack a certain spark. Other reviewers often seem to feel that they should be rewarded for playing the game well. I am careful to always evaluate the science, the substance rather than the “grantsmanship” or other such factors — and I try to give a boost to the outsider who seems not to know the game, as long as they can show other relevant competence.

It will be an uphill battle, and we may not make much of a dent. But good science is like a good stew. It needs time to cook, time for things to settle in and come together just right. And if we’re patient, sometimes we’ll reap the rewards.

Maximilien never really learned to play the game. He’s a clinical research coordinator now, not a professor, and the field is worse off for it. But he was right about the manuscript, right about the anomalies you can’t rush past, and right about almost everything that mattered.

Anyone who doesn’t bend to the system is out of the system. That’s not a compliment to the system. It’s an indictment of it.

Scientists think that falsification is essential

Most practicing scientists I’ve met think that falsification and hypothesis testing are the core of the scientific method. I can’t count the number of conversations I’ve had where someone asks the question, “But what’s your hypothesis?” or “How would you falsify that?” And yet, when I talk with philosophers of science, they acknowledge some debate but generally concede that most practicing scientists don’t actually do falsification and clear hypothesis testing (more on that below).

This has led to an unusual situation: grants, scientific articles, and other scientific outputs and proposals are judged based on a criterion that no one really adheres to. Everyone goes to substantial lengths to go through the motions of the process, without realizing they aren’t really doing it most of the time. And, in my opinion, the best science is almost never done in the textbook version of the method when it is done correctly. What do I mean by all this? Keep reading…

Falsification and hypotheses aren’t necessary

I bet many of you, scientists and non-scientists alike, are thinking, “Wait, but scientists really do use falsification. Think of all those controlled experiments!” And it’s true, scientists do a lot of controlled experiments, though these experiments are not always geared toward falsification.

For example, in aging research, a researcher may study whether giving a mouse the compound rapamycin extends its lifespan; they might then repeat the experiment in another strain of mice that is missing a key gene that could bind with rapamycin and see if the effects are similar. These are certainly controlled experiments, but is there falsification?

The studies are not designed to try as hard as possible to falsify the hypothesis that rapamycin extends lifespan; rather, they are designed to elucidate what genes and pathways are involved in this process.

Yes, we might frame the whole process as a series of mini-falsifications, but any negative result is rarely taken as undermining the whole idea; rather, it simply shapes the next question. There isn’t much distinction between falsification and proof, in the sense that each result, positive or negative, is key information that leads to the next test. There is a continuous process of evidence and reasoning happening that really doesn’t make much use of falsification.

Hypotheses and falsification are even less relevant in many other non-experimental situations. Naturalists sit around for hours observing the natural world, and often simply cataloging things. For example, during my Ph.D., I relied on published data about the stomach contents of birds to figure out what they eat (insectivores, frugivores, granivores, omnivores, etc.). No one would deny that these catalogs are science, and yet there really was no hypothesis and no falsification. Similarly, there are many recent studies in computational biology that look for genes or proteins that are higher or lower in different settings.

In most cases, there is no a priori idea about what will be found — it’s purely exploratory. And yet these studies are changing our understanding of biology. We come in without hypotheses and nothing to falsify and we emerge (when the studies are well done, which is admittedly infrequent) with a better understanding of the biology.

Most broadly of all, there are whole fields of observational science where controlled experiments are either difficult or impossible: astronomy, geology, evolutionary biology, ecology, and epidemiology (randomized controlled trials aside). While it is possible in theory to generate a hypothesis in these fields and then look for observational data that might falsify it, it is often harder, and again it is rarely what happens.

Consider two of the most famous examples: Copernicus and Darwin. Copernicus didn’t come up with the idea that a heliocentric solar system might be a better model and then try to falsify it; he simply perceived that it was a more parsimonious explanation for existing data. Darwin didn’t come up with the idea of evolution by natural selection and then try to test it; he recognized that it was an efficient way to explain many observations he had been making for years, and gradually worked out the logic and supporting evidence.

Other less famous scientists in these fields function similarly: there is lots of back-and-forth between theory and data, but it’s rare that a scientist says, “Wait, maybe it’s this! Let me see if I can find any data that might disprove it.”

What scientists actually do (hint: it’s almost never falsification)

As should be obvious by now, scientists do a lot of things. We read the literature, observe nature, gather data, run experiments, build models, notice patterns, argue with colleagues, revise our ideas, and then gather more data. Sometimes we begin with clear hypotheses but many times we do not. Usually, there is back-and-forth between ideas and evidence.

In other words, science as practiced is largely an iterative process, with back-and-forth between ideas and data. In some cases, these are very small ideas, like the role of a particular gene in a particular pathway under particular conditions. In others, they are big ideas, like a theory of evolution or a theory of consciousness. Usually, there is some adaptation of the ideas to the data (we rarely get everything right before we get more data).

There’s also a darker side here. Once in a while scientists cheat to advance their careers or preferred ideas. More often, they are biased in favor of these ideas and don’t think critically, glossing over contrary evidence and interpreting everything in a favorable light. We all are subject to these biases and good scientists know this and figure out how to fight them and work around them. More on that below.

The art of pretending you have hypotheses

If you buy what I’ve said so far, you’ll recognize that scientists don’t usually do formal hypothesis tests and falsification even though they think they do. This weird situation means that they often twist themselves in knots pretending to test hypotheses, often with an unconscious form of self-deception.

The best example of this, which most scientists reading this will recognize, is inserting post-hoc hypotheses in a research article. As a hypothetical example, maybe I was analyzing whether some inflammatory markers predicted heart disease and then during my analyses I found that the relationship between inflammatory markers and heart disease was fuzzy and not so clear. However, there were clear relationships with diabetes. The problem was not in my exploration; its is in what came next.

How should I write up the results for publication? Well, everyone knows you’re not supposed to just say what you did in a narrative, and the final article won’t really be about heart disease, so I’d likely write the article as if I started off to test a hypothesis about diabetes. I’d likely do some research on why the relationship between the inflammatory markers and diabetes might have been worth exploring. Furthermore, I would write an introduction laying this out, setting myself up for a sentence like, “Here, we provide the first formal test of the hypothesis that XX inflammatory markers are associated with diabetes risk.”

Except I never had that hypothesis in the first place. I found the relationship and only after did I concoct a hypothesis. Maybe, if I did my work well, I double- and triple-checked my results, accounting for any biases due to the fact that it wasn’t my starting point. Maybe what I found is actually solid science! But the way we think about what we do makes it hard for me to be honest about the process, even if I want to be. Journals, reviewers, and readers all expect distinct Introduction-Methods-Results-Discussion sections, leaving little room to provide an honest narrative of our thought process. Even in journals with more flexible formatting, like the Nature journals, the actual process requires misrepresentation in order to squeeze in every word and key finding, while excluding extraneous results.

Misrepresenting the process we go through isn’t an afterthought — in some cases, it can be critical to evaluating the validity of the claims. If I check 1000 relationships, find one that looks interesting, and write it up without mentioning the others, I’m glossing over the fact that what I found is likely a false positive, a spurious result that won’t hold up. Reviewers, editors, and readers don’t have the information necessary to evaluate that risk. And in a world where science is all about attention, where we are more and more invested in our pet ideas, and where publishing in top journals requires strategy and compromises, this is increasingly a problem for science.

Falsification as a tool for self-reflection and critical thinking

Up until here, it probably seems like I’m completely critical of falsification in science. That’s not true; I’m against its misuse. Falsification does real damage when it is used both as a cudgel to exclude non-preferred ideas, and as a tool to enforce a narrow and incorrect view of how science is conducted. It is this dogmatism that does lots of harm.

But falsification, when practiced thoughtfully, can still be a key tool in science. We all have our biases and our preferred ideas. We’d like to prove them true and inclined to see evidence for their truth in everything we touch. Falsification can be our way to keep ourselves honest.

So, to sum up: falsifiability can’t distinguish valid empirical hypotheses from invalid ones, but the degree to which ideas can be challenged by evidence reflects the degree to which they can be considered scientific. Many valid, plausible, even likely ideas are only moderately scientific in this sense, and that does not render them unworthy of consideration.

This can happen at two levels. First, when framing our ideas, it’s crucial to do so in ways that maximally elucidate loss conditions — the conditions under which we’d admit that our ideas are wrong, or at least need major revision. It doesn’t have to be a formal test of falsification, but it goes in that direction.

For example, I don’t believe that epigenetics really controls the aging process; I think that DNA methylation and other aspects of epigenetics simply reflect many other biological processes. I think that apparent “rejuvenation” and reprogramming experiments are actually kind of trivial, showing that it’s possible to erase the accumulated experiences of cells, something that will make them behave like younger cells but not necessarily healthier cells.

Therefore, I should be willing to stipulate that if, in many species under many conditions, repeated epigenetic reprogramming allowed organisms to become essentially immortal, my ideas would be largely in error.

Second, when I conduct empirical research, rather than looking for evidence that supports my claims, I should look for evidence that refutes them. This is not because falsification is inherently better than proof or affirmative evidence, but because it keeps me honest — it keeps me from cherry-picking only the evidence I want to see. It’s almost like a discipline that good scientists should strive to achieve.

Used well, falsification can be a great tool for critical thinking. It asks us to state what would count against our own ideas, to look for evidence we would rather not see, and to let reality push back. That, more than the ritual of pretending every paper began with a pristine hypothesis, is where falsification can still help science.

The sixth grade version of science

Falsification is the idea that scientific hypotheses are only legitimate if they can be tested with clear experiments or observations that have the potential to disprove them. Anything else is not really science. For example, if I hypothesize that wolves are 100% carnivorous, the hypothesis is easy to falsify: find one wolf eating berries, grass, or some other plant material, and the hypothesis is disproven.

That doesn’t mean science is simply the business of proving things false. But it does mean that, for falsificationists, a scientific idea needs to have the potential to be falsified.

This idea was originally proposed and advocated by the famous philosopher of science Karl Popper in the early 20th century. As I understand it — thanks here to — Popper was reacting to fields like Freudian psychoanalysis and Marxist thought. Popper thought these ideas could be adjusted to conveniently explain away any pesky facts, and thus were not really falsifiable - and therefore not scientific.

The same logic though also works to try to wall off religion from science. What empirical test would falsify the idea that God exists? If we can’t name one, then in this framework the idea that God exists isn’t scientific. This doesn’t mean that God doesn’t exist, or that Freud or Marx were wrong — just that those ideas are not scientific in a Popperian framework.

