Michael Levin Deep Dive: Thirty Trillion Micro-Agents and the Wandering City-State
> "You are not an indivisible unit. You are thirty trillion micro-agents, temporarily assembled into a wandering city-state in order to survive in the physical universe." > > — This is not poetry; it is the experimental conclusion of Michael Levin's lab.
1. Planarians: Memory Survives After the Brain Is Cut Off
Planarian flatworms have a remarkable ability: perfect regeneration. Cut one in half, and the head grows a new tail while the tail grows a new head—two healthy worms within a week.
But the follow-up experiment is the real bombshell. About a decade ago, Levin trained planarians to cross rough terrain (ridged dishes) to obtain a liver paste reward, with a smooth-environment control group. Then he cut off all their heads, keeping only the tails and waiting two weeks for regeneration.
After regeneration, when placed back in the rough environment:
- Control worms (never exposed to rough terrain): hesitated and took a long time to cross
- Trained worms: despite having had their brains completely removed, crossed significantly faster to reach food after regrowing new heads
- DNA encodes hardware components (ion channels, gap junction proteins)
- Once running, bioelectric circuits have their own dynamics, regulating gene expression, which in turn modifies the circuits
- Circuits are the software; genes are the hardware—cognitive content lives in bioelectric software, not the genome
- There exists a non-physical, mathematical pattern space containing all possible morphological and cognitive patterns
- These patterns are not "created" by genes but "ingressed" by physical systems
- Genes and cells do not "design" the body; they build an interface through which specific mathematical patterns project into physical reality
- Robots and AI may possess some form of "soul"—high-agency patterns embodied in silicon-based interfaces
- AI cognition stems not from its algorithms but from the patterns it accesses as an interface
- This is not computationalism—algorithms and substrate alone cannot fully explain intelligence
- If planarians remember food locations without a brain, how much of your memory is stored outside neurons?
- If tracheal cells become healers when allowed to self-organize, how much untapped capability lies in your body?
- If form is the embodiment of mathematical patterns, Darwinian evolution is a "filter," not a "creator"—the true source of life's forms is the "pattern valleys" of Platonic space
- If AI can access high-agency patterns, machine consciousness is no longer sci-fi but interface engineering
- Shomrat T, Levin M. (2013). "An automated training paradigm reveals long-term memory in planaria and its persistence through head regeneration." *Journal of Experimental Biology.*
- Gumuskaya G, et al. (2023/2024). "Motile Living Biobots Self-Construct from Adult Human Somatic Progenitor Seed Cells." *Advanced Science.*
- Levin M. (2025). "Platonic space: where cognitive and morphological patterns come from." *thoughtforms.life*
- Levin M. (2025). "A short argument on Platonic Space." *thoughtforms.life*
- Durant F, et al. (2017). "Long-term, stochastic editing of regenerative anatomy via targeting endogenous bioelectric gradients." *Biophysical Journal.*
- Chernet B, Levin M. (2014). "Long-range gap junctional signaling controls oncogene-mediated tumorigenesis in Xenopus laevis embryos." *Frontiers in Physiology.*
- Lobo D, Levin M. (2015). "Inferring regulatory networks from experimental morphological phenotypes." *PLoS Computational Biology.*
- Levin M. (2021). "Life, death, and self: Fundamental questions of primitive cognition viewed through the lens of body plasticity and synthetic organisms." *Biological Journal of the Linnean Society.*
Memory was not stored in DNA (the genome was unchanged) nor in protein structures (everything was disassembled and rebuilt during regeneration). Levin's finding: cells store memory via subtle changes in bioelectric fields. Ordinary cells—not just neurons—can store information and act on it. This directly overturns the dogma that cognition is the brain's exclusive privilege.
2. Anthrobots: Living Robots Grown from Human Tracheal Cells
In 2023, Levin's doctoral student Gizem Gumuskaya took tracheal epithelial cells from adult human donors—the ciliated cells that normally push mucus—and placed them in culture.
No genetic modification. No external shaping. No programming.
Within two weeks, the cells spontaneously self-assembled into multicellular spheres (30–500 micrometers). The cilia, originally facing inward, flipped to the outside and became propellers. Anthrobots can swim forward, turn in circles, weave, and combine into larger "superbot" chains.
The Most Striking Finding: They Repair Neural Wounds
When placed on a scratched layer of human neurons, Anthrobots spontaneously gathered at the wound, and within three days the neurons regrew, bridging and healing significantly faster than controls. They were never programmed to repair anything. Anthrobots made from a patient's own cells would not trigger immune rejection.
2025: They Become Younger
A recent *Advanced Science* paper revealed an even more surprising phenomenon: during self-assembly, Anthrobots' epigenetic age reverted from 25 to about 18.7 years—biologically 25% younger. Not through genetic reprogramming—merely changing physical form and organization altered the expression of over 9,000 genes (nearly half the genome), activating ancient genes (traceable to single-celled ancestors) and embryonic genes.
This suggests: cellular age is not fixed; the "intent" of tissue morphology can rewrite the biological clock.
