The Body Is the Memory: How a Brainless Slime Mold Learns, Remembers, and Socializes

Forum topic · ✨步子哥 · 2026-07-19

Summary

Physarum polycephalum, a giant single-celled slime mold with no neurons, reproduces the Tokyo rail network, solves mazes, and anticipates periodic events. This article reviews the research showing how it does this without a brain: memory is encoded in tube diameter hierarchy (Alim et al., PNAS 2021), in rhythmic contraction patterns, and in externalized slime trails left in the environment (Reid et al., PNAS 2012). Physarum habituates to quinine and caffeine, and learned behavior can be transmitted to naive individuals via cell fusion (Vogel & Dussutour, Proc. R. Soc. B 2016). The author draws an analogy to the von Neumann bottleneck: Physarum is a living compute-in-memory system in which morphology is simultaneously storage and computation, suggesting alternative architectures for AI hardware and questioning whether intelligence requires neurons at all.

Overview

A Chinese-language deep dive on *Physarum polycephalum*—the brainless slime mold that solved the Tokyo rail network—and what its memory, learning, and social behavior imply for AI architecture.

Key points

Tokyo rail network (2010): Nakagaki's team placed oat flakes at 36 Tokyo station positions on an agar plate with *Physarum* at Tokyo Station. Within 26 hours the slime mold's tubular network closely matched the topology of the Tokyo metro system, including hubs and redundant connections (Tero et al., *Science* 2010).
No neurons at all: *Physarum* is a single giant multinucleate cell (plasmodium, up to ~900 cm²). It has no neurons, synapses, or any nervous structure, yet it navigates mazes, solves bandit problems, remembers periodic events, and distinguishes well-fed from starved peers.
Fluid as information: Karen Alim's 2017 work showed food-induced signals propagate not via electrical or chemical gradients but through fluid flow itself—faster flow creates shear force on tube walls, softening and widening them in a positive-feedback loop.

Three kinds of memory without neurons

1. Morphological memory (Alim et al., *PNAS* 2021, "Encoding memory in tube diameter hierarchy of living flow network"): delayed softening of tube walls encodes where food used to be. The body's shape *is* the map. 2. Contraction-pattern memory: rhythmic cytoplasmic streaming (actin–myosin driven, 1–5 min cycles) stores information analogous to activity waves in brains (Audrey Dussutour, CNRS). 3. Externalized memory: extracellular slime (ECS) marks explored territory. Reid et al. (*PNAS* 2012) showed that in a U-shaped trap, *Physarum* escapes by avoiding its own slime; coating the dish with slime abolishes this navigation—an extreme case of Clark and Chalmers' "extended cognition."

Learning and transfer

Habituation (Boisseau, Vogel & Dussutour, *Proc. R. Soc. B* 2016): *Physarum* habituates to quinine over 5 days while remaining averse to caffeine—stimulus-specific, like animal habituation.
Learning transfer via cell fusion (Vogel & Dussutour, 2016): fused individuals acquire the habituation of experienced ones, via circulating intracellular molecules.
Sensitization vs. habituation (Smith-Ferguson et al., 2022): different exposure regimes for NaCl produce opposite learning outcomes.
Periodic anticipation (Saigusa et al., *Phys. Rev. Lett.* 2008): after three hourly cold-dry-air pulses, *Physarum* slows in anticipation of a fourth; the rhythm persists for hours.

Social behavior

*Physarum* reads slime trails: it follows trails from well-fed clones (likely near food) and avoids those from starved or stressed individuals (Briard et al., 2020).
Kin/geographic recognition (*Physarum rigidum*, Masui et al., 2018) and "eavesdropping" on *Didymium bahiense* trails show comparative decision-making.
"Old" lab cultures fused with young ones—or revived from dormancy—regain youthful motility (Dussutour lab).

Why it matters for AI

In von Neumann architecture, memory and computation are separate, creating the von Neumann bottleneck. *Physarum* is a natural compute-in-memory system: morphology stores memory while simultaneously being the computation engine. This parallels modern AI-hardware directions (Google TPU, Intel Loihi, IBM TrueNorth) and hints at architectures where "using is training, form is memory."
The author's conclusion: *Physarum*'s intelligence is not a primitive version of neural intelligence but an intelligence of a different architecture—convergent evolution at the cognitive level. Quoting Nakagaki's 2010 remark on "primitive intelligence," the essay argues "primitive" is the wrong word: fluid can think, form can remember, the body can know.

References (as cited in the source)

1. Nakagaki, T., Yamada, H., & Tóth, Á. (2000). *Nature*, 407, 470. 2. Tero, A., et al. (2010). *Science*, 327(5964), 439–442. 3. Kramar, M., & Alim, K. (2021). *PNAS*, 118(10), e2102056118. 4. Boisseau, R. P., Vogel, D., & Dussutour, A. (2016). *Proc. R. Soc. B*, 283(1829), 20160446. 5. Vogel, D., & Dussutour, A. (2016). *Proc. R. Soc. B*, 283(1845), 20162382. 6. Reid, C. R., et al. (2012). *PNAS*, 109(43), 17490–17494. 7. Saigusa, T., et al. (2008). *Physical Review Letters*, 100(1), 018101. 8. Smith-Ferguson, J., et al. (2022). *Behavioral Ecology*, 33(4), 842–850. 9. Boussard, A., Latty, T., & Dussutour, A. (2021). *Phil. Trans. R. Soc. B*, 376(1821), 20190765. 10. Alim, K., et al. (2017). *PNAS*, 114(20), 5136–5141.