Falsificationism became broadly accepted among scientists and is often the main thing most scientists know about the philosophy of science. This is science as we learned it in 6th grade: clear hypotheses with clear experiments and clear results. It’s also what most scientists think they are doing throughout their careers.

Roles of falsification in science

As just described, falsification actually serves at least three roles in science, not just one.

First and most broadly, it is used as a wall to separate what is science from what is not science. Religion, the occult, Freudian thought, Marxism, and various other beliefs are often treated as non-scientific because they are not falsifiable in this framework. In this sense, falsification can become a cudgel used by opponents of an idea if it is deemed not sufficiently testable. This can extend even to ideas that on the surface appear more scientific but are hard to test cleanly, such as the existence of a multi-verse or the idea that aging is largely due to accumulation of entropy. Having been on the receiving end of criticism that the latter idea is not sufficiently falsifiable, I have some stake in this.

Second, falsification also serves as a guide for how scientists think about hypotheses and testing — a sort of methodological paradigm that, true or false, shapes how scientists understand their own work and judge the work of others.

Third, falsification can serve as a tool for individual scientists to think more critically about their own work. My next post will discuss these last two roles.

Falsifiability versus testability and degrees of evidence

Not all evidence that is scientifically relevant relates to falsification. For example, if I hypothesize that wolves eat deer, I could prove this to be true simply by observing a wolf eating a deer. But falsifying it would require proving that no wolf has ever eaten a deer, which is a much harder task.

In this sense, many hypotheses that are testable may not be falsifiable, and working scientists often use both positive and negative evidence. Very few scientific studies show falsification, in fact.

Beyond the question of falsification versus proof, very few scientific tests are definitive, particularly for broader ideas. For example, let’s take the hypothesis that a small number of genetic pathways regulate aging rate across most organisms. Cynthia Kenyon’s work in the early 1990s made this idea plausible by identifying and showing mutations in nematode worms (specifically in the daf-2 gene) that would substantially extend lifespan. Further work identified other genes and pathways, and confirmed that these had similar effects in other species.

However, this story is not so simple. There also appear to be species that don’t age; most of the species with clear evidence are very short-lived; and there appear to be many aging mechanisms that are not controlled by these genes. The evidence is thus mixed, and most scientists in the field accept that the answers to questions such as this will come from many smaller individual observations and experiments, not from a single definitive test. (For some of those individual studies, however, a series of more definitive tests are employed.)

Note: We can distinguish falsification from proof when there is a single definitive test that either fully disproves or proves a hypothesis. However, this is not possible when accumulating smaller bits of evidence: each failure to falsify adds to proof, and each failure to prove adds toward falsification.

An additional point here is that we can think broadly about two kinds of evidence - evidence that is directly observable, and evidence that comes from consistency with other things that are considered well-known. For example, the evolution of the vertebrate eye is not directly observable, but is consistent with many things we know. Both are regularly used in science, and cases where directly observable evidence is sufficient are actually somewhat rare - usually, there is a deeper interpretation of facts that relies on their relationship to a broader paradigm.

How much falsifiability or testability is needed?

Advocates of intelligent design (the pseudo-scientific branch of creationism) often point to major gaps in the fossil record as proof that evolution through natural selection could not have produced life as we know it, and therefore that there must be an intelligent designer (i.e., God). A commonly cited example is the vertebrate eye. An eye is an exquisitely complex organ that requires many genes to function just right in its development, otherwise it simply won’t work. Everyone — creationists and evolutionary biologists alike— would agree that the chance of all the necessary mutations happening in a single generation is so low as to be essentially impossible, even across hundreds of millions of years of evolution.

Evolutionary biologists would contend that eyes evolved in a series of small steps, from simpler light-sensitive structures to more complex organs. Indeed, the Pax6 gene controls eye development across very different animal groups like insects (arthropods) and vertebrates, suggesting deep evolutionary continuity. (The common ancestor of arthropods and vertebrates did not have eyes, but biologists believe it may have had light-sensitive cells controlled by Pax6, and that over time these cells evolved, step-by-step, into eyes. However, since eyes are not preserved in the fossil record, there is no proof, and no intermediate life form still exists.)

But here is the awkward part: since no one has a complete fossil movie of the eye evolving frame by frame, all these various explanations involve inference and supposition. They ask us to connect comparative anatomy, genetics, development, and evolutionary theory into a coherent story.

You see where this is going: advocates of intelligent design can now claim that the explanation proposed by evolutionary biologists is not scientific because it is not falsifiable. It requires a leap of faith. And indeed it does require such a leap, if only a small one: evolution has produced many such innovations step by step, many are well documented, and to most of us practicing evolutionary biologists, there really isn’t any question worth fretting over here. But we can’t falsify our hypothesis, if we’re honest.

Even smaller leaps of faith are required daily in science. If I were to do a controlled experiment on genetically identical mice, I would assume that my result would, at a minimum, extrapolate to other genetically identical mice under the same conditions. If we can’t make such assumptions, then why gather evidence at all? And yet it is a leap of faith.

Maybe all experiments are manipulated by divine beings, and we only have the illusion of consistency. Leaps of faith at a very small scale are a constant in science.

Science as a gradient

In this context, I’d propose a different type of boundary. The key distinction between science and non-science is not whether an idea is falsifiable, but whether we allow our ideas to bump up against reality.

Is the the idea consistent with:

• Observations?

• Experiments?

• Logic?

• Mathematics and simulations?

• Established knowledge?

• Can we reject or modify it?

• Do we keep probing it and let it evolve with new evidence?

If yes, then the idea is participating in science, at least to some degree. If not, then the idea is walling itself off from science.

But this raises an interesting problem: almost every human idea bumps up against reality at least a little. For example, it is hard to imagine how to falsify the hypothesis that God exists, but this doesn’t mean there is no evidence whatsoever. Every day that God fails to appear in the sky and speak to us all is a missed chance for proof of God’s existence, and, as such, is a minuscule bit of evidence against God’s existence. Even the sum of all days without such an event doesn’t provide much evidence, but it’s more than nothing.

Likewise, even the most devout monotheists and atheists would likely abandon their beliefs if the full Greek Pantheon suddenly appeared and demonstrated their powers. No one in this world proceeds through life with zero opportunity for their ideas and beliefs to intersect with observation.

This leads to an interesting idea: We should no longer use falsification to wall off some ideas as “non-scientific.” Rather, we should look at science as a gradient, and as a mindset. Some ideas are more able to bump up against reality than others, and therefore more scientific. And when we try to encourage this bumping up against reality, that is a scientific mindset.

When we try to wall ourselves off from it to protect a preferred belief, that is less scientific.

All of us have our preferred beliefs and our biases — we are scientific to the extent that we are open-minded about them and try as much as possible to bring empirical observations, logic, math, and evidence to bear in a quest to find out if we are indeed right.

As long as good evidence can be brought to bear, falsification and falsifiability are not useful tests of how scientific an idea is, since evidence can be both positive and negative.

On absence of evidence versus evidence of absence

Most of us have heard of the distinction between absence of evidence (e.g., there’s no empirical evidence God exists) and evidence of absence (e.g., this doesn’t prove that God doesn’t exist). Despite this, many scientists consistently confuse the two, at many scales — from claiming that God doesn’t exist, to ruling out hypotheses that lack substantial evidence, to misinterpreting a p-value of 0.06 as evidence against the presence of a given effect. For example, I have had grant proposals on infections and Alzheimer’s disease criticized on the grounds that “there is no evidence” for the relationship, despite the fact that the lack of evidence is largely because, for decades, these same scientists consistently refused to fund research on anything other than their preferred hypothesis — the amyloid hypothesis — which was at best incomplete.

Discussions on falsifiability and testability can clarify such situations: we can identify the amount of evidence (broadly construed) both for and against a hypothesis, and then assess the extent to which new evidence could help shed light on it.

On physical explanations of the world as a default

One last note here on science versus non-science: many scientists assume that only physical explanations can be scientific. However, science is a method or a disposition, not a set of facts. There may be scientific explanations for phenomena we observe that are beyond our current understanding of the physical world. In fact, if history is any guide, there probably are discoveries coming that will rattle our understanding of nature in fundamental ways.

Excluding ideas simply because they are not easily explained in our current framework is not scientific, but rather the opposite. It may well hinder us from a better understanding and faster advances.

An interesting example is eastern medical frameworks such as traditional Chinese medicine (TCM) and ayurvedic medicine. The internal logic of these frameworks (qi, chakras, etc.) seems to be counter to Western medicine and physical explanations of the world, and they are therefore often dismissed as being “unscientific.” However, there is a non-negligible possibility that TCM and/or ayurvedic medicine could be reconciled with physical explanations, at least partially. For example, perhaps TCM doctors intuit larger patterns in the organization of biology. It should be possible to test whether, for example, patient proteomics profiles align with TCM diagnostic categories. If there is alignment, perhaps new understanding will emerge. Or perhaps there is no such alignment — but we won’t know if we don’t study it.

That said, physical explanations for phenomena we observe have done very well to date, and to my knowledge there are no clear examples where other explanations have been definitively substantiated. In this context, we should have some strong priors (in a Bayesian perspective) to prefer physical explanations when they can be proposed. Strong evidence would be required for any non-physical explanations. And of course, if there are explanations that go beyond our current physical understanding of the universe, it may simply be that we need to change our definition of “physical.”

What it all means

To sum up, scientists are sometimes too territorial and defensive, trying to exclude any idea they don’t agree with or fully understand because it is not falsifiable (even if empirical evidence can be brought to bear). However, this cannot be an excuse to open science to all manner of quackery. In this sense, we should abandon falsifiability as a way to wall off science from non-science. Instead, we should be explicit about the gradient of science, and should encourage all ideas that want to be considered empirically to maximize their chance of bumping up against reality — whether that means falsification, proof, or accumulation of many small bits of evidence. We should encourage an empirical, critical mindset that is open to all evidence.

Water, but not quite

In Japanese, there is no everyday word for “water” that encompasses both the water in the ocean and the water in a bath. Rather, Japanese distinguishes between “mizu” (water that is not hot) and “oyu” (water that is hot).

Bath water? Oyu.

Water for tea? Oyu.

Swimming in the lake? Mizu.

A glass of ice cold ___?

You guessed it, Mizu.

Obviously, there’s no scientific distinction between oyu and mizu. This is not like the clear phase changes between ice, liquid water, and steam. In that sense, English maps better onto the scientific categories. But that is exactly the point: for a Japanese speaker, the distinction between oyu and mizu feels natural and intuitive, even though it does not track scientific boundaries.

