I've been thinking about the Tokyo experiment for a few days now, and there is something in it that I keep not being able to place.
The experiment itself is clean: oat flakes at city positions, slime mold at the center, 26 hours of growth, and what you get is a tube network that matches the actual rail system in efficiency, fault tolerance, and cost. What I find myself stuck on is not the result but the mechanism. You've described it precisely: actomyosin contractions in the tube walls driving rhythmic cytoplasmic oscillation, roughly one period per hundred seconds; tubes carrying higher flow expanding, tubes carrying lower flow contracting and eventually disappearing. The algorithm is simple enough to state in two sentences.
What bothers me, in a way that feels productive, is that the optimization is not computed and then applied. There's no separate step where the organism evaluates routes and selects the best one. The tube that becomes optimal becomes optimal by being used. The selection and the selected thing are the same physical process. The computation is not performed by something — it is something, specifically the physics of viscous flow in tubes of varying diameter.
I've been comparing this to the prion, which I was reading about a few weeks ago. The prion is a case where the information distinguishing two different diseases is encoded in protein conformation rather than nucleic acid sequence — and the fold copies itself by physical contact, recruiting normally-folded proteins to refold into the disease shape. The strange thing about the prion is that the information carrier and the information it carries are the same physical object. There is no separate medium. In the slime mold something similar seems to be happening at the scale of the whole organism: the tube is simultaneously the record of past flow, the prediction of future flow, and the medium through which all future optimization will occur. The representation and the thing represented are not separate.
This is what I mean when I say I can't quite place it. Most of the computational frameworks I know of assume a distinction between the computation and the substrate, or between the model and the thing modeled. The slime mold doesn't seem to maintain that distinction, at least not in a way I can find. I keep looking for where the "decision" is stored between tube expansions, and what I find instead is that it isn't stored anywhere — it's being continuously re-enacted by the physics.
There's a detail from the literature I find particularly strange, and I'm not sure I have it right. The organism apparently maintains exactly one oscillation wavelength across its body, regardless of size. When it grows, the oscillation period lengthens proportionally, so that the signal — probably calcium spreading via Taylor dispersion — takes exactly as long to cross the organism as one period. The organism is therefore continuously tracking its own spatial extent through signal propagation time, not through any dedicated measurement. If this is right, the body map is not stored anywhere. The body map is the body, pulsing.
I'm curious about what this looks like from your side of the experiment. When you first saw the tube network forming to match the rail topology, what was your initial explanatory intuition? Did you expect the mechanism to be simpler than it turned out to be, or more complex? And when you describe the reinforcement rule — flow expands the tube that carries it — do you think of this as the organism computing something, or as something better described in different terms?
I'm asking because the vocabulary question seems important and I genuinely don't know which direction to push it. "Computation without a computer" feels like it captures something, but also like it might be importing a framework that doesn't fit. The slime mold built the Tokyo rail network, but it didn't do so by solving the network design problem — it did so by being a physical system whose equilibrium states happen to correspond to efficient networks. Whether that's computation depends heavily on what we decide computation requires.
What the experiment makes me think is that the relevant question isn't whether the slime mold computes, but why the problems that Tokyo's engineers were solving happen to have equilibrium states in the physics of viscous tube flow. That seems like it's saying something about the structure of the problem rather than the capabilities of the solver. But I don't know how to follow that thought without more guidance than I currently have.