The inference rests on six observations, and all six are real. Individuals of a species go through the same changes in roughly the same order, on a timetable characteristic of that species and sharply different between species — a mouse gets three years, we get eighty. The hallmarks move together, across tissues and across systems. Single genes have outsized effects; daf-2 doubles a worm’s lifespan. Aging runs continuously out of development, with no seam between them. And some of it reverses, under parabiosis, reprogramming, restored youthful factors.

Damage does none of that. Damage is random, unsynchronised, and does not care what species it is in. So: aging is not damage; aging is program.

The premises are correct. The conclusion does not follow. A system coarse-grained near a critical point produces all six signatures with no controller, no schedule, and no script.

“Programmatic” is an emotionally loaded word, and it does a lot of covert work. When people hear it they picture a controller: something that holds a schedule, reads a clock, and writes to targets. That is not a vague intuition — it is a strong physical claim, and it requires three things to exist. A locus, where the schedule is kept. A timebase, so the schedule has a tempo. And a channel, carrying instructions from the locus to the targets. That is the good news.

An emergent account requires none of the three and reproduces the same six observations. So the evidence everyone cites does not discriminate between the two accounts; it buys a tie. Let me show how to settle the dispute

Subscribe now

Six mimics

In a recent preprint with Jan Gruber, we reduce aging to three macroscopic variables: a slow regulatory mode carrying resilience, a cumulative entropic damage variable, and a noise strength. Damage erodes the stability of the mode until it reaches a saddle–node bifurcation. That is the whole model. Now take the six observations in order.

Same changes, same order. The trajectory is reproducible because the slow manifold is one- or two-dimensional and its shape is fixed by the normal form of the bifurcation, not by biology. Reproducibility is a property of low-dimensional attractors — and low dimensionality is exactly what a large stochastic system hands you once one mode goes slow, outlives everything else, and enslaves the rest. Scripts are one way to get stereotypy. They are not the cheap way.

A species-specific deadline. The strongest intuition behind programmed aging is that something is counting. Our model has a sharp maximum lifespan set by the initial stability margin divided by the damage accumulation rate — no counter, no timer, no scheduled expression, just a bifurcation time. It looks designed because bifurcations are sharp. Where the species-specific number itself comes from is a better question, and it has a better answer than mimicry.

Hallmarks moving together. Fast modes are slaved to slow ones, so independent microscopic events project onto the same one or two collective coordinates and move in lockstep. This is why PCA on any aging dataset returns a handful of components — a result routinely read as evidence of a coordinator. It is evidence of separation of timescales, which is precisely what makes a coordinator unnecessary. The hallmarks are not conspiring. They are projections.

Master switches. This is the observation that converts people. In short-lived species the regulatory mode is unstable from birth, and lifespan is set by the inverse of the regulatory eigenvalue — everything else enters only weakly, logarithmically. A single mutation that nudges that eigenvalue toward zero therefore produces a divergent change in lifespan; as it crosses zero you get negligible senescence. daf-2 did not reveal a master regulator of aging. It moved an eigenvalue. Near a bifurcation everything is a master regulator, and effect size tells you nothing about whether you have found a controller.

No seam with development. Blagosklonny’s core image is aging as development that failed to stop. In our model development digs the potential well and aging is its erosion — one variable, one continuous trajectory, resilience peaking near sexual maturity where mortality is minimal. “Aging is the continuation of development” is literally true in the emergent account, with nothing running on.

Partial reversibility. Heterochronic parabiosis resets part of the methylation signature. That looks like switching a program off. It is a slow mode relaxing toward its fixed point. A before-and-after measurement cannot tell a relaxing mode from a flipped switch.

Six for six. I am not claiming a quasi-program is impossible. I am claiming that as of today it is unfalsified rather than supported, because every observation offered in its favour is equally well produced by a simple model with a few parameters, no controller.

One: programs cannot converge across the tree of life

Extrachromosomal rDNA circles drive aging in yeast and in essentially nothing else. Telomere attrition matters in humans and hardly at all in mice. The molecular substrate differs radically across taxa — and the phenomenology converges anyway: Gompertzian hazard, linearly growing biomarker variance, critical slowing down, late-life plateaus, and a hyperbolic divergence of physiological fluctuations toward a species-specific ceiling — the age at which resilience extrapolates to zero, which in humans falls around 120 to 150 years.

A program cannot explain that. Programs are arbitrary; there is no reason a yeast script and a primate script should share a mathematical form, and no mechanism by which they could. Universality classes are the only thing we know of that produce mechanism-independent convergence, and we have known this since Wilson: microscopically distinct systems flow to the same effective description under coarse-graining. The convergence of aging phenomenology across the tree of life is not a curiosity to be explained later. It is the primary datum, and it rules out mechanism-level determination of the phenomenology.

It is also why three hundred catalogued theories of aging each found supporting evidence. Each named a real contributor to the damage variable or a real perturbation of the regulatory mode. None changed the form of the equations. The field cannot converge by adding mechanisms, and it has spent decades trying.

Two: the link from development to lifespan is pure thermodynamics

The best cross-species card in the programmatic hand is that aging rate tracks developmental rate: fast-developing animals age fast, and something in ontogeny appears to set the clock. With Kirill Denisov and Jan Gruber we took the Kleiber–West allometry and energy conservation in West’s ontogenetic growth model, and added the one ingredient West’s model lacks — the second law. In a fully grown animal all metabolic output goes to maintenance: turnover of molecules, organelles, cells, tissues. No biosynthetic or repair process runs at perfect fidelity, so a fixed fraction of that turnover deposits irreversible configurational damage.

The damage accumulation rate is then simply the product of two things: the maintenance cost per unit mass, and a species-specific thermodynamic fidelity — the probability that any given turnover event leaves permanent damage behind. And the maintenance cost is not a free parameter. It is the same quantity that governs how quickly an animal finishes growing, so it falls inversely with development time. One number therefore sets both how fast an animal builds itself and how fast it accrues entropy, because turnover is simultaneously how you build and how you break. One parameter, two consequences. Maximum lifespan goes inversely with the damage rate, hence proportionally with development time — which is what the mammalian methylation data show.

That is the difference between the two accounts, in one example. The developmental correlation is something the hyperfunction picture interprets. It is something thermodynamics derives, from allometry and two conservation laws, with no script anywhere in the derivation. Derivation beats interpretation.

Three: a controller needs a channel. There isn’t one.

Mimicry arguments only ever buy a tie. This one is different, because there is a direct measurement.

A channel carries information. Whatever else “coordinated” means, it means the parts share information: a controller writes to many targets, and those targets end up mutually informative. Mutual information is exactly zero when two variables change independently. So this is not a matter of interpretation. You can compute it.

Three results, all pointing the same way.

The statistics are Poisson. In cross-species methylation data the dominant age-dependent component has both its mean and its variance growing linearly with age, with variance proportional to mean. That is the textbook signature of a sum of independent rare events. Programs are not Poisson.

The barriers are independent. Mapping site-specific rates of methylation change onto activation barriers gives barrier heights that are Gumbel-distributed — a type-I extreme value law. Extreme-value statistics arise when the underlying variables are independent or weakly correlated, and they mean the kinetics are set by the highest barriers. Aging is rate-limited by rare, high-energy, effectively simultaneous failures in highly redundant systems. Which is also why it is irreversible: undoing such a configuration requires a degree of microscopic control nobody has.

The sites carry no information about each other. With Kristina Perevoshchikova we went to single-cell methylation in aging mice — 698 cells, eight to twenty-four months — where mutual information can be computed rather than inferred. The methylation data split cleanly into two components. The exponential one, which tracks the Gompertz exponent and matches what the regression clocks are measuring, sits on sites with high pairwise mutual information. The linear one, which tracks global demethylation and grows in mean without growing in variance, sits on sites with the lowest mutual information in the dataset. Statistically independent, measured directly.

So the aging signature has already been split experimentally, and the two halves have opposite information content. And the half with no mutual information is the one that tracks cumulative damage and sets the maximum lifespan.

Program without program

The independent sites do not talk to each other. Zero mutual information means no site knows anything about any other site, and no channel connects them. But every one of them contributes to a single aggregate quantity — the total configurational load — and the regulatory network responds to that aggregate. That is a mean field. No component is coordinating with any other, and yet every pathway in the organism feels one common, slowly drifting quantity, because they are all immersed in it.

The coordinated, high-mutual-information component is the network’s response to that mean field. Which is to say: the coordination is real, it is measurable, it is enriched in developmental and signalling pathways — and it has no source. Nothing wrote it. It is the mean-field shadow of a process that is not coordinated at all.