Mouse or rat?

There are also cases where Japanese tracks biology better than English. In English, we distinguish words for “mouse” and “rat.” If you see a rodent with a pointy nose and a skinny, hairless tail, your brain will almost automatically classify it into one category, usually based on size. In Japanese, there’s just one word for all such rodents: “nezumi.”

While there is a genus of mice (Mus) and a genus of rats (Rattus), they are actually quite closely related, and many things we would instinctively call mice (such as dormice) or rats (such as pack rats) are not closely related to this group at all. In fact, there’s actually a Wikipedia page as of this writing entitled “Types of animals known as mice,” illustrating just how poorly the word “mouse” captures biological categories of organisms.

One word, one illusion

By refusing to impose an arbitrary size distinction, nezumi arguably fares better than mouse/rat. But just try seeing one of these rodents and not mentally classifying it as either a mouse or a rat! When a large rodent crawled its way through the ceiling of my apartment and onto the countertop of my kitchen, I can most definitely tell you it was a rat, not a mouse! And also that I was definitely in New York, not Tokyo!

Both Japanese and English—and for that matter every other language I know—have a single word for aging. So our minds naturally assume there must be a single process behind it. But linguistics may be leading us astray.

What if there is no unified process of aging, but rather many partially overlapping processes unfolding in parallel?

So here’s the challenge I lay out to the field of researchers on aging: come up with a definition of “aging” that draws on borders that exist within the natural world, not just in our minds — one that can separate aging from non-aging and other related phenomena. What about trees that collapse under their own weight as they grow big? Age-related diseases that accelerate decline — are they aging mechanisms?

I think what we’ll find as we try to define aging is that we cannot do so without first developing a clear theoretical framework to understand the phenomena that relate to aging. Such a framework does not yet exist, and it makes aging one of the great remaining scientific mysteries. But I think the framework is coming, and I’m hard at work on it.

Why dimensionality is important

If aging is not a single process, then much of the way we talk about “slowing aging” begins to look suspiciously off target. It may still be useful shorthand, but shorthand can mislead.

If what we call aging is really a collection of partially distinct processes—some molecular, some structural, some disease-related, some environmental—then no single intervention should be expected to target all of it at once.

Think of it this way: if you’re traveling across the country, a 2-dimensional map will work pretty well but a one-dimensional map not so much! Knowing just your longitude wouldn’t get you very far toward your destination. And if you’re flying a plane, that third dimension becomes pretty important too (at least if you want to be able to land).

In any system we wish to study, understand, or manipulate, it’s crucial to understand its dimensionality. Is it one dimensional? Two? Five? One hundred? Fewer dimensions are easier to understand and handle, but too few can be profoundly misleading. Unfortunately, many scientists never ask themselves this question and end up doing the scientific equivalent of trying to fly a plane across the country with a one-dimensional map.

One caveat before we go further: even if measures appear separate or correlate only weakly, that doesn’t prove they’re separate. We might be measuring them imprecisely, or failing to realize that a single upstream cause can produce diverse downstream effects. We need to be cautious in our conclusions.

If aging is one-dimensional, we can probably safely say there is “an aging process.” If it has more dimensions, then there are multiple facets to aging. There are many ways we can ask whether aging is one-dimensional. Here, we’ll look at three:

(1) whether our knowledge of aging biology is consistent with a single dimension.

(2) whether various measures of aging correlate with each other well.

(3) what the structure of biomolecular signaling pathways related to aging can tell us.

Multiple biological mechanisms?

A review article from 1990 identified over 300 theories of aging, many of which were mechanistic theories. The list has only grown since then! But multiple mechanisms don’t prove multiple aging processes, if all these mechanisms have a single upstream cause. If DNA damage, for example, is the root of everything else, then there’s just one aging process with diverse manifestations.

Luckily, it’s actually pretty easy to argue that there are many aging mechanisms that are unlikely to share a common cause - in fact, we’ve already mentioned them. Tooth wear in mammals and wing wear in insects don’t depend on DNA damage, just physical abrasion. Likewise, changes in DNA methylation (one of the Hallmarks of Aging) don’t depend (much) on DNA damage, accruing instead due to a mix of regulation and stochastic processes of methylation and de-methylation that are, as far as we know, largely independent of DNA damage.

I could list many, many more, but it’s not necessary. Multiple, separate mechanistic processes contribute to aging.

Independence of aging metrics

If there is a single aging process, we should expect that we could try to measure it in different ways and still get a pretty similar answer. It doesn’t matter whether I measure your height in inches or in centimeters, with a ruler or with a tape measure - all the measures will be very tightly correlated with each other. And we know this will work because we know that height is a “thing,” a clear concept that has some objective reality to it.

So what if we try this with aging? Well, it’s been tried many times now by many different research groups. The first and probably most important study was by my colleague Dan Belsky (I was a co-author). He took telomere length, three different epigenetic clocks, and three different biological aging metrics based on clinical biomarkers (albumin, cholesterol, blood cell counts, and the like). All of these purported to be, to some degree or another, measures of biological age or of the biological aging process.

What Dan found was that there were moderate (but not strong) correlations among the three epigenetic clock measures, moderate (but not strong) correlations among the clinical biomarker measures, and no correlation among these groups, or with telomere length:

Since then, many other works have replicated these findings. Sometimes, a couple of the measures are a bit more strongly correlated, but sometimes some of the correlations Dan found to be moderate become weak. Here’s a similar result from a recent paper led by my student Guillaume Provost, comparing clinical biomarker metrics across five datasets:

Recently, many papers are also coming out with “organ aging clocks” (e.g., here and here and here) — they use proteins or DNA methylation signatures from different organs to calculate a biological age specific to each organ. This is interesting because the measures aren’t that tightly correlated — different organs age at different rates in different people.

In other words, the more we learn, the more we confirm that different aging metrics give us different information about aging. We aren’t looking at different facets of the same phenomenon, we are looking at different phenomena that may somewhat overlap.

Molecular signaling pathways of aging suggest aging is not a “thing”

The last argument for why aging is multidimensional is the most subtle, the most difficult to explain, but also the most interesting (feel free to skip ahead if you want to cut to the chase). It starts with this figure:

The bow-tie motif

The starting point for this figure is the five ovals in the blue rectangle in the center. These are the names of key genes in key pathways that many researchers think regulate aging. Then, to the left, we have the pink rectangle with the inputs. These are some of the things that regulate the signaling pathways in the middle. And on the right, we have the outputs, the things that those signaling pathways control.

The figure is vastly oversimplified. First, there are many more inputs and outputs. Second, many other molecules are involved in the signaling pathways. Third, the signaling pathways aren’t fully parallel, and lastly, the outputs influence the inputs. Nonetheless, at a conceptual level, this figure illustrates how these regulatory processes work.

The key takeaway is that natural selection produced these signaling pathways for a reason: to provide a regulatory structure with feedback that keeps the organism stable and healthy (thus the need for feedback from the outputs to the inputs). And the structure tells us something about that reason, about what’s happening here.

Anyone who knows machine learning knows that this figure looks like an autoencoder, a type of neural network used in statistics. And that is telling: autoencoders are powerful computational tools, and they were inspired by networks in the brain. This kind of network in biology is also called a “bow-tie motif,” for obvious reasons.

What bow-ties do

Bow-tie motifs are very powerful because they take a lot of upstream information (the inputs), pass it through a set of processing functions (the signaling pathways in the middle), and then use it to generate a coherent set of outputs. In other words, the organism needs to make a lot of decisions: how much energy should I invest in DNA repair? Do I need to mount an inflammatory response? Am I hungry? Should I reproduce? Should I increase or decrease immune surveillance?

The right answer to these questions depends on a lot of things — the inputs — but also crucially depends on what I choose for the other decisions. If I decide to mount a strong inflammatory response without sufficient protection against ancillary damage, I might do more harm than good, so maybe I need to upregulate DNA repair too.

It’s for these kinds of joint decisions that bow-tie motifs shine. Essentially, they take in all the upstream information, integrate it, and then make a coordinated set of decisions that are, jointly, the best way to respond to all that incoming information. And they do it again and again, every second of every day. They maintain our bodies in balance.

What the bow-tie means for aging

Now step back and look at the figure again and think about what it means. All those inputs — and many more — are funneling into networks related to aging. Those networks are controlling decisions on lots of aging-related processes, including the hallmarks, but also on other things, like reproduction and immune response. And there’s not just one node in the middle — there are at least five!

If there were a single node, we might argue that node controls aging, and maybe balances aging against things like reproduction. But five nodes? That means that the body has no single control point for aging-related processes. The only reason to evolve such a network is because, under some circumstances, the body needs to upregulate some aging-related processes and down-regulate others. For example, maybe in some circumstances we’ll need more DNA repair and more inflammation, as mentioned above. There are trade-offs among aging processes!

In other words, from the body’s perspective, and from evolution’s perspective, there’s no such thing as aging. Just a lot of processes that need to be coordinated, some of which might under some circumstances accelerate aspects of aging.

Aging is an afterthought: too fast a physiological decline isn’t good of course, but natural selection doesn’t solve this by trying to slow aging. The costs of each process are just part of the optimization algorithm.

Pulling it together

As should be apparent, there are very good reasons now to think that aging is not a one-dimensional process. This is not quite the same thing as saying there’s no such thing as aging, but it comes close. It certainly means that we should not trust any single metric that purports to give you a comprehensive score for your biological age.

I do think we can still think of “aging” as a field of inquiry. I think we can study aging-related processes and delve into potential ways we could define aging. I think entropy comes pretty close to providing a reasonable framework to see aging (though not quite exhaustive). But we certainly need to move away from the notion that aging is a single biological process, or something we can measure, or something we can “solve.”

So if aging isn’t one thing, why do we all talk about it as though it is? That’s not really a biology question; it’s a language question.

We hear a lot about aging and longevity these days: longevity bros, the anti-aging industry, and a steady stream of advice on how to eat, exercise, socialize, have sex, and otherwise optimize to live a longer life. The message is clear: aging is something we can measure, manage, and perhaps even slow. Sometimes the rhetoric goes further, suggesting that Ponce de Leon’s Fountain of Youth may not be a fantasy but just a few breakthroughs away.

The assumption everyone makes

All of this buzz rests on a major assumption: that there is a single biological process called aging. If such a process does indeed exist, perhaps it could be measured, tweaked, slowed, or reversed. And at first glance that seems both accurate and plausible. I mean, we get old, our hair turns gray, our skin wrinkles, our movement slows, our strength declines, and our risk of diseases rises until our ultimate decline: death. We recognize aging when we see it. Other animals appear to age too. Surely, there must be a biological aging process, tying all of this together.

