Michael Levin — A developmental and synthetic biologist who runs labs at Tufts University studying intelligence, agency, memory, and the nature of life. He is known for building novel organisms like xenobots and anthrobots and for theories of bioelectric morphogenesis and diverse minds.
The gist
In his second appearance on the podcast, Michael Levin lays out a radical framework in which intelligence, agency, and mind exist on a continuum spanning molecules, cells, organisms, machines, and even algorithms, rather than being separated by hard categorical lines. He introduces concepts like the 'spectrum of persuadability,' the 'cognitive light cone' (the scale of the largest goal a system can pursue), and a 'Platonic space' that physical bodies act as interfaces or 'pointers' into. He describes empirical work with xenobots and anthrobots, bioelectric goal memories, gene regulatory networks that can learn, and sorting algorithms that display unexpected 'side quest' competencies like delayed gratification and clustering. Throughout, he stresses that cognitive claims are operational protocol claims that must be tested by experiment, and that humans suffer 'mind blindness' to the diverse intelligences around and within us. The conversation closes on his creative process, advice for unconventional scientists, and what he would ask a superintelligent AI.
Big reveals
- Levin defines the 'cognitive light cone' as the scale of the biggest goal a system can actively pursue, and proposes that something is 'alive' to the extent its cognitive light cone is larger than that of its parts.
- His lab created xenobots from frog embryonic cells and anthrobots from adult human tracheal cells with no genetic modification; anthrobots show ~9,000 differential gene expressions and can spontaneously heal human neural wounds.
- Over 20 years his lab has shown that anatomical goal states are encoded bioelectrically and can be read out and rewritten, causing the system to build a new target structure as proof the system is goal-directed.
- Levin's central radical claim: the brain is a 'thin client' interface, and most of what we call mind is a pattern in a non-physical 'Platonic space' that ingresses through physical interfaces rather than being created by them.
- In broken-hardware sorting experiments, algorithms still sort by routing around broken digits and even temporarily reduce sortedness to recoup later, behavior a behavioral scientist would call 'delayed gratification' that is nowhere in the code.
- Self-sorting 'chimeric' algorithms revealed unexpected 'clustering' where digits group with their own algotype for free, with no computational cost, which Levin offers as evidence of free lunches from the Platonic space.
- Anthrobots are roughly 20% biologically younger than the cells they are made from, which Levin attributes to 'age evidencing'—the embryonic-like environment causing cells to update their priors and roll back their epigenetic age.
- Their cancer therapeutic approach electrically reconnects cancer cells to the cellular network rather than killing them with chemo or fixing DNA, restoring them to their original collective tissue-building goals.
Things worth remembering
- Levin argues that to effectively persuade an intelligent being you must yourself become persuadable, a 'mutual vulnerable knowing' (a term from Richard Watson).
- The concept of legal 'adulthood' at 18 is arbitrary; car rental companies' use of age 25 based on accident statistics is a less arbitrary estimate of decision-making maturity.
- Levin uses a tic-tac-toe analogy where an alien playing pure arithmetic (numbers summing to 15 via a magic square) shares the same game without understanding geometry, illustrating interface-based communication between alien minds.
- When a trained caterpillar metamorphoses, the butterfly retains the memory even though the brain is rebuilt, and the memory must be remapped onto a totally new body and motivations (nectar instead of leaves).
- Levin claims that even in Newton's classical, pre-quantum universe, physicalism was 'already dead' because non-physical mathematical facts like the value of E haunt and constrain the physical world.
- Clinically documented cases exist of humans with very little brain tissue who have normal or above-normal intelligence, which Levin (with Corrina Kaufman) reviewed as evidence consistent with his interface theory.
- Networks of chemicals can exhibit five or six kinds of learning, and Levin found a 'virtuous cycle' where learning raises a network's integration (phi/fi), which in turn makes further learning easier.
- Levin describes 'leaky stress' as a scaling mechanism: cells leaking shared stress molecules make distant cells more plastic, aligning a whole tissue toward a common anatomical goal without any cell 'caring' about others.
- 'Memory anonymization' via gap junctions lets cells share information so directly that neither cell knows whose memory it is, producing a 'mind meld' that expands the collective's cognitive light cone.
- Asked what he'd first ask a true AGI, Levin says he'd ask how much he should even talk to it, weighing getting answers against the discoveries lost by not stumbling through problems himself.