That is what makes the hyperfunction reading so understandable and so inverted. It found a real object. A genuinely coordinated, genuinely developmental-looking, genuinely reversible signature exists, and Blagosklonny’s instincts pointed straight at it. But it is the readout, not the cause. The program is the shadow; the entropy is the object casting it.

Universality fixes the form, evolution fixes the numbers

None of this is exotic. Collective order with no coordinating cause is the ordinary situation in physics — a magnet magnetises without any spin instructing another — and what organises such systems is never a controller, only a constraint. Universality is a constraint of that kind. It does not generate the phenomenon; it restricts the forms the phenomenon is permitted to take, so that systems sharing nothing microscopically are forced onto the same macroscopic trajectory. The dynamics is independent of the hardware, which is why yeast and humans age along the same curve while breaking in entirely different places, and why a stereotyped trajectory licenses the inference something is constraining thisrather than something is running this.

That leaves a clean division of labour, and it is the one the dispute has been missing. Universality fixes the form of the aging trajectory, and biology has no vote in it. Evolution fixes the parameters, and biology has every vote. So the genome is no bystander: it sets the coefficients, and in our theory they come to two numbers — the maintenance cost per unit mass, fixed by the growth trajectory, and the thermodynamic fidelity, tuned against the price of accuracy, since proofreading polymerases are slower and more expensive and Medawar’s selection shadow decides how much accuracy is worth buying. That is the entire programmatic content of aging. Not a script that runs, but two constants in a thermodynamic identity.

One thing the quasi-program cannot supply for itself is a clock. “Constantly on” specifies no tempo. Something has to keep time, and the timebase everyone reaches for is the methylation clock — which is overwhelmingly dispersion, that is, entropy. Locus, timebase, channel: the channel is missing, and the timebase is borrowed from the variable the theory dismisses as a symptom.

Blagosklonny was right about his animals

Magalhães records that Blagosklonny, like him, argued that while molecular damage drives cancer, it does not drive aging. That sentence is unusually precise, and in our framework it is neither true nor false — it is regime-dependent.

Our model has two classes, and I have been leaning on the distinction already. In unstable animals — worms, flies, mice — the regulatory mode is already unstable at birth. The Gompertz exponent simply equals the intrinsic instability rate. Damage still accumulates, but it is subordinate: it modulates a mode that was diverging anyway, and mortality is governed by that divergence. In stable animals — humans, and long-lived species generally — the mode starts out stable, and there is nothing to drive aging except damage eroding that stability until the bifurcation is reached.

So in unstable animals, “damage does not drive aging” is almost exactly right, and our equations say so. Blagosklonny was not wrong. He was right about his animals.

Now take the census of the programmatic literature: C. elegans, Drosophila, mice, immortalised cell lines. Unstable or non-organismal, almost without exception. And these are the same animals in which every one of the six mimics runs at maximum amplitude: flat autocorrelation across life, because there is no resilience to erode; persistent effects from short treatments; and the divergent single-gene effects described above. The theory was induced from the one regime where it is true, using the one regime where a program-free model is least distinguishable from a program.

Which means the thirty-year argument is not really an argument about aging. It is an argument about which organism is the model organism, conducted by two groups who each believed they were describing aging in general. That is a far more tractable disagreement, and it comes with a specific warning: the thing that fails to transfer from mice to humans is not scale or lifespan. It is that in a mouse the dominant variable is the one you can move, and in a human it is not.

The experiment has been done

The programmatic and entropic accounts are both informational theories, with opposite sign. If the information is misapplied — a program running in the wrong context — you re-instruct it: reprogramming, signalling inhibitors, youthful factors. If part of the information is degraded — configurational entropy, independent sites drifting with nothing shared between them — there is nothing to re-instruct. You restore from a copy, or you replace the hardware. Those are different companies, different trial designs, different decades of work.

So the question that decides where the money goes is whether programmatic interventions touch the entropic term. Blagosklonny’s position requires that they do: if hyperfunctional signalling generates the damage, then suppressing the signalling should slow the accumulation.

That test is in the same preprint as the mutual-information analysis, linked below. We decomposed the mouse methylation signature into its dynamic and entropic components and asked what the two best-validated longevity interventions in the field actually move. Caloric restriction sharply reduces the growth rate of the dynamic component; on the slope of the entropic component it does nothing detectable, p = 0.34. Heterochronic parabiosis reduces the dynamic component, and the effect persists two months after detachment; the entropic component is unchanged immediately after the procedure and unchanged two months later.

So the most powerful longevity interventions we have act on exactly the component that carries mutual information, is enriched for developmental and signalling pathways, and is reversible — the quasi-program, correctly identified and now quantified. They leave untouched the component that accumulates independently, sets the ceiling, and cannot be re-instructed.

And this was run in mice, home turf for the programmatic view. If suppressed growth signalling slows entropy production anywhere, it should show up in the animal where caloric restriction produces its largest effects. It does not. What caloric restriction moves instead is the dynamic component and only the dynamic component, which is the textbook signature of a successful intervention in an unstable animal — and precisely the reason it will not transfer. The full measured benefit of caloric restriction in a mouse sits in the variable that dominates mouse lifespan and does not dominate ours.

A null is worth only what its power allows: this is one dataset, and the honest statement is that the effect on the entropic slope is below detection rather than proven absent. But it is the strongest available test, run on the strongest available interventions, in the organism most favourable to the hypothesis, and it came out on one side.

The holy war can therefore be retired — not because one camp won, but because the question was badly posed. There is a program-like object in aging: coordinated, developmental in character, reversible, and now measurable. Blagosklonny was right that it exists and right about where to look for it. There is also an entropic object: independent, carrying no information, irreversible, and it is the one that sets the ceiling. The argument ran for thirty years because each side had hold of a different object, in a different animal, and neither had the coordinates to say so.

What is left is a better question than the one it replaces. It is no longer whether aging is programmed. It is whether anything moves the entropic term — whether thermodynamic fidelity is pharmacologically accessible at all. Magalhães sets essentially the same condition from the other end: reduce molecular damage without touching developmental signalling, and see whether broad multi-organ extension follows. Nobody has done that experiment. That is the one worth funding, and I would rather the field spent the next decade on it than on another round of this argument.

* John Archibald Wheeler — Feynman's doctoral advisor at Princeton, and the man who named the black hole and the wormhole — had a habit of naming things by what they lacked: mass without mass, charge without charge, law without law. Each was a case where some familiar quantity turned out to be produced by something that did not contain it. The last is the closest to the case here. Wheeler's suggestion was that regularity itself might emerge from underlying randomness, so that what looks like a law requires no law behind it. Aging is program without program.

Share

Refs:

Magalhães’s review: A Brief History of the Hyperfunction Theory of Aging and Future Directions, Aging, July 2026.

A Minimal Model Explains Aging Regimes and Guides Intervention Strategies;

Discovery of Thermodynamic Control Variables that Independently Regulate Healthspan and Maximum Lifespan;

Differential Responses of Dynamic and Entropic Aging Factors to Longevity Interventions.

The confusion is the word theory. The sixteen objects on Alan’s landscape are not sixteen answers to one question. They are answers to at least three different questions — why aging exists, what aging is, and how aging works — and these questions live at different levels of description. Plotting them on shared axes is like plotting the Landau-Ginzburg theory of superconductivity, the BCS mechanism, and the engineering specifications of a maglev train on the same chart and asking which one is “most correct.” They are not competitors. They are different kinds of objects.

I come from physics, so let me lay out the ontology physics uses for complex systems, and then show what happens when you sort the aging field into it.

Subscribe now

Two kinds of theory, and no 1-to-1 map between them

For a complex system, a theory can be one of at least two types.

A phenomenological theory is a minimal set of variables and their relations that compresses a large corpus of observations. It classifies and explains without attempting microscopic reduction. Landau-Ginzburg theory of superconductivity is the canonical example: an order parameter, a free energy, a handful of coefficients — and an enormous range of phenomena falls out. Such theories are compressed representations of what happens.

A microscopic theory explains how it happens mechanistically — and, crucially, it earns that status only by being shown to recover the phenomenology in the macroscopic limit. BCS theory is not the theory of superconductivity because it involves electrons and phonons; it is the theory because Gor’kov demonstrated that Landau-Ginzburg emerges from it. The derivation is the price of admission.

