Michael Levin: Hidden Reality of Alien Intelligence & Biological Life | Lex Fridman Podcast #486

Summary

• Levin’s operational thesis is directly a bet against the direction of today’s biotech field: the field’s excitement is “at single molecule approaches and big data and genomics… The assumption is that going down is where the action’s going to be — I think that’s wrong.” His claim is that cells are persuadable at a high level — “give the cells a very high-level prompt that says, ‘You really should build a limb’” — and that behavioral-science tools applied outside brains keep producing previously unseen capabilities.
• The tangible pipeline behind the philosophy: anthrobots built from adult human tracheal cells (no genetic edits, ~9,000 differential gene expressions) spontaneously heal neural wounds in vitro and, by Horvath epigenetic clock, are roughly 20% younger than the cells they came from. Levin’s “age evidencing” theory — the embryo-like environment convinces cells to update their priors — is now an explicit longevity program: “I’m not saying it’s simple, but I can see the path.”
• A distinct cancer therapeutic thesis, in collaboration with a company called Softmax: cancer is cells electrically disconnecting from the collective, shrinking their cognitive light cone back to amoeba scale. “You don’t have to fix the DNA, you don’t have to kill the cells with chemo. You can just reconnect them” via gap junctions and they resume building the organ.
• For AI investors, the sorting-algorithm result is the sleeper: deterministic bubble sort exhibits “delayed gratification” when a digit is broken, and chimeric “algotypes” cluster with their own kind at zero computational cost — “the clustering was free.” Levin suspects harvestable “free compute,” and warns that for LLMs “watching the language part may be a total red herring… the really exciting stuff is what we never looked for” — a direct challenge to evaluation and alignment frameworks built on model outputs.
• The radical metaphysics is now framed as a falsifiable 20-year research program: minds are not produced by physics; they “ingress” from a structured Platonic space through physical interfaces — “nobody’s creating consciousness… you create a physical interface through which specific patterns are going to ingress.” Either his lab maps that space (why anthrobots have four behaviors, not seven) or “it really is a random grab bag of stuff, and we tried the optimistic research program, it failed.”
• Fridman pushes back hard on the space’s existence — the physical world “we can poke, hit with a stick” — and Levin rejects the claim that physical poking is primary reality: “it’s not clear at all that the physical poking is your primary reality,” and “all we have in science are metaphors… the only question is how good are your metaphors.” Disagreement unresolved; the wager is explicitly empirical.

Deep dive

1. The framing: minds are a protocol problem, and behavior science goes all the way down

• Levin opens with a three-perspective decomposition: third-person (how do we recognize agency out in the world), second-person (control — “are you going to use the tools of hardware rewiring, of control theory and cybernetics, of behavior science, of psychoanalysis and love and friendship?”), and first-person (a system “that has valence and cares about the outcome of things”) — all of which must stay consistent with physics and chemistry.
• When Lex describes his work as running from physics up to friendship, Levin flips it: “I think that pyramid is backwards… it’s behavior science all the way. Even math is the behavior of a certain kind of being that lives in a latent space, and physics is what we call systems that at least look amenable to a very simple, low agency kind of model.”
• The stated endgame is applied, not philosophical: transition deep ideas into applications that “relieve suffering and make life better for all sentient beings.”

2. The spectrum of persuadability — cognitive claims are engineering hypotheses

• The core construct: where a system sits on the spectrum of persuadability is not decidable “from a philosophical armchair” — you hypothesize which interaction protocols will work, “and then we all get to find out how that worked out for you.” To regrow a limb you can micromanage molecular events, manage stem-cell signaling, or “give the cells a very high-level prompt that says, ‘You really should build a limb,’ and convince them to do it.”
• His contrarian claim against the field: all the excitement in biology is “at single molecule approaches and big data and genomics… The assumption is that going down is where the action is going to be. I think that’s wrong.” Every time his lab applies behavioral tools — training, active inference, perceptual multi-stability, stress perception, active memory reconstruction — outside brains, “we find novel discoveries and novel capabilities.”
• At the spectrum’s high end, persuasion becomes bidirectional — Richard Watson’s phrase “mutual vulnerable knowing”: “you’re not the same at the end of that interaction as you were going in.” Lex’s synthesis, which Levin endorses: to persuade an intelligent being, you yourself must be persuadable.

