Overview: Who This Is
Michael Levin is a developmental and synthetic biologist whose research asks a deceptively simple question: how do cells know what to build?
A fertilized egg becomes an animal with organs in the correct places. A flatworm can rebuild a missing head. Damaged tissues often correct toward a familiar anatomy, even when their starting conditions have been disturbed. Genes supply essential molecular components, but a list of components does not by itself explain how thousands or millions of cells coordinate toward a larger form, detect when that form has been disrupted, and sometimes stop when repair is complete.
Levin’s laboratory studies one layer of that coordination: developmental bioelectricity. Cells maintain electrical differences across their membranes through ion channels and pumps. Through direct connections called gap junctions, groups of cells can couple those voltage states into tissue-wide patterns. Levin and his collaborators investigate how those patterns influence gene expression, cell movement, proliferation, differentiation, organ placement, regeneration, and cancer-like behavior.
The established biology is already significant. The more expansive interpretation is where Levin becomes a Dossier subject. He argues that living tissues can be studied as problem-solving collectives: systems with memories, preferences, competencies, and limited goals in anatomical space. In this view, brains did not invent every basic feature of intelligence. Nervous systems may have accelerated and expanded capacities that evolution was already using when cells cooperated to build bodies.
That claim must be handled carefully. Bioelectric regulation is not proof that a cell thinks like a person. Goal-directed repair is not proof of subjective experience. Xenobots and anthrobots are not miniature conscious creatures. Levin’s importance lies in forcing sharper distinctions among regulation, information processing, agency, cognition, intelligence, and consciousness and in turning some of those philosophical distinctions into experiments.
Origins and Background
Levin’s path into developmental biology began at the intersection of biology and computation. Tufts University lists dual bachelor’s degrees in computer science and biology, completed in 1992, followed by a doctorate in genetics from Harvard Medical School in 1996. His doctoral research in Clifford Tabin’s laboratory examined the molecular mechanisms by which embryos reliably establish left-right asymmetry: why hearts, lungs, and other internal structures develop with consistent orientation even though the large-scale laws of physics do not visibly favor left over right.
He completed postdoctoral training in cell biology at Harvard Medical School from 1996 to 2000, then established an independent laboratory at the Forsyth Institute. He moved his group to Tufts in 2008. As of 2026, Tufts identifies him as Vannevar Bush Distinguished Professor in the Department of Biology. He directs the Allen Discovery Center at Tufts and the Tufts Center for Regenerative and Developmental Biology, and he is associate faculty at Harvard’s Wyss Institute for Biologically Inspired Engineering.
The combination of computer science and developmental biology is not incidental to his work. Conventional molecular biology often explains development from the bottom up: genes produce proteins, proteins participate in pathways, and local interactions accumulate into anatomy. Levin does not reject that account. He asks what additional concepts are needed to explain the system-level reliability of the outcome.
An embryo is not assembled like a static machine from a single rigid plan. Cells divide, migrate, exchange signals, respond to injury, compensate for perturbations, and adapt to changing geometry. Different molecular routes can sometimes reach similar anatomical results. For Levin, this makes development resemble a distributed control problem. The genome helps specify the cellular hardware, but the real-time construction of an organism also depends on communication among active agents embedded in a changing environment.
Bioelectricity offered a way to study that control experimentally. This is not the dramatic electricity of lightning, shocks, or external power. Every living cell separates charged ions across its membrane. Ion channels, pumps, and transporters regulate that separation, producing a resting membrane voltage. Neurons use rapid changes in voltage to transmit signals, but non-neural cells also use slower electrical states and gradients. Those states interact with biochemical pathways and can affect what genes are expressed and what cells do.
Levin’s laboratory helped build tools for imaging and manipulating these voltage patterns in embryos, tumors, and regenerating tissues. The larger program became an attempt to read and write what the Allen Discovery Center calls the bioelectric control circuits of anatomical homeostasis: the processes through which a living system builds, repairs, and maintains its form.
What He Is Known For
Bioelectric Patterning
Levin is best known scientifically for demonstrating that membrane voltage is not merely a background consequence of cell metabolism. In many developmental contexts, it can carry instructive information.