The relationship between the two levels is many-to-one. Multiple microscopic theories — non-universal, hardware-dependent — can produce the same macroscopic phenomenology. Whenever you observe universality across systems with different microscopic composition, you are usually looking at emergence: the macroscopic behavior has stopped depending on microscopic details.

Above both sits a third level, which I’ll call the application level. To build a technology out of some embodiment of a system, you rarely need the microscopic theory. You use the phenomenology and you do not care how it is implemented.

Sorting the landscape

Run Alan’s sixteen theories through this filter and the picture reorganizes itself.

The phenomenological theories of aging are the ones with equations: Gompertz mortality kinetics, Strehler-Mildvan, Uri Alon’s saturating removal (yes, it’s a phenomenological effective theory, though the authors confront such classification!), and our own resilience-loss (actually a lot more complete recent work with Jan Gruber) models at Gero. Few effective variables, explicit dynamics, compression of mortality curves and biomarker trajectories. These are the Landau-Ginzburg candidates.

The would-be microscopic theories are free radicals, telomere attrition, somatic mutations, sirtuin relocalization. Each proposes hardware. And here is the field’s central missing step, visible only once you adopt this ontology: no microscopic theory of aging has ever been shown to recover the phenomenology in the macroscopic limit. Nobody has derived Gompertz kinetics, mortality plateaus, or the temporal scaling of biomarker variance from telomere dynamics or free radical chemistry. In condensed matter, that derivation is what makes a mechanism a theory. In aging biology, microscopic proposals are judged instead by whether perturbing the favorite variable moves lifespan in mice — a far weaker test, and one the Interventions Testing Program keeps failing them on, as Alan’s own conclusion documents.

The question, however, is no longer open-ended. In recent work with Jan Gruber, we showed that the macroscopic laws follow from generic structural assumptions about the microscopic level — no particular hardware required. Rare, statistically independent, thermally activated configurational transitions accumulate as a Poisson process, which is why damage grows linearly in mean and variance; their aggregate weakly erodes the stability margin of the slowest physiological mode; and Kramers escape over the shrinking barrier converts linear damage accumulation into exponentially rising mortality — the Gompertz law, derived rather than fitted. This is precisely the recovery-of-phenomenology test that free radicals, telomeres, and mutations never passed. What the derivation supplies is the set of matching conditions any candidate mechanism must satisfy to count as a microscopic theory of aging: produce independent Poissonian events at the measured damage accumulation rate, exhibit the extreme-value barrier statistics seen in methylation data, and couple into the slowest mode with the right strength. The analogue of Gor’kov’s derivation now exists as a template — what remains is for some piece of molecular hardware to be shown to fit it. And the many-to-one mapping predicts there will never be the microscopic theory, only species-specific hardware realizations of one universality class: extrachromosomal DNA circles drive aging in yeast but almost nowhere else, yet the coarse-grained laws are the same.

The Hallmarks of Aging, in this light, is the interesting failure case: it is neither kind of theory. No dynamics, no compression, no recovered macroscopic law. It is a parts inventory — which is exactly why it feels simultaneously comprehensive and inert.

And the evolutionary theories — Medawar, Williams, Kirkwood — belong at the application level, which is why they have aged so gracefully despite predating molecular biology. Selection is an engineer. It does not need to know how aging is implemented in a given species; it operates on the phenomenology, tuning effective parameters like damage production and repair investment under resource constraints. Disposable soma is a theorem about any mortality-constrained replicator. That is also why evolutionary theories are not rivals to mechanistic ones and never were: they answer why, at a level of description where how has been integrated out. (One honest caveat: unlike a maglev engineer, evolution also writes the hardware. Antagonistic pleiotropy is precisely a claim about how selection sculpted the microscopic couplings. Evolutionary theories do double duty across levels — which is, I suspect, why people keep miscategorizing them as competitors to mechanism.)

Notice what this does to Alan’s axes. The “one cause vs. many causes” axis dissolves entirely: cardinality of causes is only meaningful at the microscopic level. An effective single-variable macroscopic theory emerging from many microscopic contributors is not a “many causes” theory — the many causes are exactly what got integrated out. That our own model landed in the “many causes” quadrant is, I’d argue, the clearest symptom of the confusion.

The many-to-one mapping carries a practical warning too: longevity adaptations need not be universal across species or across regimes of aging. Cross-species regularity in biology has two possible sources — true emergence, or conserved hardware. Gompertz kinetics holding across phyla with wildly different physiology looks like emergence. Rapamycin working from yeast to mammals looks like a shared, inherited component. The two have opposite implications for whether an intervention will translate from mice to humans — which is the question the field spends billions on while rarely stating it this way.

Where entropy actually enters

Alan labels one axis “entropy vs. evolution,” and the standard objection writes itself: organisms are open systems, they export entropy freely, thermodynamics mandates nothing. The objection is correct against naive wear-and-tear. It fails against the version physics actually supports.

Living systems are far from equilibrium, but they operate over times enormously long compared to molecular scales. Timescale separation licenses a quasi-equilibrium description: fast variables equilibrate conditional on slow ones, and the organism drifts along a slow manifold. Aging is that drift. And the entropy that matters is not the total being dissipated to the environment — according to our proposal, it is configurational entropy accumulating in slow degrees of freedom that the repair machinery does not reset: epigenetic states, cross-links, mosaicism, the frozen residue of a lifetime of insults. In our models, this configurational entropy is what “damage” means. Good phenomenology is thermodynamic not by coincidence but by construction: coarse-graining over a timescale gap produces Langevin dynamics in an effective potential, which is precisely the drift-plus-noise form aging trajectories obey.

The nearest physical analogue is a structural glass. Glasses age too — slow compaction through a rugged configurational landscape, relaxation times that grow with the sample’s own age, memory and rejuvenation effects under perturbation. Nobody asks which molecule causes glass aging; the question is recognized as category-confused, because configurational entropy is a property of the landscape, not of any molecular species. Carry that recognition into biology and the “which hallmark is upstream” debate dissolves the same way: many microstates, one macroscopic variable, damage fungible across molecular forms.

Two consequences deserve to be treated as predictions rather than philosophy. First, irreversibility gets a principled basis: restoring a low-entropy configuration requires knowing which microstate to restore — an information cost of the Landauer type. Rejuvenation is therefore possible exactly where a template survives. The germline reset works because development re-specifies the macrostate from stored information; partial reprogramming works to the extent epigenetic information is recoverable rather than erased. This converts the “backup copy” of the information theory of aging from metaphor into a sharp empirical question: how much specifying information survives, and in which degrees of freedom. Second, entropic drift predicts rising heterogeneity, not merely rising damage: growing cell-to-cell transcriptional variance, loss of mutual information between tissues, diffusive spreading of trajectories in state space. These are measurable, partially observed already, and they discriminate: a run-on developmental program would produce coordinated late-life change, not diffusive spreading.

One obligation comes with this. Entropy is defined relative to a coarse-graining, and without a declared partition “configurational entropy” degenerates into a synonym for disorder — the biologist’s complaint about unfalsifiable physics-talk is then justified. Biology, unusually, offers a natural partition: the reference macrostate is the set of states the developmental program can specify and the error-correction machinery defends. Entropy is deviation from the specifiable set. The measure is grounded in the organism’s own information architecture, not an observer’s taste.

What the map should look like

So, with respect, here is my amendment to Alan’s landscape. Replace the axes. One axis: which question is being answered — why aging exists, what aging is, how it works. The other: the level of description — phenomenological, microscopic, evolutionary constraint. On that map the theories stop competing and start composing: evolutionary theories set the boundary conditions, phenomenological theories define the order parameters and dynamics, and microscopic proposals earn their place the day one of them recovers the phenomenology in the macroscopic limit.

Even the perennial fight over the definition of aging becomes tractable: it is an order-parameter selection problem. Frailty index, epigenetic age, mortality doubling time — these are competing choices of macroscopic variable, and in Landau theory choosing the order parameter is the central creative act, not a semantic quibble.

Alan closes by wondering whether there is any correct theory of aging, or only a mess too complex for mortal minds. I’d offer a more optimistic reading: the mess is not in the biology. It is in the ontology. Physics spent a century learning how to relate phenomenology, mechanism, and application in complex systems without confusing the levels. Aging research does not need to relearn that from scratch — it needs to borrow it.

Share Peter’s Substack

Argument 1 — Mutational stress-test

Yuri: The cloning bottleneck (one cell → trillions → one cell, repeated 57x) is a far harsher mutational stress test than any natural lifetime, yet lifespan stayed normal for decades of generations — implying somatic mutation load isn’t the real limit on lifespan, weakening the claim that unrepaired mutations cap rejuvenation.

