Bioelectricity and How Cells Compute Without Neurons

Every cell holds a voltage across its membrane. In neurons that voltage carries signals in milliseconds, which is why it was studied there first. The same voltage exists in cells that have no neurons at all, where it changes over minutes and hours, and it carries a different kind of information. It tells groups of cells what to build.

That is developmental bioelectricity, and the reason it appears on a site about consciousness is that it breaks the assumption connecting minds to brains. Michael Levin’s laboratory has shown that altering a voltage pattern for a few hours can change what a flatworm regenerates, and that the altered pattern persists after the intervention stops. This section collects that work, the people doing it, and the open-source tools for modelling it. The flagship overview of the field’s attempt to define consciousness is in the race to define AI consciousness; the specific case for bioelectric substrates runs through Levin’s bioelectric cognition and the cognitive light cone and the mapping of consciousness theories onto unconventional embodiments.

Bioelectricity at a glance

Why this section belongs on this site

This site treats consciousness as an emergent property that does not depend on a particular material. Developmental bioelectricity is the sharpest available test of that claim, because it is the one case where goal-directed behaviour is measured in tissue that has no nervous system.

The argument runs in four steps, and it is worth stating precisely, because the field is often summarized in a way its own researchers reject.

1. Bioelectric signalling is older than neurons. Ion channels and gap junctions predate nervous systems. Neurons did not invent electrical signalling; they made an existing system fast.
2. Cognition therefore scales continuously. Levin’s framework commits to gradualism. No sharp line separates problem-solving in tissue from physics, and the selves it describes are nested and overlapping. The cognitive light cone measures the scale of goals a system can pursue, by deliberate analogy with special relativity.
3. Sentience is tied to that, by proposal rather than proof. Levin suggests animal sentience can be understood as an instance of collective intelligence among cell groups, and argues it is likely impossible to understand sentience without first understanding cognition.
4. So bioelectricity tests substrate independence directly. If the major theories of consciousness are about integration, recurrence or global broadcast, those properties can be looked for in non-neural tissue. Rouleau and Levin did exactly that and found that none of the main theories names neurons in its own definitions.

One clarification that the primary source insists on. Levin’s framework states that its focus is on cognitive function, not on phenomenal or access consciousness, and it cites Chalmers when naming the hard problem. The work shows goal-directed competence in unexpected places. It does not claim to have explained experience. Anyone summarizing it as a theory of consciousness is reporting something the authors did not say.

What the experiments show

Each of these is a published result, not an interpretation.

• A flatworm can be made to regenerate a second head by changing its bioelectric state for a few hours, and the change persists through later rounds of regeneration. This is the clearest demonstration that the voltage pattern, rather than the genome, carries the instruction for which shape to build.
• Learned behaviour survives the loss of the brain. Shomrat and Levin trained planaria, removed their heads, and found the training still present after regeneration.
• Brain defects can be repaired from a distance. Misexpressing a single ion channel in tissue far from the developing brain restored anatomy, gene expression and learning in frog embryos exposed to nicotine, and small-molecule activators of the same channel produced a similar rescue.
• Cells reassemble into things that are not embryos. Xenobots, built from frog skin cells, and anthrobots, built from adult human airway cells, move and behave without being the organism their genome encodes.

The bioelectricity scenes on this site

Cells, signals and cognition without neurons

• Bioelectricity SignalsArthur Reber's Cellular Basis of Consciousness and the Attack on AI ConsciousnessArthur S. Reber, William B. Miller Jr., František BaluškaSep 5, 2026
• Bioelectricity SignalsPlant Cognition and the Boundaries of Intelligence. Lessons for AI SubstratesPaco CalvoJul 1, 2026

Pattern memory and regeneration

• Bioelectricity MemoryMichael Levin's Bioelectric Memory Research From Planaria to XenobotsMichael Levin, Tal Shomrat, Douglas Blackiston, Vaibhav PaiSep 5, 2026

Scaling from cells to minds

• Bioelectricity MindsLarge Language Models as a Cognitive Virus in Solé and Levin's ModelRicard Solé, Michael Levin, Giulio Ruffini, David KrakauerOct 17, 2026
• Bioelectricity MindsMichael Levin Bioelectric Cognition and the Cognitive Light Cone as an AI Consciousness CriterionMichael Levin, Pamela Lyon, Fred KeijzerAug 14, 2026
• Bioelectricity MindsConsciousness as Symmetry-Breaking Safron and Levin's View of Generative ModelsMichael Levin, Adam SafronJun 28, 2026
• Bioelectricity MindsBrains and Where Else Why Major Consciousness Theories Do Not Require Neural TissueMichael LevinMay 15, 2026