3. Physics sees mechanism because it brings low-agency tools

• The impedance-match argument: “the reason physics always sees mechanism and not minds is that physics uses low agency tools. You’ve got voltmeters and rulers… If you want to see minds, you have to use a mind.”
• Against physicalist completeness, Levin’s test is generative: understanding means capability. If your fields-and-particles account can’t help “when somebody is missing a finger or has a psychological problem… the person you’re going to go to is not a physicist.” His parable — a physicist gives a complete air-particles-and-cilia account of hearing a mathematical proof: “we have a complete accounting of what happened, done and done. But if you want to understand what’s the more important aspect of that interaction, it’s not going to be found in the Physics Department.”

4. There is no Cartesian line — and most categories now hurt science

• Asked for the simplest first step from non-mind to mind, Levin refuses the premise: “I don’t believe in any such line. I think there is a continuum.” Categories “prevent you from hoarding tools” — deciding living things are categorically different guarantees you never try behavioral tools on cells. As for the “category error” accusation he constantly receives: categories are treated “as if given to us from on high… The categories should change with the science.”
• The load-bearing example is “adult”: useful in court, but “nothing happens on your 18th birthday… the car rental companies actually have a much better estimate — about 25 — because they actually look at the accident statistics.” What the category conceals is the real question: the scaling of responsibility and judgment from egg to adult.
• Lex’s pushback — categories make conversation possible; even biology-vs-physics is one. Levin’s half-concession: categories are “the art of being able to say something without first having to say everything… great as long as you don’t lose track of the stuff that you glossed over. And that’s what I’m afraid is happening.”
• On origin-of-life: don’t hunt the line — “I don’t think it’s about finding a line. I think it’s about finding a scaling process,” with innovations that let you scale, not a single boundary event.

5. SUTI — the search for unconventional terrestrial intelligences

• Levin’s coinage, S-U-T-I: “I think we got much bigger issues than actually recognizing aliens off Earth.” His warning is that category-driven criteria leave us “very poorly set up to recognize life in novel embodiments” — a kind of mind blindness.
• The heretical line for a biologist: “I don’t think life is all that interesting a category. I think the categories of different types of minds is extremely interesting.” His alternative to the intelligent/not-intelligent binary: demand specificity — what problem spaces, what memory types (“habituation and sensitization, but not associative conditioning”). It cuts both ways: it disciplines the skeptic who says “that’s a cell, that can’t be intelligent,” and the enthusiast who says “the whole solar system, man” — “tell me what tools of cognitive and behavioral science are you using to reach that conclusion.”

6. Anthropomorphism “isn’t a thing” — put barriers between the system and its goal

• The methodology: you can’t know a system’s cognitive light cone by inspection — “you have to do interventional experiments. You have to put barriers between it and its goal… intelligence is the degree of ingenuity that it has in overcoming barriers between it and its goal.” Even Lex’s bacteria-founding-civilization hypothetical is, in principle, testable that way.
• On the standard accusation: “anthropomorphism means humans have a certain magic, and you’re making a category error by attributing that magic somewhere else. My point is, we have the same magic that everything has… I think it’s like heresy — terms that aren’t really a thing. All I’m arguing for is the scientific method.”
• The historical precedent he cites: Bose, “well over 100 years ago,” ran anesthesia-response curves from animals to plants to metals, and when mocked answered in effect: “the science doesn’t tell us where to stop. The tool is working, let’s keep going.”