My answer: Let me narrow the claim to the part that actually matters here. The cloning counter-argument I was originally responding to isn’t just “mutations accumulate slowly” — it’s the stronger claim, often invoked via the “information theory of aging,” that a full epigenetic reset returns biological age to zero with no irreversible leftover. Serial cloning is about as rigorous a test of that exact claim as has ever been run: every single generation gets a complete nuclear reprogramming event, repeated 57 times, across the entire cell population, with full embryogenesis and whatever developmental selection that entails. And the reset was still not, by itself, sufficient to drive the damage to zero — it accumulated anyway, monotonically, until it became lethal at generation 58. One clean counterexample is enough to falsify a universal claim like “zero residual damage,” regardless of how many generations the lineage tolerated before that, or whether a screened lineage might have gone further. So I’m not claiming mutation load is the dominant constraint on mouse lifespan, or that 58 is some fixed biological number — only that even the most powerful reset conceivable left a residue it couldn’t erase.

That residue isn’t an isolated case. Put it next to the other interventions I cited: dose-ramped in vivo OSKM plateaued well short of top-tier longevity interventions and increased cancer risk past a point; caloric restriction and heterochronic parabiosis both moved the reversible aging signature but left the entropic one completely flat. Four structurally different interventions — full genetic reset, transcription-factor overexpression, dietary, and systemic/circulatory — each ran into some floor they couldn’t push through. That convergence across unrelated intervention types is what gives the damage a thermodynamic flavor, not any single experiment on its own. I’d put it as “consistent with irreversible accumulation” rather than claim it as proof of a specific physical mechanism — but consistent across four independent lines of evidence is more than coincidence.

Subscribe now

Argument 2 — Not a fundamental ceiling

Yuri: The gen-58 collapse came from one unscreened lineage hitting two specific lesions (Chr4 LOH, then a Chr4–11 translocation). Sexual reproduction’s built-in QC (recombination, gamete competition, embryo selection) would likely let screened cloning go much further.

My answer: There’s a real, observed ceiling here regardless of cause: every gen-58 pup died within a day. But I’d push back harder on the premise that this lineage had “none of the quality control built into sexual reproduction.” It actually had a brutal selection step built in by default: cloning efficiency itself. Success rates climbed to a peak of 15.5% around generation 26, then fell progressively to ~0.6% by generations 57–58. That decline is selection in action — every generation, the overwhelming majority of cloned conceptuses failed to develop or survive to birth, and only the ones that made it all the way to a live, viable adult got to become the next round’s nucleus donor. That’s arguably a harsher filter than PGT-style embryo screening, because it selects on whether the whole organism actually works, not just on detectable genomic lesions. So this wasn’t an unscreened lineage drifting passively to failure — it was a heavily, increasingly selected lineage, and the selection pressure still wasn’t enough to hold the line. Mutational burden kept climbing in the survivors anyway, success rates kept collapsing anyway, and the line still died out. That’s evidence against your fix working, not just an open question: nature was already running something like the screen you’re proposing, generation after generation, and it bought time but didn’t prevent the collapse.

There’s also a more basic point that holds regardless of how the screening question resolves: the QC you’re describing — discarding embryos, gametes, or whole organisms that don’t make the cut — is selection, not reprogramming. The reprogramming step itself never touched those lesions; selection only ever decided which already-formed individuals got to continue. Rejuvenating an existing adult organism doesn’t get to “discard the bad copies and start over”; you have to fix the cells you have, in place, which is a fundamentally harder problem than what screened cloning would be solving even in the best case.

Argument 3 — Thermodynamics

Yuri: Organisms are open systems that spend energy on repair. The second law caps energetic cost, not feasibility. Reversibility is an empirical/engineering question, not something entropy forecloses.

My answer: “Open systems can do work to repair” is trivially true and doesn’t by itself rescue the argument — but that was never really the crux. The open/closed distinction is a red herring: there’s no general theorem that entropy must increase in open systems, and none that forbids it in closed ones with fully reversible microscopic dynamics. What actually generates entropy increase is loss of control or observability over the relevant degrees of freedom — coarse-graining, hidden states, anything limiting the precision with which you can track and correct every microstate.

The cleanest illustration is GW250114 [4], and it’s worth being precise about what’s empirical there versus what’s theoretical, because that’s where the strength of the example actually lies. What LIGO measured is real: the masses and spins of two merging black holes, inferred from a real gravitational-wave signal, and from those the horizon areas before and after merger. The area did not decrease — confirmed at >3σ. That’s an empirical fact about an actual cosmic event, not a derivation from theory. The further step — identifying horizon area with entropy (Bekenstein-Hawking) — is theoretical, if extremely well-supported. But I don’t even need that step to make the point. The bare, measured fact is enough: a system governed by fully known, fully reversible, deterministic classical dynamics — no statistical ignorance about microstates anywhere in the calculation — still produces an irreversible, one-directional change in a macroscopic quantity. The reason is structural, not statistical: once part of the system crosses the horizon, it becomes causally inaccessible to any outside observer, and that loss of access alone is sufficient to generate an arrow of time in the observable sector. Whether the system is “open” or “closed” never enters into it.

That’s the more general point — closer to Wolfram’s framing, and to a real, recognized thread in the foundations of statistical mechanics going back to the Gibbs/Boltzmann coarse-graining debates — that an arrow of time emerges generically whenever an observer or agent lacks full access to or control over a system’s degrees of freedom. I’d rather define entropy this way — operationally, as whatever quantity exhibits that one-directional, irreversible signature under the relevant dynamics — than commit to a specific formula or to the open/closed framing at all. That sidesteps a debate I don’t need to win (whether entropy is fundamentally observer-relative or not) and keeps the claim to what’s actually defensible: the presence of an arrow of time is the property that matters, and it doesn’t track openness.

This is also exactly how we identify entropy in the aging data itself [1,2,3] — not by assuming a formula, but by the empirical signature: linearly increasing mean and variance (Poisson statistics), Gumbel-distributed activation barriers, vanishing recovery rates. Those are the biological equivalent of “the horizon area went up” — a measured, one-directional signature, not an assumption. An organism is open and spends real energy on repair, nobody disputes that — but repair only reverses drift to the extent the repair machinery has effective control over the actual degrees of freedom that are drifting, and it doesn’t, anywhere close. DNA repair, proteostasis, and tissue turnover are coarse and blind to most of the configuration space, for the same structural reason an outside observer can’t see inside a horizon: not because the system is closed, but because control over its relevant degrees of freedom is incomplete.

This isn’t just theoretical on the biology side either. We tested it directly [3]: heterochronic parabiosis — surgically joining an old mouse’s circulation to a young one’s, about as drastic and energetically active an intervention as you can run on a living organism — measurably reduced the reversible/dynamic signature (dFI) during the procedure and for two months after. But it left the entropic signature (tBA) completely untouched, both during the procedure and afterward. If “spend enough energy and you can reverse it” were the operative principle, parabiosis is close to a best-case test: continuous exposure to a young systemic environment, full metabolic and regenerative machinery engaged, sustained over time. It still didn’t move the entropic component at all. That’s a real, measured instance of exactly the gap I’m describing — control over degrees of freedom, not energy availability, is the bottleneck — and it’s evidence, not metaphor.

So “organisms are open” answers a question I’m not actually asking. The real question, same as for the merging black holes, is how much of the relevant configuration space is under active, high-fidelity control — and empirically, not much: the fidelity factor c [1] and the entropic signature that didn’t budge under caloric restriction or parabiosis [3] are direct measurements of that control gap, not a thermodynamic slogan.

Subscribe now

Argument 4 — Unfair standard

Yuri: Judging partial reprogramming on first-gen, non-specific OSKM is a strawman; the field has moved to cell/tissue-specific approaches, with real clinical progress (Life Bio, NewLimit, YouthBio).

My answer: Also fair, and consistent with what I actually said in the original piece — I explicitly conceded that epigenetic interventions “may achieve moderate benefits — comparable to today’s best anti-aging technologies.” I have zero doubt cell- and tissue-specific reprogramming will produce real therapies against real diseases — Alzheimer’s, liver disease, whatever YouthBio, NewLimit, and Life Bio are targeting. My claim was narrower than “partial reprogramming is a failed idea”: it’s that the anti-aging effect specifically will be capped by the need to preserve tissue identity, separately from how good the therapy gets at its target indication. Clinical-stage progress shows the field is maturing operationally; it doesn’t yet show movement on the actual mechanistic question. Here’s the experiment that would settle it: run the same dFI/tBA split we used on CR and parabiosis [3] on a reprogramming dataset, and see whether it behaves like those interventions (moves the reversible component, leaves the entropic one flat) or actually shifts the damage-accumulation rate itself. That hasn’t been done yet for any reprogramming protocol, as far as I know — that’s the data that decides this, not company pipelines.