A field without agreement

It turns out that the matter may not be that simple. In 2020, some colleagues and I at a symposium conducted a survey among ourselves to see whether we agreed on what aging actually is. We did not. In fact, we didn’t agree on much of anything (except the fact that we disagreed)! We ended up publishing a paper on the survey showing just how little consensus there is in the field. A key component of the disagreement was definitional: we could not even agree on what aging means or on whether we truly know what it is.

Around the same time, I published another paper, with the same title as this essay, What if there’s no such thing as “aging”? It was meant to be provocative, but it also illustrates a serious point.

The more closely I looked, the harder it became to defend the idea that aging has one clear biological definition. (It’s not one of my best-cited papers, but it’s my personal favorite. It lays out just how challenging it is to come to a single biological definition of aging.)

The problem of definition

Part of the challenge is that there are many ways to define aging, and they do not line up neatly. Something we clearly intuit as aging might not count as aging under a given definition, or something we think is not aging might be included under it. None of the definitions do a very good job of lining up with our intuition. Ideally, we’d like the world to be composed of natural categories (like clusters of dots in the figure below), and we’d like to have a vocabulary that describes these categories. If there aren’t any real categories at all (like in A and C) then it really doesn’t matter how we define something — our definitions are just arbitrary. If there are true categories, then we need to identify them and use the right words to describe them — like in B, but not D.

In the context of aging, the question is whether there are any true categories, and if so what definition can help us clarify them. Let’s look at a few examples to see how challenging this is.

A demographic definition

One common definition says that a species is aging if its risk of death rises with age. (Juveniles are allowed to become less likely to die as they mature, but in adults mortality risk should progressively increase.)

Indeed, there are at least a few species, such as the hydra, whose risk of death does not appear to increase with age. But essentially all mammals, birds, and many other organisms (including some plants!) clearly show a mortality risk that not only increases with age, but increases exponentially over time. We can conclude these species are aging, right?

The apparently immortal hydra

But now consider a tree whose risk of death rises simply because it has grown large enough to be struck by lightning, toppled by a strong wind, or to simply collapse under its own weight. Its demography may resemble that of other aging species, but is that really aging? We might feel intuitively that it’s not, unless there is some basic degradation happening.

A definition based on damage

Consider another definition. What about defining aging as the accumulation of molecular damage within our cells? As we age, DNA gets damaged, proteins misfold and aggregate — sounds like a plausible definition for aging, right? But that definition also seems incomplete. Some of the things we probably all agree constitute aging don’t depend on these molecular processes. For example, teeth wear down, or in insects, wings wear down. These changes may reflect simple physical wear and tear rather than cellular failure. Molecular damage may be part of the story, but it’s certainly not the whole story.

Maybe the problem is that the definition is too narrow. Maybe it’s not just molecular. Perhaps aging is the accumulation of damage across many levels. That would allow us to include both cellular damage and something like wing wear in insects. But if this is the case, then another problem appears: what about losing an arm in an accident? Is that aging? And if not, how is it fundamentally different from a DNA mutation? Each mutation is a kind of micro-accident. There are intermediate cases too: what about a scar that never fully heals? Isn’t wing wear or tooth wear an accumulation of small injuries? If aging is simply the accumulation of damage, why would losing an arm not count too?

Aging mechanisms and diseases:cause or consequence?

Consider another difficult case. Is Alzheimer’s disease a cause of aging? A consequence of aging? Both? Many people get old without ever developing Alzheimer’s disease, so it certainly could not be a necessary feature of aging. But it can clearly accelerate someone’s decline. So where does Alzheimer’s disease belong?

This gets at a key conundrum: is aging simply anything that tends to go wrong with time, or does it refer to some specific underlying process? If it’s the former, Alzheimer’s belongs inside our definition of aging. If it’s the latter, perhaps it does not.

Many researchers seem to believe that there is a core process or set of processes, more-or-less universal across species, that we could call aging. This view is epitomized by the well-known work on The Hallmarks of Aging. If so, Alzheimer’s should be a consequence, but not a cause. But certainly we know that Alzheimer’s, and many other chronic diseases, accelerate the process of decline. And none of these diseases are universal. So does this mean that there are aging mechanisms that are not universal? This is hard to square with the idea of a single underlying process.

“Aging” as an arbitrary word

These are just a few examples, but they point to a core issue. There may be no single “right” definition of aging. It’s just a word, invented by humans, that means whatever we say it means. We intuitively want the term to include things like rising mortality risk, functional decline, breakdown of our physiological systems, and wear and tear. We also want to exclude things like trees dying from lightning or losing an arm in an accident. But every definition we propose seems to either exclude some things we intuitively think are aging, or to include some things we think are not.

This raises an intriguing possibility: the everyday word “aging” may not correspond to any single natural category in biology. This does not mean that we don’t age (of course we do). It means that what we casually call aging may actually be a loose bundle of processes, only partially connected to one another. There may be no single biological process called aging at all.

The fact that scientists are trying to understand aging, to measure aging, to cure aging has nothing to do with whether an aging process actually exists; it may simply be a linguistic artifact: we happen to have this word, and so we assume there must be some reality behind it.

If that is right, then many debates in aging biology are not simply debates about evidence — they are also debates about category formation. We may still need the word “aging,” but we should be more careful about assuming that it names a single process waiting to be isolated. It may instead point to a family of processes — overlapping, interacting, but not reducible to one thing.

In the next post (Part 2), I’ll explore whether there is evidence for more than one aging process and what that all means.

When I launched Dynamic Entropy, I thought I was starting with a fairly simple task: expressing my numerous thoughts on how the body is not a machine, but a complex dynamic system. While the ideas were familiar, expressing them was not always straightforward.

To explain them properly, one quickly needs to discuss entropy, aging, health, physiology, evolution, moderation, optimization culture, biomedical reductionism, and possibly why the scientific enterprise sometimes behaves like a large committee trying to steer a bicycle.

So before adding yet another post to the pile, I thought I’d pause and offer a short digest of where we’ve been so far.

Subscribe now

The basic idea

The body is not a collection of independent parts waiting to be optimized one by one. It is a system in motion: constantly regulating, adapting, compensating, repairing, and occasionally making decisions that look very strange until we remember that natural selection was not optimizing us for wellness podcasts.

This is the central thread running through the first month of Dynamic Entropy.

In the introductory post, I argued that our bodies are complex dynamic systems: cells, tissues, organs, immune signals, hormones, metabolites, neurons, and behaviors all interacting continuously. The reductionist approach has taught us a tremendous amount, but it is often poorly suited to understanding what we casually refer to as “health.”

From there, we moved to aging.

If the body is constantly working to maintain order, then aging can be understood partly as the accumulation of disorder — entropy — across many biological levels. Not just in DNA. Not just in the epigenome. Not just in one tissue or pathway. Everywhere structure, information, and regulation have to be maintained, entropy is trying to make a mess of things.

The body fights back with repair systems, immune responses, wound healing, cellular turnover, metabolic regulation, and all sorts of error-correction machinery. But those systems are imperfect. Aging may be what happens as the imperfections accumulate.

This also complicates the idea that aging can be “reversed” by fixing one layer of biology. If aging is distributed across the whole system, then the solution is unlikely to be a single molecular Ctrl-Z.

Another theme this month was the strange modern urge to optimize everything.

We tend to imagine health as a mountain: climb higher, get better, optimize harder. Better diet. Better sleep. Better supplements. Better biomarkers. Better morning routine. Better afternoon routine. Better evening routine. Better way to worry about all three.

But biology rarely works that way.

Most physiological variables have ranges, not infinite directions of improvement. Too little can be bad. Too much can also be bad. The goal is not always more, or less, but the right amount in the right context.

This is why “everything in moderation” has more scientific depth than it usually gets credit for. It is not just folk wisdom. It reflects the fact that living systems are regulated systems maintaining a balance.

The next step was the idea that health is more like a plateau than a mountain. Most of the benefit comes from avoiding the cliff: sleeping enough, moving enough, eating reasonably well, maintaining social connection, avoiding obvious harms, and not treating every biomarker as a personal insult.

There may be a center of the plateau, and perhaps it is better to be near it. But most of us probably do not need to become elite health mountaineers. We just need to stay away from the edge.

Subscribe now

A major goal of this project is to move toward a true science of health.

That requires defining health in a way that is more precise than “not currently diagnosed with something” and more measurable than “complete physical, mental, and social well-being,” which is a lovely aspiration but a difficult research program.

The idea of intrinsic health is one attempt to solve this.

Intrinsic health is the body’s capacity to maintain dynamic equilibrium. Like a tightrope walker, the body does not stay balanced by freezing in place. It stays balanced through constant micro-adjustments.

Health, then, is not a static state. It is a dynamic capacity.

The challenge is to measure that capacity. Not by finding one magic biomarker, but by looking for patterns of resilience, robustness, plasticity, performance, and sustainability across time.

And then, naturally, the system of science….

Once one starts thinking in systems, it is hard to stop.

So we also looked at science itself as a system. Funding agencies, journals, universities, career incentives, publication metrics, and peer review all shape what science gets done.

The current system produces an enormous amount of research. But it also creates competition, standardization, groupthink, and bureaucracy. It rewards certain kinds of productivity while often making it harder to pursue the weird, risky, conceptual work that may be needed for genuine paradigm shifts.

This matters because if we want a new science of health, we may need institutions capable of supporting new scientific frameworks.

Apparently, even the study of complex systems is embedded in a complex system. This is either elegant or annoying. Possibly both.

Where we are going

So far, Dynamic Entropy has been circling one central idea from several angles:

The body is a complex dynamic system shaped by evolution to maintain itself against disorder.

Health is the capacity to maintain that dynamic equilibrium.

Aging is what happens as disorder accumulates and the systems maintaining order become less effective.

And medicine, if it wants to understand health rather than only treat disease, will need to take the whole system seriously.

That should keep us busy for a while, as we examine concepts like resilience, inflammation, the evolution of aging, falsification in science, and whether axolotls are immortal. Stay tuned, there’s lots more coming!

Thank you for reading during this first month, and for your excellent engagement and comments. If you are new here, this digest is a reasonable place to start. If you have been here from the beginning, congratulations: you are now part of a small experimental cohort in conceptual aging biology communication.

No IRB approval was obtained for your cohort ;).

If health is a plateau, what should you do to be healthy?