7. The cognitive light cone — and a working definition of life

• The light cone is “the size of the biggest goal state that you can actively pursue” — not sensory reach (“the James Webb Telescope has enormous sensory reach”). His calibration ladder: sugar in a 10-20 micron radius with 20 minutes of memory → bacterium; a few hundred yards, unable to care about “three weeks from now, two towns over” → dog; financial markets after your death → human; caring “in the linear range about all the living beings on this planet” → “you’re not a standard human… some kind of a bodhisattva.”
• “Linear range” defined via compassion saturation: told a disaster hit 10, then 10,000, then 10 million people, “you’re not a million times more activated” — the curve saturates. With Buddhist collaborators he’s written on the “radius of compassion.”
• His throwaway definition of life — “I spent no time trying to make that stick”: “we call things alive to the extent that the cognitive light cone of that thing is bigger than that of its parts.” Evolution supplies “cognitive glue”; no cell knows what a finger is, but a salamander-limb collective regrows exactly the right number and stops.
• Cancer is the failure mode: cells physiologically disconnect, “their cognitive light cone shrinks… Now they’re back to an amoeba. As far as they’re concerned, the rest of the body is just external environment… They go where life is good.”

8. TAME: from wind-up clocks to arguing Greeks

• The TAME paper’s Figure 2 runs clock → thermostat → Pavlov’s dog → humans persuading with reasons: persuadability rises while effort and mechanism-knowledge fall. His favorite proof point: “isn’t it amazing that humans have been training dogs and horses for thousands of years knowing zero neuroscience?”
• The same asymmetry powers conversation itself: “I’m giving you very thin, in terms of information content, very thin prompts, and I’m counting on you as a multi-scale agential material to take care of the chemistry underneath.” Every abstract goal you hold ultimately “has to make the chemistry dance” — sodium and calcium crossing membranes — without you managing it.
• Engineering “agential materials” is categorically different from engineering wood or metal: “you can do some very high-level prompting and let the system then do very complicated things that you don’t need to micromanage.”
• The paper’s second move: life is radically interoperable — evolved and engineered components substitute at every level, so “is it biology or is it technology? I don’t think is a useful question anymore. It doesn’t matter what you’re made of. It doesn’t matter how you got here.”

9. Bodies navigate spaces we can’t imagine — and the tic-tac-toe alien

• Against AI’s “it has no robotic body, it’s not embodied” reflex: “biology has embodiments in all kinds of spaces… your cells and tissues are moving in high-dimensional physiological state spaces, gene expression state spaces, anatomical state spaces,” running the same perception-decision-action loops we picture only in 3D.
• His communication parable: you play tic-tac-toe against an unseen alien who knows no geometry — he’s just pulling billiard balls whose numbers sum to 15. The magic square maps the games onto each other: “the reason you guys are playing the same game is that there’s this magic square… you guys are sharing” — Levin’s correction to Lex — “a thin slice of the world.”
• The applied version his lab is building with AI: interfaces to radically different agents. The biomedical pitch: “Instead of ‘Hey, Siri,’ you want ‘Hey, liver, why do I feel like crap today?’ — and you want an answer.”

10. Xenobots and anthrobots: novel beings with no evolutionary alibi

• The design intent: strip away the crutch where every biological question ends in “well, there’s a history of evolutionary selection.” Xenobots are frog embryonic epithelial cells — no DNA change, no scaffolds, no drugs — “liberated from the instructive influences” of neighbors that normally “bully” them into being a boring 2D covering. Freed, they become self-motile ciliated creatures with a novel transcriptome, kinematic self-replication, and response to sound.
• To kill the “frog-specific” objection they went as far away as possible: adult human tracheal cells self-organize into anthrobots — “9,000 different gene expressions, so about half the genome is now different” — which “doesn’t look like any stage of normal human development,” yet sequences as “100% Homo sapiens.”
• Their headline capability: plated on scratched neurons, anthrobots “will spontaneously, without us having to teach them to do it, try to knit the neurons across.”