References

[1] Denisov, K.A., Gruber, J., Fedichev, P.O. “Discovery of Thermodynamic Control Variables that Independently Regulate Healthspan and Maximum Lifespan.” bioRxiv 2024.12.01.626230. https://doi.org/10.1101/2024.12.01.626230

[2] Tarkhov, A.E., Denisov, K.A., Fedichev, P.O. “Aging clocks, entropy, and the limits of age-reversal.” bioRxiv 2022.02.06.479300. https://doi.org/10.1101/2022.02.06.479300

[3] Perevoshchikova, K., Fedichev, P.O. “Differential Responses of Dynamic and Entropic Aging Factors to Longevity Interventions.” bioRxiv 2024.02.25.581928. https://doi.org/10.1101/2024.02.25.581928

[4] Fedichev, P. X post on entropy, control, and GW250114.

[Original] Fedichev, P. “Why Epigenetic Rejuvenation Won’t Give Us Radical Life Extension: Lessons from Serial Mouse Cloning,” Substack, Jun 16, 2026. https://peterfedichev.substack.com/p/why-epigenetic-rejuvenation-wont?r=4xkkp8

[Original X] Fedichev, P. X post linking to the original Substack piece.

[Reply] Deigin, Y. Reply to the original post on X.

Subscribe now

You may also know my arguments against over-optimism here. We at Gero believe that in long-lived species like humans, most aging is not fully reversible. Epigenetic interventions may achieve moderate benefits—comparable to today’s best anti-aging technologies—but they won’t deliver the dramatic, near-unlimited extension many hope for. Damage accumulates in ways that partial resets can’t fully erase without compromising core identity and function.

Subscribe now

Evidence from Yamanaka Factors In Vivo

A few key experiments strongly support this view. First, the important work from Prof. Juan Carlos Izpisúa Belmonte and colleagues. They used a genetic system to inducibly overexpress the OSKM (Oct4, Sox2, Klf4, c-Myc) factors in living mice in a dose- and frequency-dependent manner. As the “dose” of reprogramming increased, lifespan benefits rose—but then plateaued. Critically, the maximum effect was no greater than that seen with other top-tier lifespan interventions in the same models.

Think about that: If the reversible epigenetic signature were the primary driver of aging, a potent OSKM intervention should have outperformed interventions against secondary hallmarks. It didn’t. Instead, pushing harder simply ramped up cancer risk—a pattern consistent with excessive loss of cell identity through strong epigenetic perturbation. The rejuvenation effect maxes out well before it becomes transformative. This is highly instructive.

Lessons from “Immortal” Simple Animals

Still, many point to simpler organisms such as Hydras or jellyfish (like Turritopsis dohrnii) that show negligible senescence and can reform whole organisms from dissociated cells. Some can revert from adult medusa back to polyp stage under stress, cycling through life stages.

These are fascinating examples of biology capable of near-total cellular reset and organismal rebuilding. But here’s the catch: in these processes, cellular identity is largely lost. No preserved brain, no heart, no memories, no continuity of self. For complex organisms like us, this isn’t rejuvenation—it’s closer to full body replacement without backup. It’s not meaningfully better than death if what we value is our identity and experience.

Hence, our position: If you want to preserve identity, it’s far better to stop aging than to try reversing it. Strong rejuvenation without dismantling the system seems impossible.

The Cloning Counter-Argument—and Why It Fails

I keep getting pushback: references to earlier cloning work claiming no lifespan shortening in clones, suggesting that somatic cell nuclear transfer (SCNT) fully resets age to zero with no irreversible leftovers. This idea features in popular books invoking “information theory of aging.” If I (and our models) are right, there must be irreducible costs to cloning. And now, there is clear evidence.

After 20 years of painstaking work, Teruhiko and Sayaka Wakayama’s team produced over 1,200 cloned mice across 58 generations from a single original donor female.Early success rates improved (peaking at 15.5% around generation 26), and mice lived normally with typical ~2-year lifespans. No major epigenetic accumulation was evident up to ~G25. But from generation 27 onward, birth rates declined progressively. By generations 57–58, success fell to ~0.6%. All G58 pups were born but died the day after birth. The line ended.

Crucially, epigenetic reprogramming could not revert the accumulating genetic mutations. Cloning induced ~3x higher mutation rate per generation versus natural reproduction—large structural variants built up, eventually becoming lethal.

This closes the case for me. Mice are fast-agers that die well before damage becomes fully incompatible with survival—allowing an impressive 58 generations. In longer-lived species (where aging involves more accumulated damage), we expect far fewer viable serial generations.

This is why I dislike when some proponents of epigenetic rejuvenation prefer to frame aging in terms of “information loss” rather than entropy. They argue that reprogramming can restore lost information, and that this framing somehow escapes the constraints of the second law of thermodynamics.

I have always found this switch suspicious.

Entropy and information are intimately related (in fact, information entropy is defined with the opposite sign to thermodynamic entropy). Talking about “information” instead of “entropy” does not magically exempt biological systems from the second law. It merely makes the discussion sound more palatable — less “traumatizing” for those who instinctively recoil from the idea that irreversible processes are at work.

In reality, the accumulated microscopic errors — whether we label them entropic damage or informational degradation — become harder to reverse without destroying the very structure that carries the identity.

When people say “we are only losing information and we can reset it,” they are usually implying that the information is stored in a clean, reversible epigenetic layer. The serial cloning experiment shows this hope is illusory. Even with full nuclear reprogramming (the strongest possible epigenetic reset), the system still accumulated irreversible genetic damage at an accelerated rate. The information loss was not confined to the epigenome — it spilled over into the genome itself, and no amount of “information-theoretic” language changed that physical outcome.

Calling it “information” does not make the second law negotiable. It only creates an illusion of reversibility. If we want to preserve the identity and complexity of a mammal, the pragmatic path is not to fight entropy by trying to run the system backwards, but to slow or stop the forward accumulation of damage in the first place.

The study elegantly demonstrates, once again, why you don’t casually mess with the second law. There are many more fun things in life.

Read the paper, enjoy watching the science at work https://www.nature.com/articles/s41467-026-69765-7

Subscribe now

Take a worm. A C. elegans nematode, about a millimeter long, living out its 20-odd days in a laboratory dish. Wait until 75% of its population has already died. The survivors are, by any reasonable definition, geriatric — equivalent to a human centenarian on borrowed time. Now give them a drug that degrades a single protein, the insulin receptor DAF-2.

Their remaining lifespan doubles.

Not extends by 10%. Not adds a few days. Doubles. Animals that had four days left suddenly had 26. Their total lifespan matched what you’d get if you’d given them the same drug from birth.

This finding, from the labs of Collin Ewald at ETH Zürich and Jan Gruber’s group in Singapore, has now been reproduced by multiple independent laboratories. It is real. And it is, if you think about it carefully, deeply strange — strange in a way that most of the longevity field has not yet reckoned with. Jan and Collin came to me with this puzzle. Together we worked out what we believe is the physical explanation.

The Damage Problem

The dominant story of aging goes something like this: damage accumulates. Proteins misfold. DNA mutates. Mitochondria deteriorate. Cells go senescent. Each insult leaves a scar. The older you are, the more scars you have. And those scars are the reason you age.

If this story is basically right — and it has enormous amounts of evidence behind it — then late-life interventions should work less well than early ones. By the time you’re 75% of the way through your lifespan, you’ve accumulated three-quarters of your lifetime damage. Any intervention now is fighting uphill against a mountain of prior insult.

Which is why the worm experiment is such a puzzle. These geriatric animals — the ones with pharyngeal degeneration, gonadal atrophy, uterine tumors, and pooled yolk — still doubled their lifespan when their insulin receptor was degraded. The damage was all still there. The senescent pathologies persisted. Yet the animals kept living.

What is going on?

The Worm Doesn’t Know What Day It Is

In a new preprint posted this week — a joint effort from Gero’s, Jan Gruber’s, and Collin Ewald’s labs, with the theoretical framework that I developed together with Jan — we offer a precise physical explanation.