This framework gives us several broad principles for how to live healthy lives:

1. Solve the biggest problems first. Are you a gym rat who nonetheless hates cooking and survives on fast food three meals a day? Stop optimizing your workout and start working on your diet. Are you sleeping only 4 hours a day so that you can have time for your 3-hour workout and to prepare fresh vegan meals from scratch? Maybe worry a bit more about your sleep, and less about your diet and physical activity.

2. Don’t do anything too weird. There are much greater risks following a strange diet far from anything our ancestors ever ate than from being slightly less than optimum.

   

3. Don’t sweat the details. Each of has different genes, a different history, different bodies. The science just isn’t good enough now to provide very specific advice on the details of health, and as I’ve argued, it won’t make much difference anyway. So stop worrying about it! Instead you should…

4. Relax and enjoy life! While a bit separate from the main theme of this post, there’s great evidence that psychological state is as important to physical health as anything. It’s certainly not worth being stressed in order to try to get every aspect of health perfect.

So stop trying to be the world’s most intrepid health mountaineer. Instead, do things you love, avoid doing things that are obviously harmful, and enjoy the ride!

Lifestyle

Yes, in fact. Emerging research (including mine, self-congratulatory back pat) is increasingly showing both the evolutionary and physiological bases for this principle, at least in regards to diet and exercise. Let’s look at this from three perspectives:

Subscribe now

Modeling moderation

Before we start looking at biology, let’s consider how we normally conceive of the relationship between quantity and quality. Most of us have a model in our heads that looks something like this:

The reason we think this way is because it’s simple. Our brains are built to find simple patterns in what they observe, so we need to train them to understand and detect anything more complex. In fact, even this quantitative model is a bit more than our brains like. Ideally it would be simply lots=good, little=bad.

This model provides us with simple general rules on how to judge things. It can be easily inverted when appropriate: in the case of trans-fats or fuel consumption, lots=bad, little=good. There is also cultural variation in how we see things. Americans tend to view large things and large quantities as good much more than in some other places, notably Japan. Just look at portion sizes, for example… Still, everywhere in the world there is a clear tendency to use simple rules to judge things.

Of course, the real world is sometimes simple and sometimes complex. Most of the time, when there is reason behind a simple rule such as more is better, the rule still has its limits – it is just that our normal experience exposes us just to some of the possibilities. For example, for any healthy food we can imagine, it would presumably be possible to eat so much we die, just as a function of overstressing the capacity of our guts to process so much. If you don’t believe me, imagine trying to force down 100 pounds of spinach per day.

In other words, the graph above, when valid, is almost always a close-up of a more general trend, such as these:

All this really means is that an infinite amount of anything would not be infinitely better, and in many cases would actually be worse (as represented by the right hand graph). However, the rule “Everything in moderation” is based on a different scenario, one in which our day-to-day experience offers us the chance to have either too much or too little:

This principle of non-linear dose effects has a fancy name, hormesis, a concept often applied to medications. The fact that we give it a special name indicates that we don’t expect this sort of pattern, though in reality we should. It’s just that it goes against our linear pre-conceptions.

Subscribe now

The physiology of moderation

If you go to your doctor and she takes a blood sample, she’s likely to measure things like cholesterol, HDL, sodium, calcium, and so forth. For many of these things, our general rule is more is better (HDL) or less is better (total cholesterol), and there are established medical thresholds for defining risky levels. But for many other things such as sodium and blood cell counts, the thresholds go in both directions. There is a normal range, and you can have too much or too little. Even in most cases where there is a clear direction that is considered bad, this is usually because our lifestyle has changed since we lived like our hunter-gatherer ancestors:

Based on some studies of modern subsistence populations (hunter-gatherers and others that are close to this), it seems that cholesterol levels, for example, are much lower than in developed countries. Cholesterol is essential for metabolism, and it should certainly be possible to have levels that are too low. It’s just that people living in developed countries are unlikely to have those levels, but are very likely to have high levels that would start to cause problems.

In fact, in order for our bodies to function well, there needs to be tight regulation of an entire network of things – hormones, lipids, electrolytes, vitamins, the immune system, and so forth. These things regulate each other in complex, still poorly understood ways. For example, in chickens, eating more vitamin E increases activity of one part of the immune system and decreases another, BUT this relationship is inverted if the chicken also eats more omega-3 fatty acids. For the moment, we don’t know much more than that there is regulation, lots of it, and it is complicated.

For this regulation to function well, it needs to respond to external conditions. When we are stressed, we need to have the fight-or-flight response. When we eat, we need to absorb the nutrition. When we are cold, we need to reduce blood flow to extremities. Most of these responses pass through the regulatory network described above. So the system is not static – it is dynamic. Nonetheless, generally speaking, we can identify a range of normal values for almost any molecule in the network, and in most cases we can define both upper and lower thresholds beyond which things won’t work as well. Levels that are too high or too low may be directly problematic, or they may simply be indicators that something else in the system is not going well.

The relationship between lifestyle and the levels of these molecules is not always straightforward. For example, it is not clear that eating more cholesterol increases the level of cholesterol in the blood, based on how it is digested. Excess vitamin C is simply excreted, so it’s hard (though not impossible) to overdose. Nonetheless, for most components of our diet, there is a normal range of intake, and deviating too far in either direction could have consequences. Likewise, our physiology changes when we exercise. It’s probably good to avoid extremely high or low levels of exercise in order to avoid the extreme physiological effects that could result.

All of this is to say, based on relatively simple knowledge of how our bodies work, that it makes sense in most cases to consider moderation as a guiding principle.

The evolutionary basis for moderation

The physiology I just described is fairly basic, but it makes sense to ask why our bodies function in this particular way. Why not build bodies that just work better the more good stuff we eat? Why not build bodies that are fine no matter what our intake of different things, or no matter how much we exercise? Evolutionary theory provides some answers to these questions.

One popular misconception of evolution is that it is always building the best possible (i.e. optimal) organism for each environment. If this were true, humans would have wings like angels – it would be better, right? In fact, evolution works slowly over long periods of time making (mostly) small modifications to existing forms, and this creates a fair amount of inertia. It’s not easy to make the jump from four limbs to six limbs (I believe I’m correct in saying that it hasn’t even happened once in the last 500 million years), and it’s not easy, when you use testosterone and estrogen to regulate sexual development and breeding cycles, to find new hormones to take their place.

So the starting point for any organism is that it has a network of molecules that regulate each other and maintain it in homeostasis. In order for this regulation to work, and in order for the system to respond to changing conditions, there need to be physiological consequences to changing the levels of different molecules. And this means that some levels must be better than others, at least in certain situations.

It is possible – and indeed often observed – for the normal levels of different molecules to change across species. For example, unlike most mammals, humans do not synthesize vitamin C in their livers, meaning they have to get it from the diet. As a result, circulating levels of vitamin C are much lower in humans – presumably we have evolved to be perfectly healthy with much lower levels than are needed in other species. There are many such examples. At the same time, natural selection favors individuals who can tolerate a wide range of environments. Our ancestors lived in a wide range of environments and ate diverse diets. Some ate mostly meat, others ate huge quantities of shellfish, and others still large quantities of grain. Conditions could change from year to year or century to century, and so flexibility in nutritional and exercise needs must have been a good thing.

But in evolution, as elsewhere, everything has its price. Evolutionary biologists call these trade-offs: you can’t have your cake and eat it too, and you can’t (if you’re a diving bird) build a body that is light enough to fly well and heavy enough to dive well. It appears that there are trade-offs operating on flexibility too. Many evolutionary biologists think that it is often possible to build bodies that tolerate a wide range of conditions, but that doing so makes them function less perfectly in any of the conditions. For example, birds that migrate or need to survive a cold northern winter are often shorter lived than birds that stay in the tropics year-round.

Moderation in practice

What does all this mean for moderation? It means that over time, as environments change, organisms will change their physiology in response. The optimal level of cholesterol, or protein, or vitamin E, will change so that it corresponds to what the animal is likely to and able to take in. There will be some tolerance for straying a bit from this optimum – not many organisms could survive if a slight deviation was fatal – but there will not be infinite tolerance.

And so, in the end, organisms will be healthiest if they live lives not too different from those of their ancestors. Their physiology is optimized for it.

This general principle, easy to see for diet, should apply to a whole host of scenarios. We are accustomed to thinking of marathon running as a healthy hobby, and of people who do it as extremely fit. It is true that marathon runners are able to train their bodies amazingly, and have stamina beyond most of us – it is this fact that makes us admire them. But for exactly the same reason – that it requires so much effort above and beyond what seems the typical task of our bodies – that we should doubt if it is really good for our health in the long run.

Evolution surely gave us bodies that could, as necessary, adapt to the physical hardship of different environments (such as marathon training), but that does not mean we are better off in those environments, or that we pay no price for these adjustments.

If marathons are a reasonable approximation of the physiological exertion our ancestors faced, they are probably healthy to train for, but if they exceed that, they are probably unhealthy.

Of course, it is hard to know exactly what kind of life our ancestors led. Modern hunter-gatherers and similar subsistence populations may or may not be similar to our ancestors, and even they are quite diverse for just about every aspect of their lives we study. Still, many aspects of our modern lives are clearly far from what our bodies are built for, and to the extent that this is true we are probably hurting our health and accelerating the aging process.

Caveat: I am NOT arguing that we need to replicate our ancestors lives in every way. Getting attacked by jaguars, suffering from malaria, and starving once in a while are hardly desirable. We just need to recognize that our bodies were designed to make the most possible out of that world, not the one we live in. Give us 100,000 years of modern life and we’ll probably have changed that…

Musing

We know a great deal about disease. But ask someone to define health, and often the conversation becomes strangely awkward. Their mouth begins to fumble, brow begins to furrow and eventually they utter some version of “well…not disease?”

Modern biomedicine has become extraordinarily good at identifying diseases, mechanisms, pathways, risk factors, and broken parts. This has produced remarkable successes. But it has also left us with a surprisingly thin idea of health: health as the absence of diagnosable disease.

Subscribe now

The wellness industry offers a different but equally incomplete picture. A constant barrage of ads would have you believe health is something like a mountain peak: a state you reach by optimizing your workout, diet, sleep scores, supplements, bloodwork, and probably your cold plunge schedule too.

Both views talk constantly about health without quite answering the basic question: what is health?

Bodies are not machines assembled from separable parts. They are complex dynamic systems, shaped by evolution to maintain themselves in changing and unpredictable environments. They regulate, compensate, repair, buffer, and adjust. They do this constantly, usually without asking our permission or notifying us in the morning.