The researchers

• Michael LevinAllen Discovery Center, Tufts UniversityBioelectric signalling, the cognitive light cone, and diverse intelligence Functionalist12 articles
• VP Vaibhav PaiTufts University (Levin laboratory)Memory mechanisms in synthetic living constructs, bioelectric pattern control, xenobot lineage research Biological naturalist1 article
• TS Tal ShomratTufts University (Levin laboratory)Automated planarian training, memory persistence through head regeneration, in vivo learning instrumentation Biological naturalist1 article
• DB Douglas BlackistonAllen Discovery Center at Tufts UniversityMemory stability during brain remodeling, xenobot platform design, amphibian memory transplantation work Biological naturalist1 article
• FP Federico PigozziTufts University (Levin laboratory)The causally emergent alignment hypothesis, causal emergence in gene regulatory networks, AI-guided memory resetting, causal architecture before self-replicators Integrated information2 articles
• PL Pamela LyonFlinders UniversityBasal cognition, the biogenic approach to mind, valence and affect in organisms without brains, Stress-Induced Senescence theory Biological naturalist1 article
• FK Fred KeijzerUniversity of GroningenBasal cognition in bacteria, early nervous systems as action systems, moving and sensing without input-output decomposition Biological naturalist1 article
• PC Paco CalvoUniversity of MurciaPlant cognition and plant neurobiology, minimal cognition without neurons, Planta Sapiens, cognitive ecology of plants Biological naturalist1 article
• FB František BaluškaUniversity of BonnCellular consciousness, subcellular nanobrains, plant neurobiology, sentient cells as units of physiology Biological naturalist1 article
• RS Ricard SoléICREA, Universitat Pompeu FabraComplex systems theory of cognition across cells, brains and machines, phase transitions in living systems, cognitive graphs, language origins Biological naturalist1 article

They sit under different positions in the researcher index, which is itself informative. Levin is indexed as a functionalist, most of his laboratory as biological naturalists. The disagreement about what their results imply for consciousness runs through the field, not around it.

Open-source tools

Three pieces of software do most of the modelling in this field, and all of them are free.

• BETSE, from Alexis Pietak and Michael Levin, simulates bioelectric tissue with ion channels, pumps, gap junctions and extracellular space. It is the reference implementation.
• BioElectricNetwork and ElectricMorphogenesis, from Santosh Manicka and Michael Levin, model minimal bioelectric networks and tissue-level voltage patterning.
• bioelectrics, built for this project, is a PyTorch engine that makes the physics differentiable. Because every quantity carries a gradient, it can answer the inverse question directly. Given a target voltage pattern, which realizable intervention produces it? A channel misexpression, a drug dose, a gap junction blocker, or a light pattern, printed as a protocol. It reproduces the Cervera, Alcaraz and Mafe model under test, finds attractors and their basins, and reads BETSE configurations so the two tools can be compared on the same parameters.

The code is at github.com/tlcdv/bioelectrics under an MIT licence, with documentation, four tutorials that run in the browser, and a validation page that recomputes every published number from the code on each build.

Where the field stands

The mechanism is established and the interpretation is contested. That a voltage pattern instructs anatomy is demonstrated in several species by several laboratories. What that implies about cognition is argued, and the argument is the interesting part.

Two reproductions this project attempted are instructive about the state of the literature. The equations in the 2019 minimal bioelectric network paper are dimensionally inconsistent as printed, and the parameters for the 2020 frog brain patterning model are in supplementary material the paper says is available from the corresponding author. Neither model can be reproduced from the public record alone. That is common in computational biology and it is worth saying plainly, because reproducibility is the standard the field applies to its own experimental claims.

A second finding came out of building the drug library. Of the compounds with a published potency on a target this physics models, every one acts on a polarizing element, so blocking it depolarizes cells rather than repairing a depolarized region. Nothing currently in that library normalizes an abnormal voltage domain. That is a statement about which measurements exist, not about what bioelectric medicine can do, and it points at a specific gap. The protein carrying the depolarizing current in a persistent abnormal domain has not been named in a way a modeller can use.

Related sections

The substrate-independence claim that this section tests is set out across Brain Emulation and Mind Uploading, which asks the same question about copying a known brain, and Project Consciousness to Orbit, which asks where a mind can go. The theories being tested are indexed in Theories of Consciousness. For the measurement problem that all of them run into, the Phi Lab Console computes integrated information for every wiring of three logic gates, and finds that every feed-forward one scores zero.