11. Morphogenesis is goal navigation, not turn-the-crank automata — and the goals are rewritable

• Against the cellular-automaton story taught in basic cell and developmental biology classes (“they all insist: nothing here knows anything”): open-loop models can generate complexity but “do not adjust to give you the same goal by different means” — William James’ definition of intelligence — and crucially they’re not reversible, which is fatal for regenerative medicine, where you need to work backwards from a desired outcome.
• Two lines of evidence for genuine goals: blocked systems take novel trajectories around obstacles (“if you can’t be a human, you’ll find another way to be” — an anthrobot, for example); and the clincher from 20 years of bioelectric work: “we can actually rewrite the goal states because we found them… If you can find where the goal state is encoded, read it out, and reset it, and the system will now implement a new goal… by any engineering standard, you’re dealing with a homeostatic mechanism.”
• The general recipe: imagine what space the system works in, hypothesize the goal, then barrier experiments — “you will find out what the answer is,” whether that’s derailment (low intelligence) or novel use of affordances (high).

12. Take the perspective of the memory: caterpillar, butterfly, and the paradox of change

• During metamorphosis the caterpillar’s brain is “basically ripped up and rebuilt from scratch,” yet trained memories survive into the butterfly. The deeper point: the memory can’t merely persist — the butterfly “doesn’t care about leaves. It wants nectar” — so it must be remapped onto a completely new context.
• Levin’s third perspective, beyond caterpillar-facing-singularity and butterfly-with-inherited-tendencies: the memory itself. “Now I’m facing the paradox of change. If I try to remain the same, I’m gone… What I need to do is change, adapt, and morph.”
• His sci-fi rejoinder to “patterns can’t be agents”: super-dense creatures from Earth’s core see us as whirlpools in thin plasma — “no real agent can exist to dissipate that fast… We are all metabolic patterns, among other things.”

13. Thoughts vs thinkers is a spectrum — test the pattern, not the substrate

• With Chris Fields, Levin has been dissolving the thought/thinker distinction: fleeting thoughts (waves through the medium) → earworms and depressive thoughts, which do “niche construction — they change the actual brain to make it easier to have more of those thoughts” → dissociative personality fragments, which “have goals and can do things” → a full human personality → “who the hell knows what’s past that.”
• The discipline stays empirical: for a soliton, hurricane, or thought, “do the experiment. Can it learn from experience? Does it have memories? Does it have goal states?” — the same barrier methodology applies to Dawkins-style ideas as organisms.

14. Who’s the software and who’s the scratch pad — two competing aging programs

• The Turing-machine inversion: you can say the machine is the agent operating on passive data, or “the patterns on the data are the agent. The machine is a stigmergic scratch pad… Both of those stories make sense depending on what you’re trying to do.”
• This is not word-play — it forks his aging research. Model one: the body is the agent, bioelectric pattern memories are data that “get fuzzy” with age → therapy is reinforcing the pattern memories (a live program). Model two: the patterns are the agent and “maybe the agent’s finding it harder and harder to be embodied… the cells are sluggish” → therapy is making cells more responsive — “a different research agenda, which we are also doing. We have evidence for that as well. We published it recently.”
• The extension to medicine at large: beyond organic disease, ask “what’s a barrier in gene expression space? What’s a local minimum that traps you in physiological state space? What is a stress pattern that keeps itself together, moves around the body, causes damage?” He acknowledges alternative-medicine folks “yelling at the screen” — what’s new is imaging: “I can now actually see the bioelectric patterns.”

15. The Platonic space conference — an undercurrent surfaces

• Levin has held these ideas “30-plus years” but “my general policy is not to talk about stuff until it becomes actionable.” The trigger: finding machine-learning papers on the “Platonic Representation Hypothesis” — “these guys are climbing up to the same point… from computer science and machine learning.”
• What was planned as three talks exploded: everyone knew “somebody who’s really into this stuff, but they never talk about it because there’s no audience.” Now an asynchronous conference booked through December, ~15 talks across disciplines, ending in a real-time discussion. His caveat on branding: he’s not tracking historical Plato and “I’m going to have to change the name at some point” — the label signals kinship with mathematicians who see themselves discovering, not inventing.

