The Xenobot Problem: When Is Something Alive?

In January 2020, a biologist at Tufts University named Michael Levin and a computer scientist at the University of Vermont named Josh Bongard announced they had built something that did not fit any existing category. They had taken skin and heart cells scraped from frog embryos, reshaped them into millimeter-scale blobs, and watched the blobs move around a petri dish, push tiny objects, and heal themselves after being cut. Nothing about the cells had been genetically altered. The DNA inside them was ordinary Xenopus laevis frog DNA, the same DNA you would find in any African clawed frog in any pond. But the thing the cells had become was not a frog, and it was not anything that had ever existed before. Levin and Bongard called it a xenobot.

The reason this matters goes past the novelty of a lab curiosity. It matters because xenobots quietly break an assumption most people carry around without examining it: that an organism’s identity, its species, its body plan, its behavior, is fixed by its genome. Xenobots have a completely standard frog genome and a body plan no frog has ever had. That gap between genetic instructions and physical outcome is the whole story, and it is a bioelectricity story before it is a robotics story or a philosophy story.

A Frog With No Frog In It

Here is what a xenobot actually is, mechanically. Researchers harvest stem cells from the embryos of Xenopus laevis at a stage before those cells have committed to becoming skin, heart, or nerve tissue. They separate the cells and let them re-aggregate into small clusters, then shape those clusters, sometimes by hand with tiny tools, sometimes according to designs generated by an evolutionary algorithm run on a supercomputer at the University of Vermont called Deep Green. The algorithm was not told what a xenobot should look like. It was told to search through millions of possible cell arrangements for shapes that could move efficiently, and it found configurations no biologist would have sketched on a whiteboard.

The resulting organism has no brain, no nervous system, and no muscle in the conventional sense. It moves because skin cells are covered in cilia, the hairlike structures a tadpole normally uses to sweep mucus and pathogens across its skin. Repurposed on the outside of a xenobot, those same cilia row the whole cluster across a dish like tiny oars. Some versions use cardiac cells instead, which contract rhythmically on their own because that is simply what heart tissue does, with or without a heart to be part of.

Same genome. A shape evolution never tried.

None of this required a single edited gene. The frog’s genetic code was left completely alone. What changed was context: which cells ended up next to which other cells, and what those cells then did in response to that new arrangement. That is the detail worth sitting with, because it is also the entry point into what Levin’s lab has spent two decades actually studying, which is not robotics at all.

The Body Plan Comes from Voltage, Not Just DNA

Every cell in your body maintains a voltage difference across its membrane, usually somewhere between negative 40 and negative 90 millivolts, generated by pumps and channels that control the flow of ions like potassium, sodium, and chloride. Neuroscience built an entire field around the fact that neurons use fast voltage spikes to send signals. What gets less attention is that non-neural cells, skin cells, gut cells, embryonic stem cells, use the same basic electrical toolkit to talk to each other, just more slowly. Cells connect through channels called gap junctions that let ions and small molecules pass directly from one cell’s interior to the next, turning a sheet of tissue into something like a low-resolution electrical network.

Levin’s research program, built up over years of work on flatworms, frog embryos, and tadpoles, argues that this network carries pattern information: instructions about where a limb should form, how many eyes a face should have, when regeneration should stop. In one well known experiment from his lab, altering the bioelectric state of cells in a tadpole’s flank caused a fully functional eye to grow there, wired into the tadpole’s nervous system, with no genetic modification involved. The genome supplies the parts. The bioelectric network decides, in real time, what to build with them and where.

The instructions for the shape were never written in DNA. They emerged from cells talking to each other with voltage.

Xenobots are what you get when you strip cells out of the developmental context that normally constrains this conversation. Freed from the surrounding tissue that would normally tell them “you are becoming a frog,” the cells fall into a new stable pattern instead, one shaped by their own bioelectric and mechanical signaling rather than by an inherited developmental script. You are not looking at reprogrammed genes. You are looking at a reprogrammed conversation between cells that never needed their genes touched in the first place.

They Copy Themselves. Sort Of.

