Slime Molds Solve Mazes With Self-Optimizing Electrical Networks

In 2000, a biophysicist named Toshiyuki Nakagaki, then at Hokkaido University’s Research Institute for Electronic Science, cut a maze into a plate of agar and let a yellow slime mold called Physarum polycephalum grow through it. He placed food at two exits and let the organism fill every corridor. Then he watched it retract from the dead ends until only the shortest path between the two food sources remained, connected by a single efficient strand of the organism’s own body. Nakagaki published the result in Nature under the blunt title “Maze-solving by an amoeboid organism.” The finding still gets cited today, not because slime mold is exotic, but because nothing in that maze had a nervous system.

Physarum polycephalum is not a fungus and not an animal. It is a single cell, sometimes the size of a dinner plate, containing millions of nuclei sharing one continuous membrane. It has no neurons, no synapses, no brain, and yet it routes itself through a labyrinth in something close to the shortest possible path. The mechanism behind that trick has spent two decades moving from a curiosity of developmental biology into a live research question for bioelectricity and unconventional computing labs, because the organism is doing something that looks a lot like network optimization, and it’s doing it entirely through oscillating voltage.

A Body That Thinks in Rhythm, Not in Circuits

Physarum moves by streaming its own cytoplasm back and forth through a network of tubes, a process biologists call shuttle streaming. The direction and speed of that streaming is set by rhythmic contractions of the tube walls, driven by actin and myosin, the same contractile proteins that make your muscles move. Those contractions are triggered by oscillating calcium concentrations inside the cell, and calcium flow across a membrane is an electrical event. Every few minutes, a wave of depolarization sweeps across the organism, thickness of tubes rises and falls with it, and the whole plasmodium pulses like something breathing.

That pulse is measurable with ordinary extracellular electrodes, the same basic technique an electrophysiologist would use on a slice of neural tissue. Researchers have recorded oscillating membrane potentials in Physarum for decades, going back to work by the Japanese physiologist Noburô Kamiya in the 1950s on protoplasmic streaming mechanics. What’s changed is the interpretation. Where that oscillation used to be filed under “cell biology curiosity,” a growing number of labs now treat it as the organism’s only computing substrate. There’s no other information-carrying system to point to.

No neurons, no brain, just an oscillating voltage that remembers where it’s been.

The tubes that see more cytoplasmic flow, and therefore more of the electrical activity that drives contraction, get thicker over time. Tubes that see less flow thin out and eventually vanish. Run that rule across a whole network for a few hours and the organism prunes itself down toward whatever configuration moves resources most efficiently between food sources. Nobody is directing that outcome. It falls out of local electrical and mechanical feedback repeated thousands of times across the organism’s body.

The Tokyo Rail Map Was the Moment This Got Serious

The maze paper was striking, but a 2010 study in Science is what convinced infrastructure engineers to pay attention. Atsushi Tero, working with Nakagaki along with Oxford researchers Mark Fricker and Dan Bebber and computer scientist Seth Bullock at the University of Southampton, placed oat flakes on a map of the Tokyo metropolitan area, positioned to mirror the region’s real population centers. Physarum grew across the map and, over about a day, built a tube network connecting the food sources that closely resembled the actual Tokyo rail system, including some of its redundant loops that make the real network resistant to a single line failure.

The team then built a mathematical model of the reinforcement and pruning rule the slime mold appeared to be following, essentially: strengthen tubes carrying flow, weaken tubes that don’t, and let a small amount of randomness keep exploring new routes. That simple local rule, run on a computer instead of a cell, generated networks with efficiency and fault tolerance comparable to networks designed by actual transportation engineers, at a fraction of the design cost. The paper’s title, “Rules for Biologically Inspired Adaptive Network Design,” made the intended audience clear. This wasn’t a biology paper dressed up for a general journal. It was an engineering result that happened to come from a cell with no engineers inside it.

The tubes carrying current survive. The rest get pruned.