Health, from this perspective, is not a perfect state. It is not the top of a mountain. It is more like the capacity to stay on the plateau: to remain stable, adaptive, and functional across a wide range of conditions.

Aging, likewise, may not be a single biological program quietly running in the background until we fall apart. It may be better understood as the gradual accumulation of disorder in systems that are constantly trying, imperfectly, to hold themselves together.

That is the idea behind Dynamic Entropy.

Entropy is the tendency of things to drift toward disorder over time. Dynamic entropy is what happens when the thing drifting toward disorder is also walking around, eating lunch, answering emails, repairing DNA, regulating inflammation, and trying to maintain a coherent sense of self.

In other words: life.

System of Science

A recent post on Systems of Human Performance provocatively asked whether science is good for us. It’s an interesting post with much to discuss.

Here, I’d like to ask a related question: whether the systems we’ve built to do science - funding agencies, academic journals, research institutions, supporting industries, career structures, etc. are actually helping advance science. It’s a long subject, so I’ll keep this post as an overview, opening up questions for future exploration.

Let’s frame it as two lists: ways the system helps, and ways it hurts:

Ways the system of science helps science:

1. Delivers huge amounts of funding. Not all research requires lots of money, but some does, and various funding agencies have truly succeeded in getting lots of money to scientists to advance their research.

2. Facilitates efficiencies and coordinates large projects. Many research projects are too massive to be undertaken by a single researcher. Centralized initiatives like the Human Genome Project are able to help scientists coordinate and make their research more efficient.

3. Draws more people into science. In 1958, the US conferred about 8800 doctoral degrees for a population of 175 million. That’s about 1 per 20,000 people. In 2022, it was 57,600 Ph.Ds for 333 million, or 1 in 5800. That’s incredible growth, and doesn’t include the numbers of non-Ph.D researchers.

4. Standardization accelerates progress. Many aspects of the system are standardized - from the formatting of scientific articles and grant applications to methods and procedures for research. This makes it easy to follow the recipes and generate more results, faster.

Overall, I don’t think we can question that lots more science is happening than used to. And in some ways, progress is accelerating: each discovery may unleash many more discoveries, in an exponentially growing system. But is the actual science itself better?

Ways the current system of science hinders science

1. Creates too much competition. At many levels, scientists compete more than ever. A little competition can be good, but when it’s too intense it ends up deforming the science toward whatever wins the competition, not toward the best science (Goodhart’s law). Those who win the competition are often not the best scientists, but rather those who are the best at playing the game as it is. And with their success comes power: power to reshape the system to even more favor those who play it like they do. Neurodivergents - those who often have the most unusual ideas and the kind that might radically change things - are more excluded now than ever, because they won’t win at the increasingly sophisticated game.

   1. For funding. My Ph.D advisor, Robert Ricklefs, was one of the world’s foremost ecologists. He used to sit alone in his office writing theoretical papers, reshaping the field. He got the occasional grant, but he didn’t need any funding to be great. In today’s world someone like him would struggle to survive: the universities want researchers who bring in big grants so they can get a cut. At US biomedical research institutions, faculty salaries are paid off by grants, not by universities. The competition is intense, and failing to win means you’re out of a job.

2. Draws too many people in. The increase in the scientific work force could be seen as good, but it also creates problems, notably:

   1. Accelerating competition. See above.

   2. Lower average researcher quality? Not all scientists back in the day were great, and many of the new people being drawn in are great - for example, they may be the cream of the crop from previously under-represented groups that rarely got the chance. That’s a good thing. But it seems likely (I have no data) that on average expanding the pool of scientists has lowered the average quality by letting in people that would never have made the cut before. However, the pace of progress is often determined not by the total volume of research, but by how quickly the reigning paradigms change. Adding more lower-quality researchers may thus result in little net gain. Lots of what is funded and published is slop that won’t change much.

   3. Increases the scale of the system. Having so many people in the system means that you are less likely to be known individually when you apply for a job. This means people rely more on statistics - h-indices, numbers of papers published, dollars of funding - rather than really knowing your work. And this increases the competition and exacerbates Goodhart’s law.

3. Standardizes what shouldn’t be standardized. As the science system has become more and more complex and interconnected, many things have become standardized in various ways. There are rules - sometimes formal, sometimes informal - for what is considered good science, and they are often applied rigidly, to the detriment of anyone thinking creatively outside the box. For example, randomized controlled trials can be a better type of experimental design under some circumstances, for some questions, but the rigid adherence to the idea that they are a gold standard prevents case-by-case consideration of context.

4. Creates groupthink, disincentivizes dissent. To get a grant funded by NIH, I need three reviewers to agree that it’s excellent. If even one of them doesn’t like it, I have essentially zero chance of getting funded. This means that I will never, ever submit a controversial idea for funding - it takes a lot of work to prepare a grant application, and if there’s much risk that any reviewer might disagree, it’s not worth my time. Top journals also aren’t likely to publish anything too radical. This kind of incentive structure means that existing dogmas get heavily reinforced, whether it’s a reductionist paradigm, the idea that amyloid beta is responsible for Alzheimer’s disease, or the idea that RCTs are a gold standard. It means that we spend decades and billions of dollars going down dead-ends, wasting careers, because we don’t allow dissent in. Some of the dogmatic ideas may be right, and some wrong, but science works only if we encourage dissent.

5. Bureaucratic nightmares. There are increasingly complex webs of interconnected rules that researchers need to navigate. In order to work with a colleague’s dataset, I may need to obtain a data transfer agreement negotiated between lawyers at the two universities, I may need an IP agreement, and ethics approval. All this may be required before I can get funding, and it can take months or even years to make it happen. More and more of our time as researchers is spent navigating these rules which are established to mitigate very small risks - and with little possibility to use human judgement and waive them when they are absurd.

These lists are partial, made just to give a flavor for some of the key issues at stake. Future posts will delve into some topics in more detail. Overall, I think it’s likely that the current system of science is net-accelerating progress, but not necessarily by much. And if we imagine a world in which funding and publication were decoupled from career success, I think we’d make much faster progress. There is certainly substantial room for improvement, and most of it would come through having fewer rules, procedures, and bottlenecks: by just letting scientists do science, however they want, with minimal interference.

The Hidden Coordination Keeping You Alive

Sit for a minute. Breathe deeply, five seconds in, five seconds out. Repeat this a few times. Notice the sounds and smells around you.

As you relax, many things are happening in your body. Your breathing and heart rate are pulled into coherence. Cascades of neurotransmitters are released in your brain, triggering not only changes in your emotions and mood, but physiological responses throughout your body. We are only beginning to scratch the surface of how profoundly such a simple act transforms your physiology.

A Body is a System in Motion

Everything we do, everything we sense, and every change that happens within our bodies is a catalyst instigating further change, part of a never-ending cascade. Our cells, tissues, organs, and bodies are never static; they are dynamic patterns in continual flow, talking to each other in ways we’ve never imagined. This is how we stay alive, adapting to our ever-changing internal and external conditions, battling the forces of entropy which would otherwise rapidly decompose us.

In other words, the body is a complex dynamic system: a system made of many interacting parts that influence one another continuously, often in non-linear ways. The whole cannot be understood simply by adding up its parts. This simple reality, so obvious upon a moment’s reflection, is nearly absent from current biomedical science.

The dominant reductionist paradigm tries to break us down into the smallest of pieces - usually molecules and cells, sometimes tissues and organs. Applied in medicine, this perspective can make us think of humans as collections of organ-specific diseases, each to be tackled one-at-a-time by specialists in distinct fields. This perspective assumes the whole can be understood merely by summing up its isolated parts, but it rarely considers how different components interact dynamically, or how higher-order processes like emotions might influence lower-order ones such as cellular function.

The reductionist paradigm has given us many major advances, both in understanding and our ability to treat disease. But it is reaching its limits. Increasingly, we are applying it not just to solve major, discrete diseases, but to fine-tune our bodies in an attempt to assert control, optimize health, and slow aging.

We are over-reaching, micromanaging in ways that may do more harm than good. Living systems are not always improved by more interventions: sometimes micromanaging a complex system creates new problems.

Complex Systems Are Everywhere

Our bodies are not the only complex dynamic systems. We are surrounded by complex dynamic systems: ecosystems, economies, traffic networks, social networks, swarms of ants. There is a whole field of research into these systems. They are not easy to study, but they can be studied. And crucially, they share features: the behavior of one complex dynamic system can often be extrapolated to understand how others behave. For example, we have applied models derived to predict ecosystem collapse to predict when patients on hemodialysis are likely to die.

Welcome to Dynamic Entropy

What would it mean for science to truly embrace this obvious reality of our bodies as complex dynamic systems? How would that impact our understanding of the aging process? How has natural selection sculpted our physiology if everything is so connected? How should we live our lives to maintain health day by day? How would this perspective change the way we think about medicine? Might it even have lessons for how we structure our societies, to help make them more resilient?

This substack, Dynamic Entropy, attempts to answer these questions by blending deep science with relatable examples, always approaching the subject with humility and fun.

What is Aging, Really?

What is aging? It might surprise you to learn that there is nothing close to scientific consensus on this seemingly simple question. I have even argued that there might not be any such thing as aging in a scientific sense. These are themes we will discuss in detail in future posts.

For now, I want to briefly consider the idea, recently growing in popularity, that aging is entropy: an accumulating increase in disorder in our bodies over time. This idea is not new - it can be found as early as 1944, espoused by Erwin Schrödinger in his book What is Life?

Entropy and the Body’s Fight Against Disorder

Briefly, the second law of thermodynamics states that any closed system will tend over time to dissipate its energy toward greater states of randomness. Ordered patterns break down. For example, if you put hot water on top of cold water, eventually the heat will diffuse and be evenly spread in the cup. The universe tends toward disorder, toward random distributions of energy.

In the 1930s and 1940s, Claude Shannon applied the same idea to information, developing the notion of information entropy. Mathematician John von Neumann famously told him, “You should call it entropy, for two reasons. In the first place your uncertainty function has been used in statistical mechanics under that name, so it already has a name. In the second place, and more important, no one really knows what entropy really is, so in a debate you will always have the advantage.”

Just as thermodynamic entropy breaks down structures in energy, information entropy breaks down organizational structure over time. For example, my desk gets more and more cluttered over time.

Your body also accumulates entropy. There are DNA mutations, shifts in position of cells within tissues, unhealed injuries, tooth wear, dysregulation of the immune system after response to certain pathogens, and many others.