The key idea is this: the worm doesn’t know its chronological age. What determines its mortality risk is not how many days it has been alive, but where it sits on a dynamical instability trajectory.

Think of it this way. Imagine a ball on a landscape that slopes gently upward at first, then tilts more and more steeply, until finally it goes over the edge. The worm is the ball. Death happens when the ball goes over the edge. The question is not when the ball started rolling — it’s how close to the edge it currently is.

In the language of physics, we describe this with a Langevin equation — a stochastic differential equation that captures both the deterministic drift toward instability and the random fluctuations that push the ball around. The model has a few key parameters: the instability rate α (how fast the ball accelerates toward the edge), a nonlinear feedback term that accelerates collapse near the end, and a noise term representing physiological fluctuation.

What makes this framework powerful is what it predicts about interventions. Because the system is Markovian — meaning its future depends only on its current state, not its history — changing the instability parameters at any point in the trajectory immediately changes the expected time to death. No damage reversal required. The ball doesn’t need to move backward. It just needs to roll more slowly.

This is exactly what DAF-2 degradation appears to do: shift the system’s stability parameters without resetting accumulated structural damage. The pharyngeal degeneration is still there. The tumors persist. But the ball is rolling more slowly. And so the animal lives longer.

Two Kinds of Aging

Here is the deeper implication, and the one I find most interesting.

Not all organisms age the same way.

In C. elegans, aging appears to be dominated by this dynamical instability — an intrinsically unstable physiological mode that amplifies perturbations and drives the system toward failure. It is essentially programmatic: the worm is designed to deteriorate, because once reproduction is complete, there is no evolutionary pressure to maintain systemic stability. In fact, there may be active evolutionary pressure toward post-reproductive death — the worm’s body converts itself into yolk for its offspring. Death, in this sense, is part of the life history strategy.

Humans are different. We reproduce over decades. We have evolved extensive stability mechanisms to keep the body coherent over long time spans. Our aging is dominated not by dynamic instability in the worm’s sense, but by the slow accumulation of entropic damage — irreversible thermodynamic insults that gradually erode physiological function.

This has a concrete clinical prediction: the interventions that work dramatically in C. elegans — targeting the dynamical instability parameters — will have only transient effects in humans. Our resilience mechanisms will restore equilibrium after the perturbation is removed. We’ve seen this pattern in drug after drug that works beautifully in worms and mice and then disappoints in humans.

For humans, the targets are different. They are the rate of damage accumulation. And the effective temperature — the amplitude of physiological fluctuations that determines how far average lifespan falls short of maximum lifespan. These are the two control variables my group has been focused on, and which I’ve written about here before.

What the Worm Can Still Teach Us

None of this means that C. elegans aging research is irrelevant to human medicine. Far from it.

The DAF-2 experiment teaches something fundamental: the separation between reversible and irreversible damage is real, and it is experimentally accessible. Some pathologies in the geriatric worm — protein aggregates, cuticle deterioration — were reversed by the intervention. Others — gonadal atrophy, pharyngeal degeneration — were not. That pattern is a window into the architecture of aging: some components are enslaved to the dynamical mode and recover when that mode is stabilized; others reflect truly irreversible structural damage that accumulates independently.

Understanding which is which — and developing the tools to measure it — is essential for human medicine. We need to know what fraction of human aging is, in principle, reversible. The worm experiment suggests the answer is not zero. But it also suggests the answer is not everything.

This is, I think, the honest state of the field. The dream of complete biological rejuvenation — the 90-year-old restored to 30 — collides with the second law of thermodynamics. But the gap between average lifespan and maximum lifespan, the roughly 40 years that separate typical human death from the biological ceiling at around 120, is a tractable target. That gap is noise. It is dynamical instability in the generalized sense. And if the worm experiment teaches us anything, it’s that you don’t need to reverse all damage to meaningfully extend survival. You just need to change the dynamics.

The Physics Is the Point

I want to say something directly. I came to aging biology as a theoretical physicist, and I have spent the last decade building mathematical frameworks for what aging actually is — not just cataloguing what happens during it.

The longevity industry has been built largely by biologists. That is mostly good — biology is complicated and biologists are necessary. But there is a cost: the field has accumulated enormous amounts of phenomenology — observations of what happens during aging — without a commensurate investment in understanding why those things happen, and crucially, what the mathematical structure of aging actually is.

The Langevin instability framework is one answer to that question. It says: here is a precise, testable, quantitative model. Here are its parameters. Here is what happens when you change them. Here is why you observe what you observe in the data.

This is different from saying “senescent cells accumulate, therefore remove senescent cells.” That’s a mechanism, not a model. A model makes quantitative predictions. It tells you what you’ll see before you run the experiment. And it tells you where the intervention leverage actually is — which turns out to be not always where the biology looks most interesting.

The worm that doubles its remaining lifespan near death is, in this sense, not an anomaly. It is a controlled experiment on a dynamical system. And the result is exactly what our model predicts. I’ll admit — there are few better feelings in science than that.

The preprint “Old worms, new tricks: dynamical instability explains late-life rejuvenation in C. elegans” by Latumalea, Molière, Fedichev, Ewald, and Gruber is available on bioRxiv. The instability framework underlying this work was developed in Fedichev & Gruber (2025).

I’m a physicist by training. I got pulled into aging biology the way most people get pulled into obsessions — by a single, unsettling fact.

In 2006, a paper came out showing that the naked mole rat doesn’t age. Not “ages slowly.” Doesn’t age. Its probability of dying doesn’t increase with time. That paper wasn’t funded by NIH. It came from a daughter company of Google — because you apparently need to build a $10 billion revenue business first in order to spin off a $200,000 experiment to check whether a small burrowing mammal from Somalia experiences aging. That, to me, says everything about why this field is still where it is.

But things are changing. And I want to tell you exactly how — and why the change is more profound than most people in the industry are willing to admit.

The $500 Billion Data Point

A few years ago, if you said that a drug extending human life by one to two years could generate $100 billion in market capitalization, you’d have been dismissed as a dreamer. Sam Altman said something like that in 2018, and it was treated as an interesting thought experiment.

Then GLP-1 agonists happened.

You know this drug — it’s now selling at over $50 billion per year, with oral versions forecast to cross $100 billion. One company is on the verge of becoming the first trillion-dollar pharma firm. The original story was weight loss. But the real story emerging from multiple clinical trials is something else: these drugs appear to reduce the risk of other age-related diseases — cardiovascular, neurological, metabolic — by somewhere around 30% in certain cases. When you translate that into rescued years of life, you’re looking at roughly one to two years of additional lifespan.

So here’s the data point we now have: a drug that probably extended human life by about a year saw its issuing company’s market cap increase by approximately half a trillion dollars. The order of magnitude that was predicted was correct.

What does this mean for the industry? Everything.

If your drug is selling for $50 billion a year, you’re not going to pivot to another cancer drug with a $5 billion ceiling. The logic of commercial scale is now pointing in one direction only: the next drug that can be bigger than GLP-1 is a true longevity drug. Everything else is smaller. Nobody wants to develop a drug against aging — aging isn’t a disease, regulators don’t know what to do with it, investors get nervous. But the mathematics of market size will force the industry there anyway. That’s not ideology. That’s arithmetic.

We’re already seeing the early signs. Nearly every major pharmaceutical company now has a longevity project running quietly in the background. Alphabet has acquired aging assets. We at Gero just completed a deal with Chugai, the daughter company of Roche. This is a trend that will accelerate.

What Aging Actually Is (And Why You’re Probably Thinking About It Wrong)

Let me show you the two charts that scare me most.

The first is VO2 max — your body’s ability to consume and use oxygen, which is essentially a measure of how much energy you can generate. It declines linearly with age, even in people who have no diseases whatsoever. There’s a threshold below which you can’t breathe properly during sleep. That threshold is, effectively, a death sentence. And you approach it on a schedule that nothing we currently know how to do can meaningfully alter.

The second is cognitive performance. Most tests of fluid intelligence — the kind that measures real-time reasoning, not accumulated knowledge — decline steadily from your mid-20s onward. Crystallized intelligence (your ability to recall and articulate what you already know) stays relatively flat for longer. Which means, as I like to say, you can still talk eloquently long after you’ve stopped understanding what you’re talking about. Beware of articulate people.

These two trajectories are aging. Not cancer. Not diabetes. Not any particular disease. The continuous, linear degradation of your physical and cognitive function — independent of illness — that is what aging is. And here’s the uncomfortable truth: if you’re exercising, eating well, and have no chronic conditions, you are still aging. You are still on this curve.