Engineers Have Started Trying to Wire the Organism Directly Into Circuits

Andrew Adamatzky, who runs the Unconventional Computing Laboratory at the University of the West of England in Bristol, has spent the better part of two decades trying to turn Physarum’s electrical behavior into working components. In 2013, Adamatzky worked with Ella Gale and Ben de Lacy Costello to demonstrate what they called a slime mold memristor, a two-terminal device whose resistance depends on the history of current that has passed through it, built from a living strand of Physarum instead of the titanium dioxide typically used in memristor research. The organism’s tubes changed conductivity in a way that depended on prior electrical exposure, the living equivalent of a component that remembers.

Adamatzky’s lab has since built Physarum-based logic gates, sensors that respond to pollutants by changing the organism’s electrical output, and small robots steered by feeding electrical signals recorded from the slime mold into a motor controller. None of this has produced anything you would call a product. A slime mold memristor degrades within days because the organism it’s made from is, at the end of the day, alive and needs food and moisture to keep functioning. Response times are measured in minutes, not nanoseconds. Nobody in this field claims Physarum is going to replace silicon. The interest is narrower and, arguably, more useful: understanding what a self-optimizing electrical network looks like when it has to run on almost nothing, no external power supply, no fabrication plant, just ion channels and geometry.

Where This Connects Back to the Rest of Bioelectric Medicine

You don’t have to squint hard to see why researchers who study bioelectricity in animals find Physarum useful as a stripped-down model. Michael Levin’s lab at Tufts University has spent years documenting how non-neural bioelectric signaling, voltage gradients passed between ordinary cells through gap junctions, directs pattern formation during development and regeneration, work that shows up in how planaria regrow heads and how frog embryos decide where a limb goes. None of that involves a nervous system either. The organizing information lives in the voltage pattern itself.

Physarum strips that idea down to its simplest possible case: one cell, one continuous membrane, one oscillation, and a visible, measurable network shape that changes in response to it. There’s no synaptic weight to obscure the mechanism, no layer of neurotransmitter chemistry sitting between the electrical signal and the physical outcome. If you want to understand how voltage alone, without any wiring diagram laid down by a genome for a brain, can produce something that looks like decision-making, this is about as clean an experimental system as biology offers.

The Part Where You Should Be Skeptical

Here’s the honest catch, and it matters. Physarum is not “solving” the maze in any sense that involves representation, planning, or evaluation of alternatives. It is doing something closer to what a soap film does when it settles into the minimal surface between wire loops, physically relaxing into the lowest-cost configuration available to it given its material properties. The word “solving” is doing a lot of rhetorical work in headlines, including versions of this one. Calling it intelligence, a word that shows up often in press coverage of this research, is a choice about vocabulary, not a settled scientific finding. Some researchers in the field are comfortable with that word. Plenty of electrophysiologists are not.

A slime mold does not solve anything. It relaxes into the cheapest shape available to it.

The engineering promise has real limits too. Every attempt to build a working device out of live Physarum runs into the same wall: biological substrates decay, need feeding, and respond on a timescale of minutes to hours, which rules out most of the applications people get excited about when they hear “living computer.” What survives contact with reality is narrower and more interesting than a headline about a maze: a demonstrated, reproducible mechanism by which oscillating membrane voltage, with no neurons involved, can prune and reinforce a physical network toward something close to optimal. That’s the transferable result. The maze was just the experiment that made it visible.

What Two Decades of This Research Actually Add Up To

Twenty five years after Nakagaki’s maze, the throughline isn’t that slime mold is smart. It’s that a single cell, using nothing but calcium-driven voltage oscillation and a rule as simple as “reinforce the tubes carrying flow, prune the ones that aren’t,” can produce network designs that took human civil engineers decades to converge on through trial, error, and cost modeling. That is the actual finding worth sitting with. The open question the field hasn’t settled is how much further this rule generalizes, whether the same stripped-down logic of voltage-driven reinforcement is quietly running underneath more complicated biological networks, including the ones with actual neurons in them, or whether Physarum is a clever exception rather than a preview.