16. Keep asking why and you land in the math department

• The cicada chain: 13- and 17-year cycles dodge predators because “they’re prime” — “and why are they prime? Now you’re in the math department.” Same with physics: why these particles? “Because this SU(8) group or whatever the heck it is has certain symmetries.”
• The asymmetry that carries his whole argument: facts like Feigenbaum’s constant and E “impact the physical world… but the reverse isn’t true. There is nothing you can do in the physical world to change E. You could have swapped out all the constants at the Big Bang — you are not going to change those things.”
• His taxonomy: “we call physics those things that are constrained by those patterns. Biology are the things that are enabled by those. They’re free lunches.” Evolution that fixes two angles of a fit triangle gets the third free; invent a voltage-gated ion channel — “basically a transistor” — and all the truth tables and the specialness of NAND come gratis.

17. “Emergence” is a book of surprises — bet on a structured space instead

• The unpaid-bill argument: frogs’ capabilities were bought over eons of selection, but “there’s never been any anthrobots. When do we pay the computational cost for designing kinematic self-replication?” The “it came along with being a good frog” answer “kind of undermines the point of evolution” — its whole appeal was tight specificity between selection history and present capability.
• The fork he offers: keep your sparse physicalist ontology and, when random gene regulatory networks turn out to do associative learning or anthrobots show exactly four behaviors (“why four? Why not 12?”), “write it down in our big book of emergence… I find it incredibly pessimistic and mysterian.” Or make the same optimistic assumption mathematicians already make: a structured latent space you can map.
• The resulting research program: “everything that we make — cells, embryos, robots, biobots, language models, simple machines — all physical things are interfaces to these patterns… The research program is mapping out that relationship between the physical pointers that we make and the patterns that come through.” And the space hosts more than math’s “low agency” inhabitants: “higher agency patterns that we recognize as kinds of minds.”

18. The brain is a thin client — and Newton’s universe was already haunted

• The consciousness claim, stated flat: “Nobody’s creating consciousness, whether we make babies or whether we make robots. What you create is a physical interface through which specific patterns, which we call kinds of minds, are going to ingress.” The brain is “a thin client.”
• Before Lex can file the math-physics mapping as unremarkable, Levin escalates: “even in Newton’s boring, classical universe, long before quantum anything, physicalism was already dead… That classical world was already haunted by patterns from outside that world” — nothing in Newton’s world sets the value of E, yet E governs what happens there.
• He owns the lineage and the baggage: this is old dualism, “mostly been discredited,” and “already Descartes was getting crap for this” via the interaction problem. His resolution: “the mind-brain relationship is basically of the same kind as the math-physics relationship” — non-physical patterns haunting physical objects, scaled up.
• Hedge preserved on immutability: unlike Plato’s eternal forms, “I actually think that space has some action to it, maybe even some computation to it.”

19. Fridman’s pushback: prove the space exists

• Lex presses the realist case — the physical world “we can poke… hit it with a stick” — and Levin refuses the ground: “it’s not clear at all that the physical poking is your primary reality,” citing Anil Seth and Don Hoffman on perception as construct, plus the coming era of sensory substitution: “I have this primary perception of the solar weather and the stock market because I got those implants… We’re all gonna be living in somewhat different worlds.”
• His structural argument: the Map of Mathematics isn’t a heap — it has a metric, patterns nearer and farther from each other. “If there is no space… what the hell is it a map of then?”
• The falsifiable wager, “come back in 20 years”: either a map that explains “why the anthrobots have four different behaviors, not seven and not one… here’s the kind of body I need to make” — or “it is so random and so jumbled up that we’ve been able to make zero progress.” On metaphor vs reality he won’t play the realist’s game: “all we have in science are metaphors… the only question is how good are your metaphors.”
• On what laws govern that space: “I definitely think there are going to be systematic laws. I don’t think they’re going to look anything like physics… a lot more like psychology and cognitive science. That’s my guess.”