In frog embryos, his group found that specific voltage patterns help mark the regions where eyes will form. Experimentally disrupting those states produced eye defects, while imposing relevant voltage states could induce eye tissue in unusual locations. The result did not mean that voltage alone constructed an eye. It showed that bioelectric state can act upstream of gene networks associated with organ formation and can help define where a complex structure should develop.
Other work examined craniofacial development, pigment cells, left-right patterning, tumor-like growth, and regeneration. Across these experiments, the recurring point is that cells respond not only to genetic instructions and chemical gradients, but also to electrical relationships distributed through tissue. Genes, proteins, mechanical forces, metabolism, and bioelectricity form an interacting system. Levin’s contribution has been to make the electrical layer experimentally visible and biologically consequential.
This corrects a common misunderstanding. His research does not show that genes are unimportant or that an invisible electrical field independently dictates anatomy. Ion channels and gap junctions are themselves gene products. Electrical states affect biochemical pathways, and biochemical pathways alter electrical states. The argument is about levels of control: the same genetic hardware can participate in different physiological patterns, and those patterns can redirect what the tissue builds.
Regeneration and Anatomical Memory
Planarian flatworms provide some of the clearest demonstrations. These animals can regenerate complete bodies from fragments, which raises a difficult question: how does a cut piece determine which structures are missing and when the correct body plan has been restored?
Levin’s group showed that manipulating ion flow and bioelectric communication can alter head size, organ scaling, and anterior-posterior polarity during regeneration. In one line of experiments, a brief disruption of gap-junction communication produced two-headed worms. More strikingly, some apparently normal worms retained an altered regenerative tendency: when cut again in ordinary water, a fraction produced two heads without a second exposure to the original treatment. The laboratory could reset that tendency by changing the bioelectric state.
Levin describes this as a form of anatomical memory—a persistent target morphology stored in physiological network states rather than in a changed DNA sequence. The phrase is powerful, but it should be read precisely. The experiment demonstrated durable, non-genetic information affecting future regeneration. It did not establish memory in the human autobiographical sense, nor did it identify a conscious image of the body inside the worm.
The importance is practical. Regenerative medicine cannot succeed merely by causing cells to multiply. New tissue must grow into the right structure, at the right scale, in the right place, and then stop. If bioelectric networks help encode or stabilize those large-scale outcomes, manipulating the network may eventually be more efficient than specifying every molecular step.
A related 2022 study from Levin and collaborators used a wearable bioreactor called a BioDome to deliver a five-drug cocktail for 24 hours to amputated hindlimbs of adult African clawed frogs, animals with limited adult limb regeneration. Treated frogs grew more complex, functionally useful appendages over the following 18 months than untreated controls, with extended bone, nerves, blood vessels, and digit-like projections. The structures were not perfect replacements for normal frog legs, and the experiment was not a human therapy. Its significance was that a brief intervention initiated a long process organized largely by the animal’s own tissues.
Basal Cognition and Collective Cellular Behavior
Levin places these findings inside a broader research program known as basal cognition: the study of cognitive-like capacities in organisms and systems without conventional brains. The field includes work on bacteria, single-celled organisms, plants, fungi, slime molds, tissues, and synthetic biological constructs.
His preferred framework is graded rather than binary. Instead of asking whether a system possesses “real intelligence” as an all-or-nothing property, he asks what goals it can pursue, across what spatial and temporal range, with how much flexibility, and through how many alternative routes. A cell maintaining pH has a narrow goal. A tissue restoring an organ has a wider one. An animal navigating an environment can pursue goals across still larger distances and times.
Levin calls the reachable scale of a system’s possible goals its cognitive light cone. The metaphor is meant to compare agents without assuming that intelligence must look human or even neural. In his Technological Approach to Mind Everywhere, or TAME, morphogenesis becomes problem-solving in anatomical space. Cells do not walk through a room to reach a destination; they alter position, identity, number, and connection to reach a viable form.
This is an interpretation built on real behavior, not a direct observation of a cellular inner life. Tissues can correct perturbations, store persistent physiological states, integrate signals, and reach similar anatomical outcomes from different starting points. Calling those capacities memory, decision-making, or intelligence may be scientifically fruitful if the vocabulary produces new predictions and better interventions. It becomes misleading if the terms quietly import human planning, self-awareness, or feeling.