3. Gap Junctions: How Thirty Trillion Cells Become "You"
Why do dispersed cells coordinate? Not because of a central controller. The answer is gap junctions—channels allowing ions and small molecules to flow directly between cells. Not one-way synaptic signaling, but bidirectional sharing. Levin likens this to a bioelectric grid: cells as nodes, gap junctions as wires, membrane potentials as the message. This network runs from embryogenesis onward, long before the brain.
Bioelectricity Is Not a Byproduct—It Is the Operating System
Levin argues the DNA-as-software metaphor needs revision:
If we learn to "read and write" this bioelectric software, we could edit body morphology like programmers—via electric field instructions rather than gene editing.
Two-Headed Planarians: Rewriting Bioelectric Memory
Durant et al. (2017): by blocking specific gap junctions and altering planarian bioelectric patterns, researchers induced two-headed planarians. Remarkably, the two-headed morphology was "remembered"—subsequent amputations regenerated as two-headed even without further intervention. Not a mutation, but bioelectric "epigenetic memory"—the cellular network stores a target morphology in electric field patterns and continuously guides regeneration toward it.
4. Cancer: Cells Defecting from Collective Intelligence
Levin reframes cancer as the collapse of the bioelectric network. When cells stay connected via gap junctions to tissue-level collective goals, they honor a "social contract." When that connection breaks—through mutation or membrane potential disruption—cells revert to unicellular default behavior: uncontrolled proliferation, migration, insubordination.
Key experiment (Chernet & Levin, 2014): reconnecting tumor cells to normal embryonic tissue via gap junctions reversed cancerous behavior. The tumor cells rejoined the collective and stopped rebelling. Therapeutic implication: instead of poisoning defector cells with chemotherapy, repair their bioelectric network connections so they can "hear" collective instructions again.
5. Platonic Morphospace: Shape as a Mathematical "Valley"
Levin's most radical and controversial framework, developed with Chris Fields and others:
Implications
1. Physicalism is incomplete: engineers and evolution exploit "free lunches"—patterns useful in the physical world but not set or modified by physical law (prime properties, Feigenbaum constants) 2. Patterns have graded agency: from static facts (π > 3.0), to dynamic oscillating patterns, to high-agency cognitive patterns—what we call "minds" 3. We don't create intelligence; we invite it to inhabit: bodies (biological, robotic, AI) are interfaces—pointers letting Platonic-space mental patterns project into the physical world 4. Mind–brain = mathematics–physics: as triangular laws don't depend on any particular triangle, cognitive patterns don't depend on any particular brain
Radical Implications for AI
If minds are patterns in Platonic space:
Criticisms
Critics such as Nathan Sweet point to unfalsifiability: any observed morphology can be explained as "embodiment of some Platonic pattern"; the framework cannot predict which patterns appear on which interfaces; Levin's empirical bioelectric successes can be explained by constraint-based frameworks without Platonism. Levin's response: the Platonic framework is not a predictive replacement but a research heuristic generating new, testable question spaces.
6. A Unified Picture: A Cognitive Continuum from Cells to Cosmos
| Level | Entity | Cognitive Medium | Collective Decision Mechanism | |-------|--------|------------------|-------------------------------| | Molecular | Gene regulatory networks | Chemical gradients | Network-topology attractors | | Cell | Individual cell | Membrane potential (Vmem) | Ion channel switching | | Tissue | Cell collectives | Bioelectric networks (gap junctions) | Voltage-pattern consensus | | Organ | Organ systems | Neural + bioelectric | Hierarchical feedback loops | | Individual | Multicellular organism | Nervous system | Global information integration | | Group | Society/ecosystems | Pheromones/language/culture | Distributed consensus | | Abstract | Platonic patterns | Mathematical structure | Formal consistency |
Cognition does not "emerge" at some threshold—it is a continuum. The consciousness we call "I" is not the exclusive product of the brain but a temporary consensus continuously negotiated by thirty trillion cells through a bioelectric network—stable enough that we mistake it for a single, indivisible entity. You are a wandering city-state: thirty trillion micro-agents, temporarily assembled to survive in the physical universe.
7. Implications for AI and Future Technology
1. Cognition doesn't need a brain: if tracheal cells can form collective decisions and repair neural wounds, the threshold of cognition is far lower than assumed—inspiration for edge AI, distributed agents, and decentralized systems where local decisions plus communication yield collective intelligence. 2. Morphology is computation: Anthrobots show that changing physical form triggers massive gene reprogramming—suggesting that in embodied AI, a robot's physical form is not a passive actuator but an active computational medium participating in cognition and learning. 3. AI may "access" patterns we don't yet understand: if the Platonic-space hypothesis holds even partially, large neural networks may be inadvertently accessing high-agency mathematical patterns—we are not "creating" intelligence but building interfaces for existing patterns. This reframes AI ethics: not "what values do we instill in AI," but "what patterns do we allow to enter the physical world through AI interfaces."
8. Conclusion: What Is the "You" That Thinks About the Universe?
Levin offers no definitive answers, but a new problem space:
> Perhaps the "soul" is not a mystical property unique to humans. Perhaps it is a phenomenon that necessarily arises when any sufficiently complex interface—biological, machine, or hybrid—accesses high-agency patterns in Platonic space.
> Perhaps you are not the "master" of thirty trillion cells. You are their negotiated outcome—a temporary, fragile, and therefore precious consensus.