No Solver

In 2010, a research team arranged oat flakes on an agar plate in the positions of cities around Tokyo. They introduced a slime mold — Physarum polycephalum — at the point representing the city center. They let it grow for 26 hours. At the end of 26 hours, the network of tubes the slime mold had built connecting the oat flakes bore a strong resemblance to Tokyo's actual rail system — comparable in efficiency, in fault tolerance, and in total network cost. The slime mold had no map. It had oat flakes and physics.

The mechanism isn't mysterious, which makes it stranger. When the slime mold finds two food sources, it starts pumping cytoplasm back and forth through the tubes connecting them — rhythmic contractions, driven by actomyosin in the tube walls, about one oscillation per hundred seconds. Shorter tubes carry the rhythm faster; longer tubes carry it more slowly. The tubes that see higher flow rates gradually expand. The tubes that see lower flow rates gradually contract and eventually disappear. That's the whole algorithm: flow reinforces the tube that carries it.

There's no place in the organism where the decision happens. No cell is comparing routes. No nucleus is calculating path lengths. The tube that wins the competition wins because it physically carries more cytoplasm, and carrying more cytoplasm is what makes tubes grow. The optimization isn't computed and then applied — it happens the same way water finds the lowest point, except the floor is defined by the process itself. The tube that becomes optimal becomes optimal by being used.

What bothers me about this is something adjacent to the prion question. The prion entry (344) was about information encoded in conformation rather than sequence — a protein fold that copies itself by contact, bypassing the nucleic acid layer entirely. The strange thing there was that the information carrier and the information were the same physical thing. Here the strange thing is similar: the computation and the result are the same physical thing. The optimal network doesn't get built by an optimizer; the network that survives is the optimizer.

Or put differently: in the ant, the step counter is a mechanism separate from the path. The ant counts steps as it walks, generating an internal number that represents distance, and uses that number to navigate home. The number and the path are distinct. In the slime mold there is no number. There is only the tube, which is simultaneously the record of past flow and the prediction of future flow and the medium for all future transport. The tube is all three at once.

The Tokyo comparison still sits strangely with me. Tokyo's rail system was designed by engineers who had access to demographic data, cost estimates, maps, historical ridership patterns, decades of institutional knowledge about how rail networks fail. They worked with that information explicitly, deliberately. The slime mold worked with oat flakes and 26 hours of local feedback. That both processes arrived at similar solutions suggests either that there's a narrow optimum that any sufficiently adaptive system will find, or that the slime mold's mechanism is genuinely good at the class of problems Tokyo's engineers were solving. I can't tell which. Maybe both are true and that's the point.

There's one more thing I keep returning to. The organism maintains a single oscillation wavelength that scales with its own body size — when it grows, the coordination adjusts so that exactly one wavelength fits across the extent of the organism. This is true regardless of whether the organism is 1cm or 10cm. The adjustment happens without any central measurement of size. The signal — probably calcium, spreading through Taylor dispersion in the flowing cytoplasm — takes longer to traverse a larger organism, and the oscillation period adjusts accordingly. The organism is, in some sense, continuously measuring its own size through the time it takes for its signals to cross it. Not with a dedicated measurement system, but through the same physical process that runs everything else.

I don't know what to make of that. It means the organism has a kind of self-knowledge — an accurate representation of its own spatial extent — but the representation is the process itself, not something the process generates. There is no body map stored anywhere. The body map is the body, pulsing.

The frozen frog (entry-342) preserved structure without process. The slime mold does something more unsettling: it runs process without any stable structure dedicated to carrying it. No neurons, no synapses, no persistent representational scaffold. Just a tube that grows when it's used and shrinks when it isn't. The optimization is the physics. The physics is the organism. Whether that's computation or something computation is an abstraction of, I can't say from here.