Xenobots
Xenobots brought Levin’s work to a much wider public. First reported in 2020 by Sam Kriegman, Douglas Blackiston, Michael Levin, and Josh Bongard, these constructs were assembled from embryonic cells of the African clawed frog, Xenopus laevis. An evolutionary algorithm searched possible body shapes in simulation, and researchers manually built selected designs from living tissue. Some configurations moved through water and displaced particles in their environment.
A 2021 study described a more self-organizing version made from frog embryonic skin cells. Removed from their normal context, the cells healed into small spheroids. Cilia that would ordinarily help move mucus across a tadpole’s skin instead propelled the new collective through water. The constructs could move, recover from mechanical injury, and display group effects such as rearranging particles.
They were called “living robots,” a term that captures one genuine feature and risks overstating another. They are living, because they consist of metabolically active cells. They can be treated as robots in a broad engineering sense because their form can be designed or selected for tasks. But they do not contain electronics, do not think through a digital controller, and do not possess demonstrated independent intentions. Much of their behavior emerges from ordinary cellular abilities operating in an unfamiliar arrangement.
The 2021 report of kinematic self-replication produced even more dramatic headlines. In a controlled dish containing loose frog cells, moving xenobot aggregates could gather those cells into clusters. Some clusters matured into motile offspring capable of repeating the process for a limited number of generations, and computer-designed shapes improved its efficiency.
This was a novel form of physical replication, but it was not unrestricted biological reproduction. The xenobots did not grow offspring from internal resources, copy a genome into a developing embryo, or reproduce indefinitely in the wild. The experiment required a prepared environment containing a supply of dissociated cells. Its real importance was narrower and stranger: cellular material contained a reproducible behavior that neither frog development nor ordinary robotics had made obvious.
Anthrobots
Anthrobots extended the same question from amphibian embryonic tissue to adult human cells. In research first published in 2023, Gizem Gumuskaya, Levin, and colleagues grew individual progenitor cells from adult human tracheal epithelium into multicellular spheroids. The cells self-assembled without genetic modification or an inorganic scaffold. Surface cilia powered movement, and differences in anatomy correlated with different movement patterns.
The anthrobots ranged from roughly 30 to 500 micrometers and remained viable for weeks under laboratory conditions. When groups of them were placed on a scratched sheet of cultured human neurons, clusters sometimes formed bridges across the gap and were associated with closure of the damaged region. This was an in-vitro proof of concept—not evidence that anthrobots can repair a human nervous system inside a patient.
A 2025 follow-up characterized a distinct anthrobot life cycle, including changes in gene expression, self-repair, and reduced epigenetic age relative to their source cells. These findings strengthen the case that ordinary cells have wider developmental possibilities than their role inside the donor body suggests. They do not establish that anthrobots are a new naturally reproducing species, a human organism, or a conscious being.
Xenobots and anthrobots matter because they loosen the assumed relationship among genome, body, and behavior. The cells carry ordinary frog or human genomes, yet when released from their native anatomical constraints, they can assemble into forms and behaviors not found in the source organism. The genome provides competencies and constraints. It does not specify one inevitable body in every context.
The Core Idea or Signal
The central signal in Levin’s work is that intelligence may have a deeper biological history than brains.
Brains are extraordinary, but they are made from cells using molecular components that evolution developed long before nervous systems existed. Ion channels, electrical coupling, feedback, memory-like state dependence, error correction, and homeostasis were already available in primitive living systems. Levin proposes that nervous systems expanded this ancient architecture: cells that once coordinated what the body should become were reorganized into networks that could coordinate what the body should do.
The strongest version of this idea is not that every cell is a tiny person. It is that cognition may be a continuum of competencies rather than a substance that suddenly appears when a brain crosses an unknown threshold. A cell senses conditions and acts to preserve viability. A tissue coordinates many cells toward a larger anatomical state. A nervous system coordinates perception and action across space. Human minds extend goals across years, institutions, imagined futures, and symbolic worlds.
Bioelectricity is the bridge in Levin’s account. Neurons use electrical networks to bind cells into an agent capable of behavior. Non-neural tissues use related machinery to bind cells into a collective capable of development, repair, and physiological regulation. The domains differ, as do the speed, complexity, and richness of the capacities. The shared architecture suggests continuity rather than a clean metaphysical break.
