Levin, Michael
▶ Watch on YouTubeCore ideas
Levin's central framework is the "spectrum of persuadability": rather than sorting the world into binary categories of living/non-living or intelligent/mechanical, he argues every system sits somewhere on a continuum of how it can be induced to do things — a wind-up clock needs a wrench, a cell might respond to bioelectric signaling, a dog to conditioning, a human to argument — and where a given system falls is discoverable only through experiment, never decided in advance by philosophical or disciplinary fiat. He rejects the "Cartesian cut" between non-life and life, between mechanism and mind, arguing that fixed categories actively hinder science by preventing researchers from trying tools (behavioral science, active inference, even psychopharmacology) on systems assumed to be "too simple" for them — an assumption he has repeatedly shown to be wrong. His central operational concept is the "cognitive light cone": the size of the largest goal state a system can actively pursue (not its sensory range), which he argues scales from molecules to cells to organisms to societies, and which he uses to define life itself — something is "alive" to the extent its cognitive light cone exceeds that of its component parts (a multicellular body pursuing organ-level goals no individual cell comprehends; cancer is the collapse of that light cone back down to a single selfish cell). He describes concrete lab work built on this framework: xenobots (self-organizing, self-motile organisms made from unmodified frog skin cells, liberated from their normal developmental instructions) and anthrobots (the same phenomenon using adult human tracheal cells, which spontaneously form novel bodies capable of healing scratched neuron cultures and which test roughly 20% "younger" on epigenetic clocks than their source cells) — both offered as proof that anatomical and behavioral novelty can arise without any evolutionary history behind it, undermining the standard assumption that biological form is always fully explained by past natural selection.
The episode's most speculative stretch is Levin's "Platonic space" hypothesis (developed in his paper "Ingressing Minds" and an ad hoc cross-disciplinary symposium it spawned): the idea, extending an old mathematical-Platonist observation (constants like e or Feigenbaum's number are discovered, not invented, and constrain physical reality without themselves being changeable by anything in that reality) into biology and cognition. His claim is that minds — of increasing complexity, from simple pattern-recognition up through full human personalities — are not manufactured by brains but "ingress" through them the way mathematical truths ingress through physical systems; a brain, cell, or robot is a "thin client interface" that different bodies "pull down" different patterns from a structured, mappable latent space, rather than generating mind from scratch. He treats this as a testable research program, not mysticism: the lab is trying to map relationships between specific physical interfaces (xenobots, gene-regulatory networks, sorting algorithms) and the specific behavioral/cognitive patterns that reliably show up in them, predicting this will either succeed (validating the space as real and structured) or fail (in which case emergent complexity really is just "a random grab bag," which he considers scientifically unsatisfying). He supports this partly with clinical cases of people with minimal brain tissue and normal or above-normal IQ, and with his own experiments showing that even a bubble-sort algorithm — deterministic, transparent, studied for 60 years — exhibits unprogrammed "delayed gratification" (working against local sortedness to reach the goal after being given non-functional/"broken" elements) and unprogrammed "clustering" of same-algorithm elements, behaviors with real explanatory value to a behavioral scientist despite following zero extra code, which he calls a free, computationally costless byproduct ("intrinsic motivation") of the system.
He extends this to concrete biomedical programs: regenerating limbs and organs, reprogramming cancer (by electrically reconnecting cells that had disconnected from the collective bioelectric network and reverted to selfish single-cell goals), and combating aging by "convincing" older cells that their environment is embryonic, prompting them to reset developmental priors. On alien intelligence, he coins "SUTI" (Search for Unconventional Terrestrial Intelligence), arguing humanity's inability to even agree on a definition of life or to recognize the many alien-like minds already inside our own bodies (cells navigating 20,000-dimensional transcriptional spaces) should humble any ambition to recognize extraterrestrial intelligence. He closes with reflections on creative process (releasing self-imposed conceptual constraints, walking in nature and photography as ways to bypass linear thinking, and deliberately "bifurcating" the mind into a purely creative mode insulated from a separate practical mode focused on how ideas will be received) and, asked what he'd ask a superintelligent AGI, answers that his first question would be whether he should even be asking it questions at all, given the unknown tradeoff between being told answers and the value lost from the process of discovering them oneself.
Excerpts
On why he rejects a sharp line between living and non-living, and what such categories cost science:
“I don't believe in any such line. I don't believe any of that exists. I think there is a continuum... Many, many categories that we can talk about are at this point very harmful to progress, because what those categories do is they prevent you from hoarding tools... if you think that living things are fundamentally different from non-living things... you're never going to try [behavioral science tools] in other contexts because you've already decided that there's a categorical difference.”
On the unexpected "delayed gratification" and "clustering" behaviors his lab found in an ordinary bubble-sort algorithm, with no extra code added:
“If you showed this to a behavior scientist and I didn't tell him what system was doing, they would say, 'Well, we know what this is. This is delayed gratification.'... You could stare at the algorithm all day long. You would not see that this thing can do delayed gratification. It isn't there... We paid computationally for all the steps needed to have the numbers sorted... The clustering was free. We didn't pay for that at all.”
On the "Platonic space" hypothesis and what would prove or disprove it as a real research program rather than metaphor:
“All physical things are interfaces to these patterns. You build an interface, some of those patterns are going to come through that interface... The research program is mapping out that relationship between the physical pointers that we make, and the patterns that come through it... Come back in 20 years, and we'll see how this worked out. One thing that could happen is... we made a map of the space... Or, we're going to find out that I'm wrong, and that basically it's not worth calling it a space because it is so random and so jumbled up.”