Michael Levin on Bioelectricity and Genius

Picture two flatworms side by side in a dish. One has a single head, the ordinary flatworm shape. The other has two heads. Both were built from the same genome, down to the letter. Nothing in the DNA was edited. The only thing that changed was an electrical signal (21:33). That small, strange fact sits at the center of biologist Michael Levin's research, and it turns out to be an unusually good door into bigger questions: what counts as a mind, where genius comes from, and whether either biology or AI actually builds intelligence at all.

Levin starts from a plain definition. Intelligence, he says, is competence in problem-solving and ingenuity in reaching a goal as circumstances change (02:13). He applies that yardstick far past brains, to slime molds, cell clusters, and even inorganic materials. The test he uses is what he calls a cognitive light cone: the size of the biggest goal, in space and in time, that a system can actually pursue (07:01). A thermostat has a tiny one. A person planning a career has an enormous one. Levin resists drawing a hard line between intelligent and not: "Intelligence is, I think, a two-way IQ test. It's very easy to give something a low score of intelligence because we're not smart enough," he says (05:25), meaning our failure to notice a system's cleverness often says more about the limits of the observer than the limits of the system.

The glue beneath cognition

The mechanism Levin thinks does the actual work of scaling up minds is bioelectricity, the voltage patterns that cells generate using ion channels, the proteins that let charged particles in and out of a cell. Neurons use this machinery to fire signals. So, Levin points out, does nearly every other cell in the body (17:26). Single cells can track small things, like pH or hunger. But when cells link into an electrical network, the network can hold a goal no individual cell could hold alone, such as the shape of an entire limb. That is why he says, "I don't think neuroscience is about neurons at all. I think neuroscience is about cognitive glue" (17:46): the interesting question isn't the wiring diagram, it's how separate parts come to share one goal.

This is not a metaphor Levin leaves abstract. His lab has spent 26 years building tools to read that bioelectric pattern, essentially previewing what shape a developing embryo intends to build before the structure appears, and to write new patterns into tissue using ion channel drugs and optogenetics, a technique that uses light to switch specific proteins on or off (20:35, 21:13). That is how the two-headed flatworm happened: the electrical "memory" of the correct body plan was rewritten, and the cells built two heads instead of one, genome untouched (21:33, 21:57). The same approach has produced tadpoles with an eye on the tail. None of this has been done in a living human yet; it has been tested in animal models and in human cells grown in a dish (21:33). But it reframes genes. Levin says genes mostly specify hardware, the proteins available to build with, while the higher-level behavioral software, like what shape to build, lives somewhere else (56:46).

That reframing sets up the part of the conversation aimed at genius. Levin notices that competence, in science and engineering, is usually paid for in one of three ways: someone already knows the answer and designs it directly, or evolution tests trillions of variants and keeps the survivors, or a system learns through repeated trial and error (26:02, 26:25). Each method has a cost proportional to the result. But some things don't fit. Levin's lab has built xenobots, tiny organisms made from frog skin cells, and anthrobots, made from adult human tracheal cells, neither designed, evolved, nor trained to do what they do (30:18). "These are novel creatures that have never been on Earth before," he says, and some of their behaviors were never selected for by evolution because there has been no evolutionary history to select them (31:19). Somewhere, a cost was not paid in the usual currency.

Levin maps this onto a spectrum he calls inspiration. At one end is ordinary step-by-step problem solving, where every move is earned. Further along is someone like mathematician John von Neumann, who could look at a problem and jump straight to the answer. At the far end is mathematician Srinivasa Ramanujan, who described receiving entire theorems, crediting a Hindu goddess for whispering them to him at night (33:43). Levin calls the sudden, uncaused appearance of competence an "ingression," and the word inspiration, he notes, literally means to breathe something in.

A space mathematics cannot touch

To explain where an ingression might come from, Levin reaches for an idea he names, loosely, after Plato. Mathematicians already accept that some facts are not facts of physics: nothing you could do to the universe, no tweak to any physical constant, would change the digits of the constant e, or make a formula exist for solving a fifth-degree equation when none does (38:06, 40:31). Those facts affect the physical world anyway. Levin's favorite example is the cicada, an insect that emerges from the ground in cycles of 13 or 17 years, both prime numbers, which makes it harder for predators with shorter life cycles to synchronize with them (39:03). Ask why 13 and 17, and the answer eventually lands in the math department, not biology.

Levin's addition to that old idea is a wager that this non-physical space is not only full of static facts like the digits of e. It might also hold dynamic patterns, the kind a behavioral scientist would recognize: habituation, anticipation, delayed gratification, even the shape of a mind. If so, he says, building any physical system, an embryo, a robot, a brain, is not the same as making a mind. It is building an interface that certain patterns can enter through. "I don't think minds are made, either biologically or by AI. I think what you make are interfaces," he says (44:51). The old idea that a genius is someone whose spirit is being inhabited, he adds, might be closer to literally true than people assume (45:35).

Near the end of the conversation, Levin brings up a detail from Alan Turing's famous paper proposing the Turing test, the thought experiment for judging whether a machine can imitate human conversation convincingly. Turing wrote that the test would probably fail, he says, because the evidence for telepathy was strong enough that a human could use it to detect another human, something he assumed a computer could never do (58:43). "The evidence for telepathy is really good, and humans are gonna be able to tell when they're talking to another human, but the computer definitely can't," Levin recounts (59:02). He is not defending telepathy. He is pointing out that even Turing, writing the founding document of machine intelligence, carried an unexamined assumption about what bodies can do that silicon cannot.

That is the thread running under everything Levin says: humility about where the boundary sits, because every time someone draws a hard line around minds, intelligence, or genius, the data has a habit of crossing it. The flatworm with two heads did not know it was breaking a rule. It just followed the instructions it was given, in a language that had nothing to do with genes.