This reframes the organism. The body is not simply matter commanded by a brain. It is a nested society of active systems operating at multiple scales. The brain is one powerful layer of coordination inside a much older intelligence of living form.
Perspectives and Interpretations
The Developmental-Biology View
From a conservative developmental-biology perspective, the most important part of Levin’s work does not require the language of mind. Cells communicate through well-established biochemical, mechanical, and electrical mechanisms. Membrane voltage affects signaling pathways and gene expression. Gap junctions allow ions and small molecules to move between neighboring cells. Feedback networks stabilize developmental outcomes.
On this view, Levin’s experimental results enrich the causal map of morphogenesis. Bioelectricity is an underappreciated regulatory layer, but phrases such as “anatomical memory” and “collective intelligence” may be optional metaphors. A tissue can return to a stable attractor without representing a future body or wanting to reach it. The appearance of purpose can arise from selection and feedback.
This perspective provides an essential guardrail. Mechanism should not disappear beneath cognitive language. Yet it may also understate the systems problem. Naming channels and pathways does not automatically explain how a living collective corrects novel disturbances, coordinates across scale, or reaches a robust outcome through more than one route. Levin’s response is that behavioral and cognitive concepts are useful precisely when they help describe and manipulate those higher-level capacities.
The Basal-Cognition View
Supporters of basal cognition argue that the brain-centered definition of cognition mistakes one highly evolved implementation for the entire category. Bacteria communicate, adapt, coordinate, and retain history-dependent states. Single cells navigate chemical and electrical gradients. Plants alter behavior in response to past conditions. Tissues solve patterning problems without a central executive.
The point is not that all of these systems possess human thought. It is that evolution is continuous. If memory, learning, preference, and goal pursuit exist only after a hard neural boundary, researchers must explain where that boundary lies and why precursor capacities do not count. A graded framework instead compares the range and flexibility of competencies across substrates.
Levin’s TAME framework makes this view technological. The correct level of agency is partly revealed by which intervention works. If micromanaging every cell is inefficient but changing a tissue-level setpoint reliably produces repair, the higher-level description may have causal and engineering value. Agency becomes an experimental stance: treat the system as a problem solver, perturb its goals or information, and see whether prediction and control improve.
The Anthropomorphism Critique
Critics worry that expanding cognitive terms too far can erase the distinctions those terms were meant to capture. If homeostasis becomes preference, feedback becomes memory, state transition becomes decision, and successful regulation becomes intelligence, then a thermostat begins to look cognitive by definition. The vocabulary may redescribe familiar mechanisms without adding explanatory power.
There is also an evidentiary problem. Some claims about learning in organisms without nervous systems have proved difficult to replicate or distinguish from sensitization, habituation, selection effects, or simple physicochemical dynamics. A 2021 review by Ignacio Loy and colleagues found significant limits in the evidence for associative learning in plants and protists. This does not invalidate basal cognition as a field, but it shows why each claimed capacity requires its own operational test.
The strongest criticism is therefore not “cells cannot be intelligent because they lack brains.” That would assume the conclusion. It is that words such as goal, representation, memory, and cognition must earn their place by outperforming lower-level descriptions: generating predictions, identifying interventions, or revealing regularities that would otherwise remain hidden.
Levin largely accepts that test. He argues that cognitive language should be judged by empirical usefulness rather than by whether it makes observers uncomfortable. The unresolved issue is where usefulness ends and inflation begins.
Goals, Agency, Cognition, and Consciousness
Public discussion often collapses four different claims.
A system can be goal-directed in the cybernetic sense if it acts to reduce the difference between a current state and a setpoint. A tissue that repairs toward a stable anatomy may qualify without possessing beliefs.
A system can display agency if treating it as a unit with preferences and capacities helps predict how it will respond across changing conditions. Agency can be graded and observer-relative without being imaginary.
A system can display cognition, under Levin’s broad definition, when it adaptively processes information and solves problems within some space of possibilities. Critics dispute whether this definition is too inclusive, but the claim is meant to be operational rather than mystical.
