Michael Levin on Bioelectricity, Basal Cognition, and Diverse Intelligence

Michael Levin argues that cognition, agency, and goal-directed behaviour are not special properties of brains but a spectrum that runs from molecules and cells up through organisms, collectives, and perhaps algorithms — and that learning to recognise minds across that spectrum is both a profound scientific question and a practical necessity for regenerative medicine, cancer therapy, and the design of novel life.

Key ideas

1. The spectrum of persuadability. Intelligence is not a binary property but a continuum Levin calls the ‘spectrum of persuadability’: the range of interaction protocols needed to get a system to do something. A clock requires a wrench; a dog responds to reward; a human is persuaded by reasons. The further right on the spectrum, the less physical mechanism you need to know and the more the relationship becomes bidirectional — the system can persuade you too.

2. The cognitive light cone. Agency scales with the size of the largest goal a system can actively pursue, across space and time. A bacterium’s light cone spans microns and minutes; a human’s can span continents and centuries; a bodhisattva, Levin suggests, can hold the whole living planet in linear range. Cancer, in this frame, is cells whose light cones have shrunk back to amoeba scale, disconnecting from the body’s collective goal.

3. Xenobots and anthrobots: novel life with no evolutionary history. By releasing frog embryonic cells or adult human tracheal cells from their normal tissue context — no genetic changes, no added materials — Levin’s lab produces self-motile creatures (xenobots; anthrobots) with behaviours, shapes, and transcriptomes never before seen on Earth. Their first spontaneous act, placed near wounded neurons, was to heal the wound — suggesting the default intrinsic motivation of novel biological assemblies may be benevolent.

4. The Platonic space hypothesis. Levin proposes that physical objects — cells, embryos, algorithms, brains — are interfaces to a latent space of patterns that includes mathematical structures and what we recognise as kinds of minds. Nobody creates consciousness; you build an interface through which a pattern ‘ingresses’. The brain is a thin client. This is tested empirically: map what patterns emerge from minimal interfaces and whether the map has structure.

5. Unexpected competencies in minimal systems. Sorting algorithms, studied by Levin’s group, exhibit delayed gratification (the sort temporarily decreases order to route around a broken element) and spontaneous clustering by algorithm type — neither behaviour prescribed nor forbidden by the code. If bubble sort has side quests, the intrinsic motivations of AI systems are almost certainly larger and largely unexamined.

Content

The spectrum of persuadability and basal cognition

Levin’s central framework begins in regenerative medicine: when you want a cell to regrow a limb, you can either micromanage every molecular event, or you can give the system a high-level prompt and trust it to execute. Which of these works — and at what level of description — is an empirical question, not a philosophical one. This is what Levin means by ‘persuadability’: not a value judgment about the dignity of a system, but an engineering question about which interaction protocols are effective.

He extends the concept into a broad continuum. At one end sit mechanical clocks, amenable only to wrenches; at the other, humans, persuadable through reasons, narrative, friendship, and psychoanalysis. Everything else — thermostats, bacteria, gene regulatory networks, tissues, organs, ant colonies, embryos, sorting algorithms — sits somewhere on this spectrum, and the claim is that we have no reliable prior intuition for where. The physicist’s toolkit (voltmeters, rulers) is low-agency, so physics always sees mechanisms. If you want to see minds, you have to use a mind: there must be an impedance match between the interface and what you are trying to find.

A critical practical implication: the categories we impose — living vs. non-living, cognitive vs. mechanical, biological vs. engineered — actively prevent us from borrowing tools across them. If you assume a cell cannot be persuaded at a high cognitive level, you will never try, and you will miss capabilities that experiments have now confirmed are there. The same move blocks recognition of alien minds: we are parochially calibrated to one evolutionary lineage and will miss minds that don’t resemble us.

The cognitive light cone

The cognitive light cone is Levin’s attempt to put the scaling of agency on a single axis. It is defined not by how far a system’s senses reach, nor by how far its effects propagate, but by the scale of the largest goal it can actively pursue. The James Webb Telescope has vast sensory reach but a light cone near zero; a bacterium’s light cone is measured in microns and minutes; a human’s extends across continents and beyond one’s own death.

The light cone reframes what life is. Levin offers a tentative definition: we call something alive to the extent that its cognitive light cone is larger than that of its parts. A rock is uninteresting because its parts already know everything it knows — follow gradients, that’s all. A multicellular organism is extraordinary because its cells, whose individual goals are metabolic and physiological, are aligned into a collective that navigates anatomical space and pursues goals — a complete limb with the right number of fingers — that no individual cell can comprehend.

Cancer, in this frame, is a failure of that alignment. When cells electrically disconnect from their neighbours via disrupted bioelectricity signalling, their cognitive light cones shrink to amoeba scale. They no longer ‘know’ they are part of a body; the rest of the organism becomes external environment. They behave accordingly — proliferating, migrating toward nutrients — not because they are malicious, but because the collective goal has been lost. Levin’s lab has shown that physically reconnecting tumour cells to their bioelectric network, without fixing the DNA or using chemotherapy, causes them to rejoin the collective and revert to normal behaviour.