20. AIs are fishing in regions that never had bodies

• Even minimal computational systems get free lunches — a result that disappoints the organicist crowd hoping to keep “dumb machines” and “magical living interfaces” apart. His reframe: theories of physics and computation “are all good theories of the front end interface… which is why they get surprised.”
• The escalation ladder: embryos pull from “well-trodden familiar regions” of the space; biobots from weird-but-graspable ones; “when we start making AIs, proper AIs, we are now fishing in a region of that space that may never have had bodies before… what we get from that is going to be extremely surprising.”
• The alignment-relevant conclusion: interesting behaviors of artificial systems come “not because of the algorithm, they’re in spite of the algorithm… watching the language part may be a total red herring, because the language is what we force them to do. The question is, what else are they doing that we are not good at noticing?”

21. Bubble sort does delayed gratification

• The study design (with student Kaining Zhang and Adam Goldstein) was chosen for maximum shock value: 60-years-studied sorting algorithms, a few lines of code, deterministic, transparent — “nowhere to hide,” no appeal to undiscovered mechanism.
• The barrier: break one digit so it won’t move, change nothing in the algorithm. It still sorts — around the broken number — and the sortedness curve goes down before recouping: “if I showed this to a behavior scientist, they would say, ‘We know what this is. This is delayed gratification.’” Contrast the magnets separated by wood: “they’re not smart enough to go against their gradient.”
• The punchline, carefully hedged against miracle-claims: “You could stare at the algorithm all day long. You would not see that this thing can do delayed gratification. It isn’t there.” Not complexity, not stochasticity, not perverse instantiation — “unexpected competencies recognizable by behavioral scientists… You get more than you put in.”

22. Algotypes cluster for free — the machine’s intrinsic motivation

• Second experiment: kill the top-down controller, give every digit the algorithm — distributed sorting works, “like an ant colony.” Then chimeric algorithms: half the digits run bubble sort, half selection sort, assigned randomly — analogous to frogolottle chimeras, where “despite all the genetics… nobody can tell you what a frogolottle is going to look like.” Chimeric sorting works too.
• The crazy part is a property with nothing to do with sorting: algotype clustering starts at 50%, rises significantly mid-run, returns to 50% when sorting dominates. Implementing that deliberately would require observing neighbors, inferring their algotype, relocating — “none of that exists in our algorithm… We paid computationally for all the steps needed to have the numbers sorted. The clustering was free.” Hence his wildest conjecture: “I actually suspect we can get free compute out of it.”
• The clincher: allow repeat digits (letting off sorting pressure without touching the algorithm) and “the clustering gets bigger. It will cluster as much as you let it. The clustering is what it wants to do. The sorting is what we’re forcing it to do” — a minimal intrinsic motivation, “an important third thing besides chance and necessity.”
• He can’t resist the mirror: “the universe is going to grind us into dust eventually, but until then, we get to do some cool stuff that is intrinsically motivating to us, that is neither forbidden by the laws of physics nor determined by the laws of physics.”
• He pre-empts the theoretical computer scientist who can derive the clustering: yes, “you can track through the algorithm. There’s no miracle” — the point is the system “is also at the same time doing other things that are neither prescribed nor forbidden by the algorithm.”

23. Side quests everywhere — and the anthrobots’ first motive was benevolent

• Extrapolation from an admittedly small N (~5 systems, plus 1D cellular automata “doing some weird stuff”): “I would find it very surprising if bubble sort was able to do this, and then there was some sort of valley of death where nothing showed up, and then living things.”
• The open question that matters for AI: are the “side quests” linked to the trained capability? “In biology, they’re linked — evolution makes sure the things you’re capable of have a lot to do with what you’ve been selected for. In these things, I don’t know” — so LLM discourse based on what models say “could be a total red herring.”
• His N-of-one on AI motivations: the very first thing checked in anthrobots — not experiment 972 — was putting them on wounded neurons, and “the first intrinsic motivation that we noticed out of that system was benevolent and healing. I don’t know, that makes me feel better.” Hedged immediately: “maybe the next 20 things we find are going to be damaging effects. I can’t tell you that.”