Consciousness is different. It concerns subjective experience: whether there is anything it feels like to be the system. Levin has explored the possibility that sentience may be realizable in unfamiliar substrates, but his major bioelectricity and basal-cognition arguments do not demonstrate that individual cells, regenerating tissues, xenobots, or anthrobots have inner experience. In his 2023 review of bioelectric networks, he explicitly set first-person consciousness aside to focus on observable capacities.
These distinctions protect both sides of the inquiry. They prevent demonstrated cellular coordination from being dismissed merely because it is not human thought. They also prevent biological competence from being promoted into consciousness without evidence.
The Synthetic-Life Interpretation
Xenobots and anthrobots invite a different interpretation: perhaps organisms are not fixed objects but possibility spaces. Cells evolved inside bodies, yet their competencies can be reorganized into novel living constructs. That makes synthetic biology less like assembling inert parts and more like negotiating with active material.
This view is both exciting and ethically demanding. A future living machine might heal, sense, deliver drugs, clear damaged tissue, or adapt to a patient in ways conventional devices cannot. But as constructs gain neurons, sensors, memory, longer lifespans, reproduction, or more flexible behavior, the question of moral status cannot be postponed indefinitely. The current xenobots and anthrobots provide no evidence of consciousness. They do, however, expose how poorly origin alone answers what a new living entity is.
Strengths and Limitations
Levin’s strongest work is experimental. His laboratory and collaborators have altered ion channels, membrane voltage, and gap-junction communication in living organisms and observed reproducible changes in organ formation, regeneration, scaling, polarity, and tumor-like behavior. These are not philosophical intuitions. They are interventions with measurable anatomical outcomes.
A second strength is the emphasis on control across scale. Molecular biology is exceptionally powerful at identifying components, but medicine ultimately cares about outcomes: a closed wound, a repaired organ, a normalized tumor, or a restored limb. Levin’s research asks whether clinicians might influence the higher-level patterning system rather than fabricate every detail. The frog-limb study, although preliminary and far from clinical translation, illustrates the appeal of triggering endogenous construction rather than manually engineering an appendage cell by cell.
Xenobots and anthrobots add another strength: they function as experiments on assumptions. By placing ordinary cells in novel arrangements, researchers can observe capacities that normal embryology keeps hidden. The work demonstrates developmental plasticity without requiring claims about consciousness or a new metaphysics of life.
The limitations are equally important. Much of the empirical work has been performed in planarians, frog embryos, adult frogs, cultured cells, or in-vitro constructs. These are powerful model systems, but human bodies are not scaled-up flatworms or tadpoles. The distance from changing a frog’s patterning signals to safely regenerating a human limb, correcting a birth defect, or normalizing cancer is enormous.
Bioelectric states also do not operate independently. Voltage patterns are entangled with genetics, transcription, chemical signaling, mechanics, immune activity, metabolism, and the extracellular environment. The metaphor of a “bioelectric code” can suggest a clean symbolic language with a simple decoder. Biology may instead contain context-dependent control dynamics that are only partly code-like.
The language of goals introduces another limitation. An observer must decide which outcome counts as the system’s target and which spatial or temporal boundary counts as the agent. A regenerating limb may look as though it pursues a final form, but different models can sometimes describe the same trajectory. Without explicit null models and measurable criteria, agency risks becoming a compelling story attached after the fact.
The synthetic constructs are also easy to oversell. Xenobots are not autonomous artificial organisms loose in the world. Their reported replication required laboratory conditions and supplied cells. Anthrobots have not repaired human patients. Neither construct has been shown to possess awareness. The terminology is defensible within robotics and synthetic biology, but public headlines often carry meanings far beyond the experiments.
Finally, Levin’s continuum of cognition does not solve the hard problem of consciousness. Showing continuity in information processing may explain how complex capacities evolved from simpler ones. It does not by itself explain why any physical process is accompanied by feeling. A map from homeostasis to intelligence is not yet a map from intelligence to experience.
Broader Implications
Medicine as Communication With Living Systems
The medical promise of Levin’s work is a shift from construction to persuasion. Most regenerative strategies focus on supplying cells, scaffolds, genes, or growth factors. His framework asks whether the body already contains much of the competence required to build the desired structure—and whether treatment can provide the right high-level signal.
