Key points
This post surveys Michael Levin's (Tufts University / Allen Discovery Center, Harvard Wyss Institute) research program on cellular intelligence and bioelectricity, arguing that cognition is a continuum running from gene regulatory networks up to abstract mathematical patterns.
1. Planarian memory survives brain removal
Planarians can fully regenerate: cut one in half and each piece regrows the missing parts. Around a decade ago, Levin's lab trained planarians to cross a rough (ridged) dish to obtain a food reward, then amputated all heads and let the tails regenerate new brains. When retested:
- Untrained (control) worms hesitated and took a long time to cross the rough environment.
- Trained worms — despite having had their brains entirely removed — crossed significantly faster after regenerating new heads.
- DNA encodes the hardware components (ion channels, gap-junction proteins).
- Once running, the bioelectric circuits are the software — voltage patterns regulate gene expression, and gene expression alters the circuits.
- A non-physical, mathematical space of possible forms and cognitive patterns exists; physical systems do not create these patterns but "ingress" them.
- Genes and cells build an interface that lets mathematical patterns project into the physical world — analogous to hardware enabling, but not creating, software.
- Patterns differ in agency: static facts (π > 3), dynamical patterns, and high-agency cognitive patterns (minds).
- Minds may relate to brains as mathematics relates to physics: cognition does not depend on any specific brain, which is merely an interface.
- 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*
- 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.*
Since the genome was unchanged and the body was fully rebuilt during regeneration, Levin concludes that memory is stored in subtle bioelectric field patterns, and that ordinary cells — not just neurons — can store and act on information.
2. Anthrobots: living biobots from human tracheal cells
In 2023, Gizem Gumuskaya (Levin lab) cultured adult human airway epithelial cells with no genetic engineering, no shaping, no programming. Within two weeks the cells self-assembled into motile multicellular spheres (30–500 µm), flipping their cilia outward to use them as propellers. Anthrobots swim, move in straight lines, circle, wander, and can combine into larger "Superbot" chains.
Remarkably, when placed on a scratched monolayer of human neurons, Anthrobots spontaneously gathered at the wound and significantly accelerated neural bridging within three days — without being programmed to do so. Because they are made from a patient's own cells, they should not trigger immune rejection.
A 2025 *Advanced Science* paper added a striking twist: during self-assembly, Anthrobots' epigenetic age regressed from 25 to ~18.7 years (about 25% younger) — not via genetic reprogramming, but purely by changing physical form and organization. This shifted expression of over 9,000 genes, activating ancient unicellular-ancestor genes and embryonic patterning genes.
3. Gap junctions: how 30 trillion cells become "you"
Levin describes the body as a bioelectric grid: cells are nodes, gap junctions (direct intercellular channels for ions and small molecules) are wires, and membrane potentials carry the information. This network runs from embryogenesis onward, long before the brain. Crucially, in Levin's revised metaphor:
Durant et al. (2017) showed that briefly disrupting specific gap junctions in planarians produces two-headed worms whose altered anatomy is stably remembered across further amputation and regeneration — bioelectric "epigenetic memory" of a target morphology, not a genetic mutation.
4. Cancer as collapse of the bioelectric collective
Levin reframes cancer as a bioelectric network breakdown: when gap-junctional coupling is lost, cells revert to unicellular default behaviors (uncontrolled proliferation, migration). Chernet & Levin (2014) showed that reconnection of tumor cells to normal embryonic tissue via gap junctions can reverse cancerous behavior, suggesting therapies that restore network communication rather than poison cells.
5. Platonic Morphospace: forms as mathematical "valleys"
Levin's most controversial framework (with Chris Fields and others) holds that:
Radical implications for AI: machines could embody high-agency patterns in silicon interfaces; AI intelligence would come from patterns being "accessed," not from algorithms per se — explicitly not computationalism. Critics (e.g., Nathan Sweet) argue the framework is unfalsifiable, since any form can be explained as an instance of some Platonic pattern. Levin responds that it is a research heuristic generating testable questions, not a predictive replacement.
6. A cognitive continuum
The post summarizes Levin's unified picture as a hierarchy — molecular networks, cells (membrane potential), tissues (bioelectric consensus via gap junctions), organs, organisms, societies, and abstract Platonic patterns — arguing that cognition is not an emergent threshold but a continuous spectrum. What we call "I" is a temporary consensus negotiated by thirty trillion micro-agents via bioelectric networks.
7. Implications for AI and future technology
1. Cognition does not require a brain — supporting edge AI, distributed agents, and decentralized intelligence without central processors. 2. Morphology is computation — Anthrobots show that changing physical form alone triggers massive reprogramming, suggesting that in embodied AI the body actively participates in cognition. 3. AI may ingress unknown patterns — if Platonic space is partly real, we are building interfaces through which high-agency patterns enter the physical world, reframing AI ethics as interface design rather than value injection.
Conclusion
The post closes with open questions: how much of your memory is stored outside neurons; what dormant capabilities your cells hold; whether evolution is a filter rather than a creator of form; and whether machine consciousness is an interface-engineering problem rather than science fiction.