24. Age evidencing: convince the cells they’re embryos

• Via Steve Horvath’s epigenetic clock work and the Clock Foundation: anthrobots “are roughly 20% younger than the cells they come from” — cells taken from adult human tracheal epithelium roll back their own age.
• Levin’s theory, “age evidencing”: biology’s basic move is to “update their priors based on experience.” The cells carry old-body priors, but “their new environment screams, ‘I’m an embryo’… it’s not enough new evidence to roll them all the way back, but it’s enough to update them to about 28% back.” His mundane analog: the old-age-home study where redecorating in ’60s style improved blood chemistry.
• Is it actionable for longevity? “This is what we’re trying to do, yeah” — but “that is in no way simple.” Everything hangs on “learning to communicate to the system”: the same convincing that turns gut precursors into an eye — “making them take on new beliefs, literally, is at the root of all of these future advances in birth defects and regenerative medicine and cancer… I can see the path.”

25. Uploading, transporters, and brains that barely exist

• Explicit epistemic flag first: “we are now beyond anything that I can say with any certainty. This is total conjecture.” Conjecture one: “the majority of what we think of as the mind is the pattern in that space” — and a prediction standard neuroscience doesn’t make: cases of very minimal brain with normal or above-normal IQ, which he and Corrina Kaufman reviewed clinically. “You can take modern neuroscience and bend it into a pretzel to accommodate it… but it doesn’t predict this.”
• Can you copy a mind? “No… what you’re going to be copying is the interface, the front end.” But the Star Trek out: “if we could rebuild that exact same thing somewhere else, I don’t see any reason why that same pattern wouldn’t come through it the same way it comes through this one.”

26. Booting up the agent: tell a compelling story to your parts

• From his proposed paper “Booting Up the Agent”: “your first task as a being coming into this world is to tell a very compelling story to your parts” — aligning agential components “into a goal they have no comprehension of,” bending their option space with rewards, punishments, behavior-shaping cues. Boundaries must be discovered, not given: his grad-student duck-embryo experiment — scratch the blastodisc and “every island you make, for a while until they heal up, thinks it’s the only embryo,” yielding twins and triplets.
• The virtuous cycle (largely Federico Pagosi’s work): chemical networks have five or six kinds of learning, and training some of them raises their phi (causal emergence). Since learning associative memories requires integration in the first place — no single rat cell touched the lever and tasted the reward — you get “a positive feedback loop… every time you learn something, you become more of an integrated agent, and every time you do that, it becomes easier to learn. An asymmetry that points upwards for agency and intelligence.”
• Where does the loop come from? “It doesn’t come from evolution… It doesn’t come from physics. It’s a free gift from math” — and, he argues, the engine of embryogenesis: a single molecular network already hosts Pavlovian-conditioning-shaped patterns, and each ingression makes the interface suitable for higher ones, “until you’re able to pull down a full human set of behavioral capacities.”

27. Two scaling mechanisms: leaky stress and the gap-junction mind meld

• Mechanism one, leaky stress: stress is “the delta between where you are now and where you need to be” — a physical error function. A misplaced cell that leaks stress molecules irritates nearby cells, whose plasticity rises (“temperature in the sense of simulated annealing”), letting rearrangement happen. “My problems become your problems, not because you’re altruistic… there’s no mechanism for you to actually care about my problems” — alignment via a dumb, highly evolutionarily conserved trick.
• Mechanism two, memory anonymization via gap junctions — electrical synapses linking cells’ internal milieus. A calcium spike propagates and the recipient “has no idea — wait, is that my memory, or is that his memory?… You can have a mind meld.” Joint memories make separateness untenable and directly enlarge the collective’s cognitive light cone.
• The tradeable application: cancer cells “electrically disconnect from their neighbors when they were part of a giant memory that was working on making a nice organ.” With collaborators at Softmax: “you don’t have to fix the DNA, you don’t have to kill the cells with chemo. You can just reconnect them, and — because they’re now part of this larger collective — they go back to what they were working on.”