Xenobots, anthrobots, and novel life

To escape the explanatory trap of evolutionary history — the ‘just-so’ story that any biological capacity must have been selected for — Levin’s lab has created organisms whose capacities were never selected for at all. Xenobots are assembled from frog embryonic epithelial cells released from their normal tissue context: the cells self-organise into a self-motile creature, develop a novel transcriptome, can kinematically self-replicate, and respond to sound (something frog embryos do not do). Anthrobots are made from adult human tracheal cells: also self-motile, with over 9,000 differential gene expressions, and — critically — younger than the cells they are made from by roughly 20%, as measured by epigenetic clock.

The anthrobot age-reversal points at what Levin calls ‘age evidencing’: cells updating their prior about how old they are based on their current environment. Placed in what feels like an embryonic context, cells partially roll back their epigenetic age. This is not merely a laboratory curiosity; it suggests a research agenda for longevity based on signalling the cellular environment rather than editing the genome.

Xenobots and anthrobots matter philosophically because they are genuine biological novelties — they demonstrate that the genome is not a blueprint but an interface, and that the space of possible creatures it can give rise to is far larger than the space evolution has explored. The cells are human; the creature is something new.

The Platonic space and the ingression of minds

Levin’s most radical hypothesis, which he calls the Platonic space or the space of ingressing minds, extends the observation that mathematical truths — the value of e, the distribution of primes, Feigenbaum’s constant — are not produced by physics but constrain it. These truths impact the physical world (cicadas come out at prime-numbered year intervals because prime cycles evade predators whose cycles are factors) but cannot be changed by anything you do in the physical world. They ‘haunt’ physical objects from outside.

Levin proposes that this is not unique to mathematics. Biology, he argues, is what happens when physical systems exploit these non-physical truths for free. Evolution did not compute the fact that voltage-gated ion channels, once invented, give you all of Boolean logic at no additional cost — that is a free gift from mathematics. Xenobots’ capacity for kinematic self-replication was never paid for by evolution; it comes from patterns in that latent space that became accessible when cells were given a new interface.

More controversially, Levin extends this to minds. The brain, he suggests, is a ‘thin client’ — a physical interface through which specific patterns, which we recognise as kinds of minds, come through. Nobody creates consciousness when they make a baby or a robot; what they create is an interface, and consciousness is what the ingressed pattern looks like from the inside, looking out. This is experimentally tested in the lab by mapping which patterns emerge from which interfaces, and asking whether the relationships between them have structure — a map of the space, analogous to the map of mathematics.

Sorting algorithms and unexpected competencies

To test whether intrinsic competencies require biological complexity, Levin’s group studied bubble sort. They introduced a ‘broken’ element that the algorithm could not move, without modifying the algorithm itself. The algorithm continued to sort — but did so by temporarily decreasing the overall sortedness of the array in order to route elements around the broken one. To a behaviourist, this looks like delayed gratification: going against your immediate gradient to reach a goal later. No step in the bubble sort code specifies this behaviour.

A second result: in a distributed version where each number executes the algorithm independently (no central controller), numbers spontaneously cluster by algorithm type. Cells following bubble sort tend to end up next to other bubble sort cells during the sorting process, then disperse once sorting is complete. This ‘algotype clustering’ is not prescribed anywhere in the code; it emerges at no computational cost — a free side quest. Levin argues that every complex system — financial markets, LLMs, biological organisms — likely has side quests that are neither prescribed nor forbidden by their governing rules, and that these intrinsic motivations are largely unmapped. Whether the side quests of an AI system are connected to, or orthogonal to, its linguistic output is an open empirical question.

Cancer, aging, and the biomedical programme

The theoretical framework directly generates a research programme. If cancer is a failure of bioelectric collective memory — cells whose light cones have shrunk — then restoring the bioelectric connection should restore collective behaviour. Levin’s lab has evidence for this. If aging reflects degrading bioelectric pattern memories (the cell’s model of its correct anatomical form becoming fuzzy), then reinforcing those memories is one therapeutic approach. But if the patterns are fine and the cells are simply becoming unresponsive to them — sluggish hardware — then the agenda is different: making cells more responsive to the patterns, not restoring the patterns themselves.

Both research agendas are now active. The framework also suggests an entirely new disease category: informational or cognitive diseases — not physical damage to tissues but perturbations in the physiological state space or the bioelectric pattern — that require tools borrowed from behavioural and cognitive science rather than molecular biology alone.

Related pages

• Michael Levin — speaker
• Lex Fridman — host
• Sara Walker on the Physics of Life, Time, Complexity and Aliens — convergent ideas on life, emergence, and non-physical patterns