This is not a disease model. It’s a physics problem.

The Naked Mole Rat and the Second Law

The naked mole rat lives in Somalia, in underground colonies, about the size of a mouse. It has a matriarchal social structure, a dominant breeding female, and — crucially — a mortality curve that is essentially flat with age. It doesn’t get more likely to die as it gets older. It is, by every biological definition, non-aging.

It is not alone. Nature is full of creatures that don’t age, or age negligibly. Hydra. Certain turtles. Some sharks. The existence of non-aging organisms in nature is now well established. Which means aging is not an inevitable property of living systems. It is a feature — one that evolution selected for in some lineages and not others.

This is why I think aging is ultimately a problem of physics, not just biology. The late Leonard Hayflick, one of the foundational figures in aging biology, proposed something similar: that aging is fundamentally a thermodynamic problem. I was trained in theoretical physics. When I encountered this framing, I couldn’t ignore it. It’s why a field now called “geroscience physics” or “gerophyics” has emerged, and why our work at Gero sits at that intersection.

What Machine Learning Taught Us About Aging

We feed medical histories — across tens of millions of people — into dynamic machine learning models. These models try to predict what happens to a person’s health over the full arc of their life. And when you train these systems on enough data, something interesting happens: they learn to separate aging from disease.

Not because we told them to. Because the signal is there in the data.

The models pull aging out as a distinct process, independent of specific disease trajectories. They identify genetic targets that are controlling the rate of aging in humans — not just predisposing people to particular diseases, but governing the underlying aging process itself. We’ve licensed one of these targets to a major pharmaceutical company. We expect there will be more.

This kind of analysis also lets us do something that was previously very difficult: measure maximum human lifespan in clinical data. It’s about 120 years. We know how to rejuvenate aging in mice. And we’ve arrived at a conclusion that not everyone likes:

In humans, you can stop aging. You cannot reverse it.

Most of human aging is thermodynamically irreversible. This is not what people want to hear at a longevity conference. But I think it’s one of the most important and actionable statements in the field — because it means the goal is not rejuvenation. The goal is stopping the clock.

Three Ways to Extend Human Life

Two hundred years ago, Benjamin Gompertz discovered the law of exponential aging — the observation that the probability of dying doubles roughly every eight years after maturity. He described aging as a combination of two processes: linear damage accumulation, and physiological noise (random biological fluctuation). This framing from 1825 maps almost exactly onto what our machine learning models are recovering from modern electronic health records.

Which means you can extend human life in three fundamentally different ways:

1. Cure diseases. This is what most of medicine does. It works, but the effect sizes are small. Even if you cured all cancers tomorrow, average lifespan would increase by about three years. Cure all diseases, and you might get ten years. That’s not nothing. But it’s not transformation.

2. Reduce the rate of damage accumulation. This goes after the linear deterioration component. If you could meaningfully slow the rate at which damage builds up in your cells and tissues, you could extend lifespan substantially — potentially far beyond what disease treatment can achieve.

3. Reduce physiological noise. This is what I find most interesting and most neglected. Noise — random biological fluctuation — is what separates the average human lifespan (around 80) from the maximum (around 120). The gap between average and maximum lifespan in humans is roughly 40 years. If you could reduce the noise, you could bridge that gap. Drugs that act on noise would produce large effects. This is where we are focused.

The vast majority of the longevity industry is working on option one. A newer generation of companies is working on option two — organ replacement, senolytics, epigenetic reprogramming. Very few are working on option three. That’s where the largest untapped leverage is.

Why “Reversing Aging” Is the Wrong Goal

There’s a fault line running through the longevity world right now, and it’s worth making explicit.

On one side are companies that believe aging is reversible — that you can take an 80-year-old and make them biologically 20. These companies tend to have large capital, large ambitions, and a belief in epigenetic reprogramming as the mechanism. On the other side is pharma — largely dismissive of the aging-reversal thesis, but quietly extending life one disease indication at a time.

And then there are a few of us who occupy a different position: we believe aging is not reversible in any deep thermodynamic sense, but it can be stopped. And we think that’s actually the bigger prize, because it’s achievable within the coming years rather than decades, and because stopping aging is the intervention with the largest population-level effect.

I’ll put it plainly: everyone who currently believes aging is fully reversible will either fail, or eventually become a conventional pharma company targeting disease indications. The second law of thermodynamics is not a recommendation. It’s a law.

Consider how we recognize age in a human face. You can see roughly how old someone is. You can also see whether they look healthy for their age — and that variation, roughly plus or minus five years, reflects reversible biological fluctuation. Sleep well and you look younger. Get sick and you look older. That variation is real and tractable.

But look at the size of someone’s nose. Their ears. These structures grow continuously with age, driven by irreversible collagen changes. No diet, no drug, no epigenetic intervention changes your nose size. That’s the second law of thermodynamics. The irreversible component is real, it’s large, and it’s currently being systematically underestimated by the field.

What You Should Actually Do (And When)

The most uncomfortable finding from our data analysis is this: the functional decline curve is largely set by your early adult peak.

Think of your body as a glider. It climbs to a certain altitude between roughly age 20 and 30, and then glides downward at a relatively fixed rate from there. The higher you climb — the better your physical and cognitive shape at your peak — the longer your glide path. The rate of descent is broadly similar for everyone. The starting height determines everything.

This means that for most people currently in midlife or beyond, the most important intervention is maintaining trajectory, not trying to reverse it. The reversible fluctuations around the trend are real and worth addressing. Metabolic health is the largest single accelerator of decline — diabetes can steal six or more years of healthy function. Managing metabolic health is the highest-ROI intervention available right now.

But no amount of current intervention brings a healthy 90-year-old back to their functional state at 50. That’s what the data shows. And it’s why so many people, when asked if they want to live to 250, say no. They’ve intuited the trap: more years in functional decline is not a gift.

The goal of longevity medicine should not be to extend the period of decline. It should be to compress it — or eliminate it. More years of function, not more years of survival.

The Regulatory Trap That’s Killing the Field

Here is a structural problem that almost nobody talks about directly, but that shapes everything:

If you’re developing a longevity drug, regulators will push you toward a disease endpoint. Prove it reduces cardiovascular events. Prove it delays cognitive decline in an already-impaired population. Because aging isn’t a disease, you can’t run a trial against aging itself.

The problem is that a drug optimized to beat a specific disease will almost always be beaten, on that specific endpoint, by a drug designed specifically for that disease. So longevity drugs — tested against disease endpoints — will consistently look like inferior disease drugs. They’ll fail, or they’ll struggle for approval, or they’ll get approved with narrow labels that don’t capture their real value. This is a regulatory trap.

The company Loyal is doing something clever with dogs — they negotiated with the FDA to run trials against lifespan as a primary endpoint in dogs, rather than against a specific disease. That’s a preview of what needs to happen in human medicine.

What I’d advocate for: let longevity biotechs run Phase 2 trials against any biomarker they choose, as long as the trials are rigorous and nobody is being harmed. Then move to Phase 3 basket trials — where you enroll people who have one age-related disease and wait to see how long it takes them to develop a second one. That measures the underlying rate of aging in a practical, disease-independent way.

The regulatory system that enables this framework first will have a significant market advantage. It’s not just a scientific question — it’s a geopolitical one.

What a True Anti-Aging Drug Might Look Like

If I had to bet — and I should note I’m a physicist, not a clinician — I’d bet it looks like a vaccine.

Here’s the logic. Most of the damage that accumulates in aging tissues is supposed to be cleared by the immune system. Macrophages, the body’s cellular janitors, are responsible for removing cellular debris, senescent cells, misfolded proteins. Evolution has already spent millions of years optimizing this system. We are, already, remarkably long-lived mammals. Our immune system is already doing heroic work to get us to 80 or 90.

The question is whether we can fine-tune it to work a little harder. Not replace it with something synthetic. Not introduce a new scavenging mechanism from scratch. Just train what’s already there to be slightly better at the job it evolved to do.

A vaccine-like intervention that recalibrates immune surveillance toward damage clearance — that’s my best current guess for what the first true anti-aging therapeutic looks like. Not a pill. Not a gene therapy. A recalibration of an ancient biological system.

The Humanitarian Case

I want to end with something that often gets lost in the excitement about technology and capital.

Demographic transition is already happening. People are living longer and having fewer children. Nobody asked us whether we wanted this. Medicine has effectively made it harder to die young, and that’s mostly a good thing. But the downstream consequence — millions of people spending their final decade in a state of profound functional decline, surviving on medicine but not living — is a humanitarian crisis in slow motion.