Our bodies (and my desk) are not closed systems. We input energy into the system to fight entropy: I clean my desk to reduce disorder, and we eat food to give us energy to fight the natural entropy occurring in our bodies. That energy from our food is directed to many processes, from DNA repair to immune responses to wound healing. In information theory parlance, these are error correction codes, mechanisms that help to maintain the incredible amount of information (i.e., non-randomness) our body needs to keep us alive. Without this energy and these error-correction codes, we die and rapidly decompose.

Note that the information in our bodies is not just our DNA. It’s all the ways in which our body and its components are not randomly organized. It’s the position of each cell in a tissue, the position of each organ and limb and feature. It’s the ratios of each pair of proteins in a cell, and all the synaptic connections between neurons in the brain. It’s the dynamic patterns of electrical signaling happening everywhere. It’s many things I’ve never thought of - but all of which, if made random, would make our bodies or their parts not function properly.

From an entropy perspective (which we could also call an information theory perspective), aging is thus not simply entropy. It is the failure of the error-correction codes to prevent entropy, to prevent the loss of this crucial information.

DNA repair enzymes don’t catch every mutation. Healed wounds still leave scars. At each level of your body, these mechanisms are imperfect. As they fail, and as entropy accumulates, our bodies become less energetically efficient and more imperfect at preventing further entropy. The process accelerates exponentially.

Under this view, we can see development as a gain of information, and thus we can map information in our bodies across the life course, with development slowing but never stopping, and aging starting at conception and slowly accelerating.

Why This Matters

If this is all true, entropy is happening at every level of biology. David Sinclair’s Information Theory of Aging assigns it principally to the epigenome: methylation patterns on the surface of the DNA. Certainly entropy happens there, but it also happens everywhere, not all traceable to the epigenome. This is a crucial observation: it means that aging cannot be meaningfully reversed simply by changing the epigenome. Entropy, in this view, will be very hard to fight.

A reductionist view invites the hope of a simple cure for aging, but a complex systems perspective suggests the very opposite.

Entropy is thus a convincing framework that explains most of what we know about aging. But as the title says, this is a brief take, and there are many nuances and caveats that future posts will explore. How does entropy relate to variation in lifespan across species, and why are some species able to avoid aging altogether? Can entropy exhaustively explain aging? How does entropy relate to complexity during aging? Is entropy always bad? How can we measure entropy? Stay tuned for more on these questions…

A note on this piece: I originally wrote this essay for Martin Picard’s The Science and Experience of Energy. We’re adding it to the Dynamic Entropy library because intrinsic health sits right at the center of this project: understanding health not as a perfect score on a dashboard, but as the body’s capacity to stay coherent, adaptive, and surprisingly competent despite the universe’s ongoing attempts to make a mess of things.

What is health? This seemingly simple question is surprisingly hard to answer.

The World Health Organization defines it as “a state of complete physical, mental and social well-being and not merely the absence of disease or infirmity.” This definition is good as far as it goes, but it has major challenges: how do we measure health? What are the limits of health? If I am unhappy because I can’t afford a luxury car, is this a decrement to my well-being, and therefore my health?

In a recent publication with Martin Picard and other colleagues, we tried to tackle this problem. We argue that health can be defined not just as a vague aspiration, but as a property of our biological and psychological systems. It’s something that can be understood scientifically and ultimately quantified.

Imagine what it would mean to have a scientific way to measure health: You could see if your new diet is working for you. We could test medical interventions based not on whether they improve one or two symptoms, but on whether they benefit you overall. We could follow the health of populations in an objective way, assessing what policies are helping or harming us.

In a recent post here, Martin alluded to the need to approach health from a first principles perspective, and to understand how energy is part of health.

In this post, we’ll take it a little further, digging into those first principles to try to build a coherent scientific framework to understand health empirically.

Health of the Body Versus Health of the Body in Its Context

Imagine two 45-year-old women who are generally in “good health.” Both get into a car accident, and both become paraplegic.

One is a writer who has never been into sports very much. She lives in a city that is very accessible and wheelchair friendly. She is wealthy enough to afford changes to her lifestyle to accommodate her new physical state.

The other is passionate about marathon running and has constructed much of her identity around an active lifestyle. She lives in a mountainous, rocky area in a small town with very few wheelchair-accessible buildings. She is not wealthy and cannot afford to renovate her house, or to move.

Most of us would probably agree that the car accident had a bigger impact on the health of the second woman than the first, even though there might be no meaningful difference in their physiological states.

In other words, health is not just about the state of our body, but about the context that body is living in: our values, preferences, environment, culture, perceptions, behaviors, and lifestyle. We call this the “lens of context.”

Of course, each of us has a lens of context that is unique, and this is a huge obstacle for trying to come up with a universal, scientific understanding of the principles of health. So we need to make a distinction between intrinsic health (the state of our body as it permits health, generally speaking) and realized health (health as we experience it, via our own individual lens of context).

Understanding realized health could be a whole academic discipline in itself. It is important, and perhaps this Substack will explore it in a future article. But here, we’ll focus on intrinsic health.

Where does intrinsic health come from? A First Principles Perspective

Rather than starting from a reductionist perspective, trying to build up a definition of health based on our understanding of the individual molecules and cells that make up our body, let’s tackle the question from a broader perspective. What is an organism? What does it mean to be alive, and how do organisms achieve this? The answer to these questions lies at the intersection of two fields: complex systems theory and evolutionary biology.

Complex systems theory is about the ways that all the components of our bodies (molecules, cells, tissues, organs, etc.) interact with each other, with feedback loops that mean you can’t always make simple predictions about how a change in one part of the system will affect things elsewhere.

For example, if I eat less because I’m trying to diet, I might lose weight. But I might also trigger my body to react to a shortage of incoming calories by both storing more fat and prompting a hunger response. I might inadvertently gain weight, due to my body’s natural mechanisms to self-adjust.

Complex systems theory is the study of many such systems—economies, ecosystems, weather systems, traffic networks, and, of course, our bodies. These systems have many interconnections and feedback loops, often making their behavior unpredictable.

Evolutionary biology adds another layer. It says, our bodies are not just complex systems. They are complex systems that have been optimized over billions of years for a reason: to propagate themselves. This optimization has led to fine-tuning of the networks in ways that help organisms survive and reproduce as well as possible.

While reproduction is of course very important in evolutionary biology, the first and most important task any organism has is to hold itself together: to survive long enough in this chaotic world that it might have a shot at reproducing.

In this perspective, an organism is a collection of molecules, cells, etc. that has been optimized to self-adjust to preserve its own internal dynamic state at least long enough to grow and reproduce.

This idea might not sound that radical—it may even sound obvious—but it is strongly at odds with the dominant biomedical paradigm. To understand why, consider the analogy of a tightrope walker.

The typical biomedical paradigm would argue that we can understand tightrope walkers and their balance by decomposing the system into its parts. For example, we might remark on the relationships between individual muscles involved in maintaining balance and note that when tightrope walkers fall, they tend to have a particular muscle in the left calf that is contracted. This might start the search for a drug to decontract that muscle in order to restore balance, without ever noticing that the contraction was in fact an attempt to restore balance.

Another key aspect of the tightrope walker is the dynamics of their muscles. Any tightrope walker who freezes all their muscles and becomes rigid will gradually tilt to one side and fall. On the other hand, a tightrope walker swinging wildly back and forth from left to right is also at high risk of falling.

The only way to maintain balance well is through microadjustments of many muscles happening all the time in an effort to maintain dynamic equilibrium. And indeed, this is what our bodies are doing every second of every day—not in two or three dimensions like the tightrope walker, but in thousands of dimensions as they adjust countless aspects of our biology to constantly changing conditions, inside and outside. Intrinsic health is thus our body’s capacity to maintain that dynamic equilibrium.

The Pillars of Intrinsic Health: Energy, Communication, and Structure

What does it take for an organism to maintain intrinsic health? Again, tackling it from a first-principles perspective, we can identify the ingredients: energy, communication, and structure.

Energy is required to do any work, and there’s lots of work to be done. The second law of thermodynamics tells us that dispersal and disorder (known as entropy) increases in any closed system. In an open system, entropy can only be countered by energy inputs that help maintain information, structure, and order.

However, sending 10,000 volts of energy into a corpse does not bring it back to life. The energy has to be guided, directed, and used in appropriate ways at appropriate moments in appropriate places. This guiding and directing requires a sophisticated internal decision-making apparatus that senses current conditions and sends energy or other resources where they are needed.

Communication is this decision-making apparatus, the software that makes our bodies run. It might seem like decision-making happens largely in the brain, and indeed the brain has a crucial role, but the decision making also happens within each cell. With molecular and bioelectric signaling happening at many scales, decision making that occurs throughout our bodies is often independent of the brain. Any information in the body that is shared and integrated to contribute to self-adjustment is part of communication—nearly every biochemical reaction is thus part of it.

However, communication and energy don’t happen in a vacuum. Again, context is key.

Energy is mediated by specific molecules (such ATP) and structures (such as our limb structure, which affects the energetic efficiency of movement). Communication also happens via specific molecules, structures, and pathways.

In other words, communication and energy need to be understood in the context of biological structures at different scales—whether it is the helical structure of DNA, organization of cells within a tissue, or the shape of anatomical structures such as blood vessels.

Structure is qualitatively different from energy and communication in that it is a limit on the others, or a context within which they occur. Nonetheless, it is impossible to understand an organism and how it maintains its health without considering how health is shaped by biological structures. And changes to these structures may affect health profoundly.

Putting It All Together—Intrinsic Health as an Emergent Property

The energy, communication, and structure that are necessary for health don’t operate independently; they work together, synergistically.

Intrinsic health is not just the sum of its parts.

In much the same way that a battery, wire, and solenoid coil come together to generate a magnetic field that can’t be easily understood or predicted from the components, intrinsic health emerges from the dynamic interplay of energy, communication, and structure within the organism.

It is, in complex systems parlance, an emergent property. All this really means is that it cannot be easily understood by decomposing it into parts. In the same way that we cannot fully understand joy by studying neurons or fully understand weather by studying atoms, we can’t fully understand intrinsic health by just studying its elements separately.

This means that there is no specific state of each component that determines health (though some states may be incompatible with health). Rather, health results from the degree of coordination and self-adjustment that happens dynamically, as every aspect of the system responds to constant slight fluctuations. This suggests that health will best be measured not by a snapshot of the organism, but by the dynamic patterns of variation over time and in response to changing conditions.

Measuring Intrinsic Health

In order for the concept of intrinsic health to gain scientific traction, we need to be able to measure it. But if we can’t break it down into constituent pieces and sum them up, how can we do this?