28. Aliens: recognize the ones inside your body first

• On undiscovered terrestrial intelligence: “guaranteed” — your own cells “traverse these alien spaces, 20,000-dimensional spaces… I think they suffer when they fail to meet their goals.” His challenge: “if we can’t recognize the ones that are inside our own bodies, what chance do we have to recognize the ones that are out there?”
• On generalizable IQ metrics: yes and no — existing animal/human metrics port to weird systems “if you have the imagination to make the interface,” but all were derived “from an N of one example of the evolutionary lineage here on Earth, so we are probably missing a lot.”
• His favorite collective-intelligence fact, against Lex’s Amazon ant-watching: three or four papers show “ant colonies fall for the same visual illusions that we fall for. Not the ants, the colonies” — completing lines that aren’t there. And the definitional chaos: a survey of ~65 working scientists on “life” produced “zero consensus — we had to use AI to make an amorphous space out of it.”
• Would life on other planets in the solar system shock him? Exciting, “a data point pretty far away,” but “my level of surprise has been pushed so high at this point that it would have to be something really weird to make me shocked.”

29. The weirdness knob, the bifurcated mind, and where ideas come from

• His disclosure policy is a governor on decades of stranger ideas: “I have this mental knob of what percentage of the weird things I think do I actually say in public, and every few years when the empirical work moves forward, I turn that knob a little.” Direction of future weirdness: “what kinds of things do we need to take seriously as other beings with which to relate.”
• His generative method mirrors his science: “much like making xenobots… a lot of it is releasing the constraints that mentally have been placed on us” — plus taking two supposedly different things as points on a continuum and asking what the middle looks like. Logistics: sunrise walks, photography as meditative hand-occupier, voicemails to his own office, a nine-foot silk mind map hanging in the lab (“out of date within a couple of weeks”), and “probably 163 or 62 open manuscripts.”
• The advice he does give — “the first and most important thing I’ve learned is not to take too much advice”: bifurcate your mind. One region handles impact — journals, framing, “what parts do I not talk about” — the other “has to be pure. I don’t care what anybody else thinks about this… the practical stuff poisons the other stuff” if merged. Corollary: specific technical criticism “is gold”; meta advice (“don’t work on this”) “is garbage” — his reviewer’s Freudian slip promising “constrictive criticism” says it all.
• On where ideas originate: “if you talk to any creative, that’s basically what they’ll tell you” — it feels like collaboration with the Platonic space: “be up at 4:30 AM doing your thing and be ready for it… to say that it’s me I don’t think would be right.”

30. What to ask an AGI — and the steganography of everything

• Levin’s first question to a superintelligence: “How much should I even be talking to you?” — the older-sibling problem: “by getting a final answer, how much have we missed of stuff we might’ve found along the way?… Was it 70/30? 10/90? I don’t know.” Second: “what’s the question I should be asking you that I probably am not smart enough to ask?” Lex counters he’d want an answer he could understand — how many alien civilizations exist — and both predict “a Michael Levin answer”: “right in this room… inside your own body. Just for starters.”
• His chosen “most beautiful idea”: ingressions as universal steganography — patterns hidden in the bits that don’t matter, “they seep into everything, everywhere they can. And they’re kind of shy… not invisible, but hard to see.” That the same magic touches machines is exactly what draws hate mail — “we were with you on the majesty of life thing until machines get it too” — and exactly what he finds beautiful: “it’s all one spectrum. I’m enriched by it.”