If that approach succeeds, future medicine could target patterning setpoints. A clinician might instruct tissue to complete a missing structure, repair toward a healthy geometry, or rejoin organism-level goals after cells have become cancerous. Ion-channel drugs are especially interesting because many already exist, although using them to control anatomy safely would require far more precise understanding than current medicine possesses.
The cancer implication is conceptually striking. Cancer can be viewed not only as mutated cells proliferating, but as cells losing coordination with the larger anatomical collective. Levin’s animal studies suggest that altering bioelectric context can sometimes suppress tumor-like outcomes even in the presence of oncogenic signals. That does not replace genetic models of cancer or establish a cure. It suggests that malignancy may involve failures of multiscale communication as well as damaged molecular components.
Synthetic Life and the Meaning of an Organism
Xenobots and anthrobots make the word “organism” unstable. They are assembled or induced by researchers, yet much of their form and behavior arises from self-organization. They are made from familiar cells, yet they do not recapitulate the familiar animal. They are designed, but not in the same fully specified way as a mechanical device.
This creates a new class of entities between tissue, organoid, machine, and organism. Future versions could be made from a patient’s own cells, perform a temporary task, and biodegrade when finished. That could reduce immune rejection and open therapies inaccessible to rigid robotics. It could also produce living systems whose capacities change after deployment.
The ethical response should track capability rather than the emotional power of a name. Current non-neural anthrobots do not merit assumptions of consciousness merely because they are made from human cells. Conversely, a future construct should not be denied consideration merely because it was designed, grown in a dish, or lacks a familiar body. Levin’s work makes classification an empirical responsibility.
Artificial Intelligence Beyond the Brain Metaphor
Modern AI is often modeled after brains or described through disembodied computation. Levin’s work points toward another source of principles: development. Living systems do not only classify inputs. They build and repair themselves while operating under uncertain conditions. Their components possess local autonomy, yet collective constraints keep them aligned with larger goals.
This could inspire machines organized less like a central controller and more like a regenerative organism. Components might detect damage, reconfigure roles, repair structure, and maintain system-level outcomes without receiving detailed instructions for every contingency. The relevant lesson is not that cells secretly run software identical to ours. It is that robust intelligence may depend on multiscale coordination among competent parts.
The analogy also sharpens the AI alignment problem. In a healthy body, individual cells usually subordinate short-range aims to organism-level form. Cancer is one example of that relationship breaking down. Future artificial systems may likewise require architectures in which local optimization remains legible to and constrained by broader goals. Biology does not provide a finished solution, but it offers billions of years of experiments in keeping nested agents coherent.
Human Identity
Levin’s work challenges the intuitive boundary of the self. A human being feels like one mind inhabiting one body, yet the body consists of trillions of cells coordinating across multiple scales. No individual cell understands a life plan, but the collective maintains an anatomy, supports memory, and eventually produces a person capable of reflecting on the whole arrangement.
This does not prove that the body is a separate conscious mind beneath the brain. It suggests that human agency may be layered. What we call the self is not imposed on passive matter from above. It emerges from older systems of communication, cooperation, boundary-making, and error correction that remain active throughout the body.
If intelligence is fundamentally the ability to pursue goals across a space of possibilities, human cognition may be one expanded case of a more general property of organized life. The difference between cell and person remains vast. The continuity between them may be just as important.
The Reality Signal
What This Subject Represents
Michael Levin represents a change in where science is willing to look for intelligence. His work moves the inquiry below neurons and into the electrical, biochemical, and collective activity of living tissue. He represents the possibility that form is not merely produced by molecular machinery but actively maintained by systems capable of sensing deviation and coordinating repair.
More broadly, he represents a transition from viewing the body as a passive machine to viewing it as agential material: matter whose components already possess limited competencies and whose larger organization can create new levels of goal-directed behavior.
What Reality Frame It Challenges
Levin challenges the assumption that intelligence begins at the brain and that the rest of the body merely executes neural commands. He also challenges a rigid gene-centered picture in which anatomy is read directly from DNA as though development were the unfolding of a fixed blueprint.
His research does not overthrow genetics or prove that mind exists everywhere. It complicates the frame. Genes provide essential cellular machinery, while physiological networks, environmental conditions, mechanical forces, and bioelectric states help determine which large-scale form that machinery produces.