In November 2021, Levin, Bongard, and a researcher named Sam Kriegman published a follow-up in Science Robotics showing that certain xenobot shapes could make copies of themselves. Left in a dish scattered with loose stem cells, a C-shaped xenobot, resembling Pac-Man, would gather those loose cells into a pile using its own motion, and the pile would compress into a ball that itself became a new, moving xenobot. Kriegman’s evolutionary search had found the C-shape specifically because it was efficient at this piling behavior, out of thousands of shapes tested in simulation.

This is where the coverage at the time got ahead of the science. What xenobots do is called kinematic self-replication: one object physically herding loose material into a new copy of its own shape, the way a mold presses a new object out of raw material. It is not reproduction in the biological sense. There is no heredity, no genetic transmission, no mutation carried from parent to offspring, and no capacity to evolve across generations on its own. The replication also runs out fast, because it depends entirely on a finite pool of loose cells sitting in the dish. After two or three generations, the raw material is gone and the process stops.

This is not evolution. It is a shape that happens to be good at building copies of itself once.

It is a genuinely strange behavior for something built from ordinary frog tissue, and it is worth taking seriously as a demonstration that self-replicating machines do not require anything like a genome or a lineage. But it is not the beginning of a new biological species, and treating it as one overstates what the experiment showed.

So, Is a Xenobot Alive?

Biology textbooks generally hand you a checklist for life: metabolism, homeostasis, growth or development, response to stimuli, reproduction, and evolution through heritable variation. Viruses already sit awkwardly on this checklist, since they cannot metabolize or reproduce without hijacking a host cell, which is why biologists have argued for a century over whether viruses count as alive at all. Xenobots force the same argument with a different set of gaps.

Notice that a xenobot clears more of the checklist than a virus does. It metabolizes, using yolk platelets stored inside the original embryonic cells, which is also why it eventually runs down and stops moving after roughly a week to ten days, since it has no mouth and cannot feed. It responds to its environment. It heals a wound in its own tissue. What it lacks is growth, development, and any mechanism for passing variation to descendants, the three things most tied to the idea of a lineage. You end up with something that is unmistakably made of living material, behaving in living ways, while failing the one test, heritable evolution, that most biologists would call non-negotiable for calling something a life form in its own right rather than a living tool.

What Happens When You Try This With Human Cells

In November 2023, a Tufts researcher named Gizem Gumuskaya, working with Levin, showed the platform was not a frog-specific trick. Her team took adult human tracheal cells, the ciliated cells that line human airways, and let them self-assemble into small multicellular structures they called Anthrobots. Left to their own devices outside a body, the cells built themselves into spheres covered in outward-facing cilia and moved around a dish the same way xenobots do. Placed next to a scratch cut into a layer of neurons in a lab dish, clusters of Anthrobots encouraged the neurons to grow back across the gap, a result the team is now examining for wound-healing and tissue-repair applications.

This is where the question stops being abstract. A frog cell rearranging itself into an unfamiliar shape is strange. A human cell doing the same thing sits closer to territory with actual stakes, involving donated tissue, consent, and eventual clinical use. Nobody is claiming Anthrobots are a new form of human life, and the researchers themselves frame the work as a tissue-engineering platform, not biology’s next organism. But the same gap that makes xenobots hard to classify, living material organized by voltage rather than pedigree, applies to something built from your own cells just as easily as a frog’s.

The honest limitation in all of this is that “is it alive” may simply be the wrong question to keep asking, because the checklist was built for organisms that arrived through evolution, not organisms assembled deliberately from cells on loan from one. Levin’s own argument is less about resurrecting old definitions and more about replacing them: that an organism’s identity is a negotiable outcome of cellular communication, not a fixed property stamped in by genome or species membership. Xenobots do not resolve whether that communication counts as life. They demonstrate that the line was always drawn around evolutionary history and reproduction, not around metabolism or motion or self-repair, and that biology can now build things on the wrong side of that line on purpose. The open question is not whether a xenobot is alive. It is whether “alive” was ever the category doing the useful work.