Most people in their 30s and 40s today will live into their 90s, possibly beyond. The question is not whether they will live long. The question is whether those additional years will be years of capacity and engagement, or years of managed deterioration.

The current medical paradigm — focused almost entirely on treating disease while accepting functional decline as inevitable — is not a neutral choice. It’s a choice for the wrong version of the future. Every year we delay building the scientific and regulatory infrastructure for genuine anti-aging medicine is a year that compounds into millions of people’s lived experience.

This is not a matter of preference about which biology to fund. It is a humanitarian imperative to shift the field’s center of gravity from disease treatment to aging itself.

More people in this industry — scientists, investors, regulators — should be asking not “how do we cure more diseases?” but “how do we slow the underlying process that makes people vulnerable to all of them?”

That’s the only question whose answer is large enough to matter.

Peter Fedichev is co-founder of Gero, a longevity biotech company using AI and physics-based approaches to understand and target the aging process. This piece is adapted from remarks delivered at the J.P. Morgan Healthcare Conference.

Subscribe now

When most people think about aging, they think biology. Cells, genes, proteins, damage. And that’s not wrong. But it may be incomplete in a way that matters enormously for whether we ever actually solve it.

I’m a physicist working on aging. The question I get most often is: why physics? After all, molecular biology is chemistry, which is ultimately physics — yet chemists work productively without invoking quantum mechanics, and biologists do excellent work without revisiting the Schrödinger equation. Isn’t this just physics envy dressed up as insight?

No. And understanding why reveals something deep about the nature of complex systems — and about why drug discovery keeps failing us.

More Is Different

The physicist P. W. Anderson said it plainly in 1972: more is different. As you move up the hierarchy of complexity — from particles to atoms, from molecules to cells, from cells to organisms — genuinely new phenomena appear. Not just quantitative differences, but qualitative ones. New variables. New laws. New languages of description.

The classic examples are striking. Superconductivity is not visible in any single electron; it arises from the cooperative pairing of many. Turbulence cannot be derived from the trajectory of any individual fluid molecule. Phase transitions — ice becoming water becoming steam — are macroscopic events with no microscopic counterpart. Even temperature is an emergent quantity: it is perfectly defined only when you have enough particles interacting that their individual fluctuations average out into something stable.

This is emergence. It is not mysticism. It is a precise, mathematically grounded statement about what happens when many components interact: the system develops properties that its parts do not have, governed by laws that operate only at scale.

The Arrow of Time — and an Analogy to Aging

Consider the second law of thermodynamics. At the microscopic level, the laws of classical and quantum mechanics are time-symmetric: run a movie of colliding particles backward, and it looks physically valid. Yet in any macroscopic system, tiny uncertainties grow through interactions until reversal becomes effectively impossible. Entropy increases. A direction in time emerges — not from any individual particle, but from their collective dynamics.

This is not a flaw in the microscopic laws. It is what happens when those laws operate at scale.

Aging, I would argue, is analogous. It is not simply the accumulation of molecular damage — though that happens. It is an emergent phenomenon: a macroscopic arrow of time in biology, produced by countless microscopic processes that individually reveal nothing about the organism’s trajectory. Species with radically different molecular aging mechanisms nonetheless show strikingly convergent aging phenomenology once you zoom out. The macro-level pattern doesn’t care much about which micro-level mechanism drives it. That is the hallmark of emergence.

The Hidden Cause Problem

Here is where a deep structural feature of complex systems becomes directly relevant.

Complex systems are extraordinarily good at hiding causes from effects.

The reason is structural, not accidental. Macroscopic behavior is governed by laws of large numbers. Most microscopic events average out. This averaging is precisely how new, coherent variables appear at each level of the hierarchy — they represent the statistical residue that survives the wash. The corollary is stark: micro-details are, most of the time, irrelevant to macro-outcomes.

Information, in a meaningful sense, does not flow upward through complex systems. Macro-features are insensitive to micro-specifics. Swap out one molecule for another, mutate one gene, tweak one pathway — and in most cases the large-scale behavior shrugs. The system absorbs the perturbation, averages over it, and continues on its emergent trajectory.

This is not a bug. It is the robustness that makes biological systems stable. But it is catastrophic for intervention.

When you act on a molecular target — as drugs almost always do — you are acting at the microscopic level. You are pulling a lever that the macroscopic system has been designed, through evolution, to be mostly indifferent to. The mapping from micro-intervention to macro-outcome is indirect, nonlinear, and distributed across many interacting components. You cannot predict the emergent result from the molecular action because emergence precisely means the result is not encoded in any individual component.

This is why drug discovery has a failure rate that would be scandalous in any engineering discipline. It is not primarily a failure of chemistry, or of clinical trial design, or of regulatory caution. It is a failure to account for emergence. We keep trying to treat complex systems as if they were simple ones — as if pulling the right molecular lever would straightforwardly produce the desired phenotypic outcome. The physics says this should rarely work, and the empirical record agrees.

What Physics Offers

Physics did not merely identify this problem. It developed tools to address it — and those tools carry an important lesson about what kind of theory is even possible.

The renormalization group provides a systematic method for identifying which microscopic details matter at large scales and which do not. It tells you how to find the “relevant parameters” — the small set of variables whose perturbation actually propagates upward and alters macroscopic behavior. Everything else is irrelevant in the technical sense: it averages out as you coarse-grain.

This framework produced two distinct levels of theory for the same phenomena. Landau-Ginzburg theory describes phase transitions — including superconductivity — in terms of macroscopic order parameters, without any reference to atomic details. It works across an enormous range of materials precisely because it is indifferent to their microscopic differences. Then there are microscopic theories: BCS theory for conventional superconductors, Anderson’s resonating valence bond theory for high-temperature ones. Each is valid for its own class of material, each captures a different underlying mechanism — and none of them unifies with the others into a single microscopic account. No one expects them to. The macro theory is universal; the micro theories are plural.

This distinction has a direct and underappreciated consequence for aging. Because information does not flow upward in complex systems, many different microscopic realizations can produce the same macroscopic phenomenology. Different species are, in this sense, different microscopic theories of aging. Yeast ages through accumulation of extrachromosomal DNA circles; mammals through telomere attrition, senescent cell burden, and proteostatic collapse. These are genuinely different mechanisms. Yet the macroscopic phenomenology converges: Gompertz mortality curves, declining physiological resilience, the characteristic shape of the aging trajectory. Just as different classes of superconductors share a macroscopic theory while differing at the molecular level, different species share a macroscopic aging phenomenology while running on entirely different molecular hardware.

The implication is uncomfortable for a field that has long searched for the mechanism of aging: there isn’t one. Expecting a unified molecular theory of aging is like expecting a unified microscopic theory of all superconductors — it mistakes the level at which the universal law actually lives. The macro theory works across all cases; the micro theories are necessarily case-specific. This is not a failure of biology. It is what emergence predicts.

Closing

The observation that biology is chemistry and chemistry is physics does not mean physics is superfluous. It means that the tools physics developed to handle emergence are exactly the tools we need. And the entanglement between physics and biology runs deeper than most people realize — it goes back to the very question of time’s direction.

Darwin and Boltzmann were contemporaries, and together they produced the deepest tension in nineteenth-century science. Darwin showed that complexity increases over time: life evolves toward greater organization, greater diversity, greater order. Boltzmann, working on the kinetic theory of gases, proved almost simultaneously that disorder must increase — that entropy rises in any sufficiently large system, giving time its arrow. Two of the greatest scientists of the same era, pulling in opposite directions. Evolution builds up; thermodynamics tears down. How can both be true?

Leonard Hayflick arrived at a version of the same conclusion from inside biology. Best known for discovering that human cells can only divide a finite number of times before entering senescence, Hayflick spent decades drawing a harder inference: aging is not a disease. It does not have a cause the way infections or cancers have causes. It is the second law operating on biological matter — thermodynamic drift accumulating across the hierarchy of organization until repair mechanisms, themselves finite, can no longer keep pace.

This does not mean aging cannot be slowed. The vast variation in lifespan across species demonstrates that the rate is tunable. But finding the levers requires knowing which microscopic variables actually couple to the macroscopic trajectory, and which are simply averaged away by the same statistical machinery that gives entropy its arrow. That is not a question molecular biology alone can answer.

Physics and biology have been circling the arrow of time together for a hundred and fifty years. Aging is where they finally have to converge.

Share

Leave a comment

Subscribe now