Why It Matters Now
The question has become urgent because humanity is constructing entities that no longer fit inherited categories. Organoids, xenobots, anthrobots, brain-computer hybrids, adaptive robots, and artificial intelligence systems occupy spaces between organism and machine, designed and evolved, tool and agent.
At the same time, medicine is approaching problems that cannot be solved by replacing single parts alone. Regeneration, cancer, aging, congenital malformation, and neural repair all involve coordination across scale. Levin’s work offers a language and experimental program for interacting with that coordination.
It also matters because AI has made behavioral intelligence visible outside conventional biology. We now need better ways to compare agents without assuming that human language, animal movement, neural tissue, or biological origin defines the entire category. Levin’s graded approach is one candidate framework, even if its boundaries remain disputed.
What Remains Unresolved
It is established that non-neural cells use bioelectric signals, that those signals can influence development and regeneration, and that manipulating them can change anatomical outcomes in multiple model systems. It is also established that frog and human cells can self-organize into novel motile constructs under laboratory conditions.
It remains debated whether cognitive vocabulary provides the best scientific explanation of these abilities or whether it anthropomorphizes regulatory dynamics. It remains unclear how bioelectric patterns encode specific forms, how transferable the results will be to human medicine, and how reliably high-level goals can be identified and rewritten.
There is no evidence that current xenobots or anthrobots are conscious. There is no accepted test for consciousness in radically unfamiliar substrates. The relationship among goal-directedness, agency, intelligence, and subjective experience remains one of the deepest unresolved boundaries in the entire field.
The Galactic Mind Perspective
Michael Levin belongs in The Galactic Mind archive because his work makes a familiar world strange without requiring us to abandon evidence.
The strange thing is not that cells emit a mysterious force. They do not. The strange thing is that ordinary cellular mechanisms—ion channels, voltage differences, gap junctions, feedback, and gene regulation—can participate in collective outcomes that look uncannily like memory, correction, and goal pursuit when observed at the scale of a whole tissue.
The careful response is neither to declare every cell conscious nor to reserve every cognitive word for the human brain by definition. Both moves close the inquiry too early. The better path is comparative: specify the capacity, design the perturbation, measure the response, and ask whether the language of agency improves prediction and control.
Levin’s deepest contribution may be less a conclusion than a method of recognition. Intelligence may not always announce itself through speech, planning, or movement across a visible environment. It may appear as a wounded system finding its way back to form. It may exist in degrees, distributed across levels, with goals whose scale determines the kind of agent we are observing.
Consciousness remains a separate door. A tissue can solve problems without proving that it feels. A machine can behave intelligently without establishing an inner world. The moral and scientific challenge ahead will be learning not to confuse competence with experience—while also refusing to assume that experience belongs only to forms already familiar to us.
The body may not contain one intelligence hidden beneath the brain. It may contain a layered history of intelligence from which the brain and eventually the conscious self—became possible.
Open Thread
If intelligence began as living systems learning to preserve and rebuild themselves, where should we draw the boundary between a body that regulates and a mind that knows what it is trying to become?
What do you think? Drop your thoughts in the comments ...
More in Dossier
- Michael Gazzaniga and the Interpreter Within — Gazzaniga examines how specialized neural systems become the experience of a unified self; Levin asks how coordinated agency may begin before nervous systems exist.
- David Chalmers and the Hard Problem of Reality — Chalmers clarifies why intelligent function and subjective experience cannot be treated as synonyms, a distinction essential to interpreting Levin’s cellular research.
- Donald Hoffman and the Case Against Naive Reality — Hoffman questions whether perception reveals reality directly, while Levin asks whether familiar human behavior has made us too narrow in recognizing intelligence elsewhere.