The answer again comes from complex systems theory. Stable, resilient systems—be they ecosystems, climate systems, or economies—tend to have specific dynamic signatures. Human biology shows the same patterns.

For example, we have shown that the same models that predict ecosystem collapse can be used to predict which hemodialysis patients are at highest risk of death, based not on the levels of blood markers but on their fluctuations over weeks and months. Again, the tightrope walker example is instructive. If we were to measure the contractions and decontractions of many muscles in a tightrope walker, we would easily detect characteristic signatures of someone keeping their balance well, versus someone swaying wildly back and forth.

There are also likely to be energetic signatures of health. In a healthy organism, energy is used efficiently and is broadly available when needed. Martin and his team’s work on the protein GDF15 and energy resistance shows that an imbalance in energy resistance creates key signatures that are broad markers of health status, and other signatures are likely to be discovered.

In the end, we don’t think there’s any one thing we’ll be able to measure that will precisely quantify intrinsic health. Rather, we think that we’ll identify many measures that capture part of the signal.

Some of them may measure energy, or communication, or structure. Some may measure what we call intrinsic health components—resilience, robustness, plasticity, performance, and sustainability. Yet others may integrate some or all of these elements.

We can then use various statistical techniques to figure out what all these measures have in common, and that is what we can call intrinsic health. Once we identify this gold standard measure, we will also need to identify simple, low-cost, and feasible ways to approximate it, and validate that these measures do indeed strongly predict all the health outcomes we care about: mortality and chronic diseases, but also quality of life, physical function, cognitive function, and independence.

What Does It Mean For You?

The intrinsic-realized health framework gives us a simple, intuitive, and theoretically justified way to understand many key notions in health.

Does health decline with age? Well, realized health might not if the contextual factors are right, but intrinsic health will certainly tend to decline.

Does having a cold impact your health? Well, as long as you recover fully, your intrinsic health never changes—your immune response did exactly what it should all through the process. But your realized health might have suffered in the meantime.

And how does trauma affect health? Early-life trauma can get biologically embedded, leading to long-term decrements in both intrinsic and realized health, though both might be modified with appropriate treatment.

Intrinsic health is not just useful at a conceptual level, though. It should also be measurable, as I described earlier. While it will take some time to derive and validate good measures, there is every reason to think this effort will succeed.

And once we have measures, we will have a single target to use in health studies. No more wondering or worrying about how coffee is good for heart disease, but bad for cancer, but good for dementia, but only if you stand on your left foot holding your nose while whistling. (I’m exaggerating, of course, but there’s a grain of truth there.) We can simply ask how coffee consumption impacts intrinsic health—across people in general, or for you specifically. Or how your new exercise regimen is working for you. Or whether subsidizing fresh produce at grocery stores improves the health of communities. Pick your question, and it’s now easier to answer…

Lastly, the conceptual work behind intrinsic health is part of a movement.

It challenges the notion that there is a pill for everything, and that improving health means identifying more pills we should take every day. It will position health as something that can be measured along a gradient, and something we can work to build, preserve, and restore across the entire life course, preventing diseases and maintaining quality of life at all ages.

The Optimization Trap

We are used to thinking about health as a mountain to climb. We try to optimize our workout, our diet, and our sleep patterns to get to the peak of fitness at the top of that mountain. We are barraged by media selling this idea, from the magazines in the check-out aisle at Whole Foods to blogs and YouTube channels to the ~$210 billion supplement industry. Each little tweak we make promises us better and better health, an idea promoted wholesale by the many industries that stand to profit from our optimization obsession.

But how good can we get our health? Is it really worth all that effort? There are both evolutionary and physiological reasons to think that investing time, money, and effort into our health will not yield indefinite benefits, but rather diminishing marginal returns. Health is more like a plateau, and our goal should not be to reach the peak – just to avoid falling off the cliff!

Diminishing marginal returns in health

“Diminishing marginal returns” is just the fancy economist’s way of saying that the more you improve things, the harder it gets to improve them more. There are tons of examples of where this principle applies in economics, in daily life, and in biology.

The example I love best is sweeping your floor. Imagine you arrive at a cabin in the woods that hasn’t been cleaned in ages. The first time you sweep, you get a lot of dirt up. But it’s still dirty! You give it another pass. This time, you still get a fair amount, but less than the first. After 3-4 passes, it’s not worth your time to keep sweeping because each successive pass gets you less dirt than the last, and at some point the additional little bit of clean isn’t worth the effort.

Other examples: beyond a certain point, spending more to air an ad campaign more will yield diminishing marginal returns because many of the people you might convert have already been converted. Producing more DNA repair enzymes will have diminishing marginal returns for reducing mutation rates because the most of the mutations are caught even with lower levels of the enzymes. You get the idea – if you start looking, diminishing marginal returns situations are all around us.

If we make a graph of diminishing marginal returns, it looks like this:

Below, we’ll dive into why health will show diminishing marginal returns for our effort

.

The best physiology is moderate

Let’s now link this idea of diminishing marginal returns up with my last post, on The Science Behind “Everything in Moderation.” There, I argued that most physiological parameters have an intermediate optimum. If your body makes it or requires it, it’s probably there for a reason. You could have too much, and you could also have too little. Almost everything needs to be at the right amount. We’ve shown that this applies to dietary nutrients, for example, in this publication.

What this means is that the effort to get ourselves healthy is not one of going as far as you can for a given nutrient or physiological parameter, it’s about being in the middle. And this means we get a graph like the one above, but symmetrical:

You’ll notice that the shape we have here looks closer to a plateau than a mountain. And of course there isn’t just one physiological parameter we need to do this for, there are many. We can expand our figure and add in an additional dimension, representing health as roughly a plateau, with the “optimum” state in the middle:

Under this analogy, longitude and latitude of the mountain represent two dimensions of physiology/health, such as physical activity level and cholesterol consumption. Altitude represents health – the higher the better. In reality, of course, there are hundreds or thousands of dimensions of health, but the principles don’t change, and two dimensions is much easier to conceptualize!

You’ll probably have noticed: diminishing marginal returns does not imply a perfectly flat plateau. Health is likely still slightly better toward the center. And the cliff is not completely vertical, it’s more like a sharp escarpment that you wouldn’t want to tumble down. But the broad principle remains: most of the benefit you can get is from not falling off, and little of the benefit is achieved by getting to the precise center.

The physiology of health plateaus

From a physiological perspective, health is achieved by all kinds of systems – immune, endocrine, metabolic, etc. – that detect our current state and make adjustments as needed to keep our physiology in the right range. These systems are heavily buffered, meaning that when something goes a little off target, there are very small consequences, and the rest of the system is able to step in and compensate. If you eat slightly more cholesterol today than yesterday, that shouldn’t make you keel over dead. Rather, cholesterol metabolism systems will simply adjust, and you won’t notice any difference. That’s how it should be!

This buffering is not just happening in one system, but simultaneously in many systems. The overall system is constantly adjusting in order to keep all the aspects stable, like a tightrope walker adjusting their muscles to maintain balance – an analogy we’ll explore more in a subsequent post. The end impact of this is that, across a relatively wide range of conditions, our physiology remains largely stable. It takes a lot to throw it off.

The evolution of health plateaus

From an evolutionary perspective, it’s easy to see why this is, and why it also applies at longer timescales. Our bodies are complex dynamic systems that have been shaped by natural selection. And natural selection of course needs to account for the unpredictability of our environment. Generally speaking, an animal that performs very well, but only under very specific conditions, will be outcompeted by one that performs a bit less well under the most common conditions but can tolerate a somewhat wider range.

There tend to be trade-offs between the range of conditions that can be tolerated and the performance under the best conditions. In other words, as tolerance to a range of conditions increases, the performance under perfect conditions decreases. In the mountain analogy, having a broader base means having a lower peak, because there’s only so much rock to be worked with. (Natural selection will long ago have done its best to increase the total amount of rock – in other words, to find ways to increase both the size of the base and the height together – so there’s usually very little scope for this to be further improved.)

In engineering, there’s a term for this: “highly optimized tolerance,” or HOT. HOT systems have evolved or been designed to be robust to a wide range of conditions, but as a consequence, they pay a price of sudden, catastrophic failure when they reach the limits of their tolerance.

One doesn’t descend the mountain gradually, one falls off the cliff, or tumbles down the rocky escarpment.

Evolution also gives us insight into where we might find the center of our plateau. It should be relatively close to whatever was normal for our ancestors. Natural selection works over many generations; we haven’t yet had time to adapt much to the industrial revolution, and probably only partially to the agricultural revolution. Our optima might be slightly different if our ancestors come from an unusual environment, such Inuit in the arctic or !Kung hunter-gatherers in the desert. But defining the precise center is likely far less important than understanding that you’re likely to be relatively healthy anywhere in the remote vicinity of the center.

Pulling it all together

We’ve seen why physiological buffering should result in relatively stable physiology across a wide range of conditions, roughly centered around some ancestral norm. We’ve seen why evolution will produce buffering and diminishing marginal returns. We can add in one more principle: the mathematical impossibility of indefinite linear growth.

If health increased linearly with effort, that would imply that there is no ceiling to health. Our daily experience strongly suggests otherwise. There’s probably a much bigger gap between the health of someone who is obese, sedentary, and depressed and your local gym rat, than between the gym rat and an Olympic athlete. And even the Olympic athlete may be optimizing their athletic performance at the expense of other dimensions of health (mental, long-term, etc.)!

However, the astute reader may have noticed some rhetorical sleight of hand. I’ve so far equated optimizing individual parameters of health (cholesterol, physical activity, vitamin consumption, etc.) with optimizing overall health. Isn’t it possible that full health only comes together when we get all the pieces right at the same time? Maybe we can get by when everything is not too far off, but to truly thrive we need to be right at the center of the plateau? Or maybe to stay healthy long-term, it helps a lot to be near the center?

There are a couple answers to this. First, natural selection isn’t working on each aspect of physiology one at a time, it’s working on the whole ensemble together. The pieces have to fit, so the principles of our health plateau should apply even as we look at all these dimensions together.

Second, centenarians appear to get to older ages in many ways, not just through living perfectly healthy lifestyles. Many have survived serious diseases. Jeanne Calment – the oldest verified person we know of – died at age 122, and smoked cigarettes for over a century.

Even with these two reasons, I think the jury is still somewhat out: there may be synergistic effects of getting a bunch of things right all at once. But even if that’s the case, it does not argue for herculean efforts to become the fittest person in town; rather, it would argue for getting most different aspects of health pretty close to right.