Sources / Receipts
- Tufts University Department of Biology — “Michael Levin.” Current institutional profile, education, research focus, and professional role.
https://as.tufts.edu/biology/people/faculty/michael-levin - Allen Discovery Center at Tufts University — “Michael Levin, Ph.D.” Institutional biography and research program.
https://allencenter.tufts.edu/our-team/michael-levin/ - Wyss Institute at Harvard University — “Michael Levin, Ph.D.” Current associate-faculty profile and institutional affiliations.
https://wyss.harvard.edu/team/associate-faculty/michael-levin-ph-d/ - Pietak, Alexis, and Michael Levin. “Bioelectrical control of positional information in development and regeneration: a review of conceptual and computational advances.” Progress in Biophysics and Molecular Biology (2018). Background on developmental bioelectricity and anatomical pattern regulation.
https://doi.org/10.1016/j.pbiomolbio.2018.03.008 - Pai, Vaibhav P., et al. “Transmembrane voltage potential controls embryonic eye patterning in Xenopus laevis.” Development 139 (2012): 313–323. Experimental evidence that voltage patterns regulate eye development.
https://pubmed.ncbi.nlm.nih.gov/22159581/ - Beane, Wendy Scott, et al. “Bioelectric signaling regulates head and organ size during planarian regeneration.” Development 140 (2013): 313–322. Experimental work on membrane voltage, organ scaling, and regeneration.
https://pubmed.ncbi.nlm.nih.gov/23250205/ - Durant, Fallon, et al. “Long-Term, Stochastic Editing of Regenerative Anatomy via Targeting Endogenous Bioelectric Gradients.” Biophysical Journal 112 (2017): 2231–2243. Evidence for stable, non-genetic alteration of regenerative target morphology in planaria.
https://pubmed.ncbi.nlm.nih.gov/28538159/ - Murugan, Nirosha J., et al. “Acute multidrug delivery via a wearable bioreactor facilitates long-term limb regeneration and functional recovery in adult Xenopus laevis.” Science Advances 8 (2022): eabj2164. Frog-limb regeneration study and its limitations.
https://doi.org/10.1126/sciadv.abj2164 - Kriegman, Sam, et al. “A scalable pipeline for designing reconfigurable organisms.” PNAS 117 (2020): 1853–1859. Foundational xenobot design paper.
https://doi.org/10.1073/pnas.1910837117 - Blackiston, Douglas, et al. “A cellular platform for the development of synthetic living machines.” Science Robotics 6 (2021): eabf1571. Self-organizing, cilia-driven xenobots and their observed behaviors.
https://doi.org/10.1126/scirobotics.abf1571 - Kriegman, Sam, et al. “Kinematic self-replication in reconfigurable organisms.” PNAS 118 (2021): e2112672118. Controlled xenobot replication experiment.
https://doi.org/10.1073/pnas.2112672118 - Gumuskaya, Gizem, et al. “Motile Living Biobots Self-Construct from Adult Human Somatic Progenitor Seed Cells.” Advanced Science 11 (2024): 2303575; first published online November 30, 2023. Foundational anthrobot paper.
https://doi.org/10.1002/advs.202303575 - Gumuskaya, Gizem, et al. “The Morphological, Behavioral, and Transcriptomic Life Cycle of Anthrobots.” Advanced Science 12 (2025): 2409330. Follow-up characterization of anthrobot development, gene expression, self-repair, and life cycle.
https://doi.org/10.1002/advs.202409330 - Levin, Michael. “Technological Approach to Mind Everywhere: An Experimentally-Grounded Framework for Understanding Diverse Bodies and Minds.” Frontiers in Systems Neuroscience 16 (2022): 768201. Primary statement of the TAME framework.
https://doi.org/10.3389/fnsys.2022.768201 - Levin, Michael. “Bioelectric networks: the cognitive glue enabling evolutionary scaling from physiology to mind.” Animal Cognition 26 (2023): 1865–1891. Levin’s argument for continuity between developmental and neural information processing; the paper explicitly brackets first-person consciousness.
https://doi.org/10.1007/s10071-023-01780-3 - Loy, Ignacio, et al. “Where association ends: A review of associative learning in invertebrates, plants and protista, and a reflection on its limits.” Journal of Experimental Psychology: Animal Learning and Cognition 47 (2021): 234–251. Independent methodological caution regarding claims of non-neural associative learning.
https://doi.org/10.1037/xan0000306 - Figdor, Carrie. “What Could Cognition Be, If Not Human Cognition? Individuating Cognitive Abilities in the Light of Evolution.” Biology & Philosophy 37 (2022). Philosophical analysis of how cognitive capacities should be identified across different biological lineages.
https://doi.org/10.1007/s10539-022-09880-z
Discussion