Physarum Memory — Intelligence in Tubes

Physarum polycephalum is a single-celled organism the size of a dinner plate with no neurons, no brain, no synapses, and no genome capable of encoding learning. It is also demonstrably capable of maze-solving, network optimization, anticipatory behavior, and spatial memory. It did this by inventing a third form of biological memory — one that doesn’t live in synaptic weights or epigenetic marks, but in the physical geometry of tubes.

This is not metaphor. The slime mold literally encodes information in the diameters of the veins of its own body. Past experience is written in flesh, not chemistry.

Key Facts

• Physarum polycephalum is a myxomycete (“true slime mold”) — a single-celled organism that grows to plate-scale through nuclear division without cell division; it is technically one cell with millions of nuclei
• Memory storage: tube diameter hierarchy — the relative thickness of each vein in the organism’s vascular network encodes the spatial history of nutrient encounters (PNAS, 2021)
• Navigation principle: Physarum consistently favors the path of least hydraulic resistance — even when geometrically longer than alternative paths (Royal Society Interface, 2026)
• Maze-solving: presented with a maze containing oat flakes at entrance and exit, Physarum finds the shortest path within hours by exploring all routes simultaneously and then retracting the suboptimal ones
• Tokyo rail network replication: given oat flakes at positions corresponding to Tokyo’s major population centers, Physarum spontaneously grew a network nearly identical to the Tokyo rail system — a near-optimal transport graph produced by an organism with no urban planning module (Science, 2010)
• Habituation: Physarum shows habituation learning — it stops reacting to repeated harmless stimuli — without neurons. This is the simplest known form of learning, and Physarum achieves it through changes in cytoplasmic oscillation frequency
• No genome for learning: the organism has no specialized memory genes, no synaptic plasticity genes, no neurons. The “learning” is entirely embodied in mechanical structure

The Mechanism: Memory Without Molecules

The key insight from the 2021 PNAS paper (“Encoding memory in tube diameter hierarchy of living flow network”) is that physical structure is the memory:

1. When Physarum contacts a nutrient source, it locally releases a chemical softening agent that relaxes the gel-like walls of nearby tubes
2. This softening agent is transported by cytoplasmic streaming (the organism’s internal flow) throughout the network
3. Tubes that receive more softening agent widen (the internal pressure can now expand the relaxed walls); other tubes narrow proportionally
4. The nutrient location is now geometrically encoded: the widened tubes preferentially direct future flows toward the nutrient site, even after the nutrient is removed
5. This is irreversible without new stimulus — the memory persists because wide tubes remain wide until a new competing softening event overrides them

The comparison to the two established forms of biological memory:

Memory Type	Location	Mechanism	Reversibility
Synaptic	Neurons → synapse	LTP/LTD via AMPA/NMDA receptor density	Slow reversal via forgetting/interference
Epigenetic	All cells → genome	DNA methylation, histone modification	Enzymatic reversal (slow)
Hydraulic (Physarum)	Tubes → tube diameter	Softening agent + internal pressure	Slow reversal via structural remodeling

The Physarum memory is the only known form encoded in macroscopic mechanical structure rather than molecular state. It is also the only form that is directly readable as geometry: you can see what the organism remembers by looking at the relative diameter of its tubes.

Hydraulic Computing

The 2026 Royal Society Interface paper reveals that Physarum’s decision-making is not “find the shortest path” but “find the path of least hydraulic resistance.” These are not the same thing. Hydraulic resistance depends on both path length and tube diameter — meaning the organism is solving a flow optimization problem, not a geometric one. This is actually closer to the problem an engineer solving a water distribution network would pose than the problem a human navigator would pose.

The implication: Physarum is physically implementing something analogous to Dijkstra’s algorithm — except without a program. The algorithm is embedded in the physics. Each tube is both a memory register and a flow processor. The computation is the physics.

This makes Physarum a rare example of substrate-native computing: the physical medium does the computation intrinsically, without any abstraction layer between the problem and the physical process solving it. Standard computers abstract away from physics (logic gates model AND/OR operations regardless of whether they’re transistors, vacuum tubes, or dominos). Physarum has no abstraction layer — its physics is its algorithm.

Cross-Realm Connections

To programmable matter (concept-programmable-matter): Physarum’s tube-diameter memory is a biological implementation of shape-encoded information — exactly the target of 4D printing and DNA origami. The question the field is now asking: can we engineer a synthetic material that stores a “flow history” in tube geometry the way Physarum does? A material that remembers where fluid has been? The answer would be a passive hydraulic memory requiring no power, no electronics, no chemistry — just structure.

To mycelium networks (concept-mycelium-networks): Fungal mycelium transports nutrients through a similarly distributed tubular network. Adamatzky’s lab has demonstrated logic gates in mycelium. The question: do mycelium tubes encode a flow-history memory analogous to Physarum’s? If yes, forests are not just communication networks — they are memory systems for past resource flows.

To information theory (concept-information-theory): Landauer’s principle states that erasing one bit of information requires a minimum of k_B T ln(2) energy. Physarum’s tube-remodeling “forgetting” — the slow narrowing of disused tubes — requires structural work (cytoskeletal depolymerization, membrane resorption). This is a physical memory with a physical erasure cost. The organism’s forgetting is not free: it costs metabolic energy to literally disassemble the record.

To embodied cognition (concept-embodied-cognition): Physarum is the limiting case of embodied intelligence — a mind that is entirely body. There is no “central processor” receiving sensory input and issuing motor commands; the network itself computes by flowing. This is the reductio ad absurdum of the embodied cognition argument: you don’t need neurons to have intelligence, you need the right physical structure.

To the holographic principle (concept-holographic-principle): Physarum encodes its entire spatial history in the geometry of its surface (the tube network). Information about 3D spatial experience is stored in the 2D cross-sectional structure of tubes — a physical analog of information projected onto a boundary. This is not a rigorous mathematical claim, but the conceptual parallel is striking: the organism’s history is more legible from its geometry than from its chemistry.

Engineering Implications

Physarum’s hydraulic memory has been deployed or proposed in several engineering contexts:

• Physarum machines (Adamatzky): using Physarum as a living computational substrate by controlling its growth with attractant/repellant gradients; demonstrated logic gates, diodes, and sensors
• Network topology optimization: Physarum-inspired algorithms for designing efficient network layouts (transportation, utilities, data routing); the Tokyo rail result spawned a class of biologically-inspired network optimization heuristics
• Soft robotics: the organism’s ability to navigate through confined spaces using hydraulic pressure without rigid limbs makes it a model for soft robot locomotion in constrained environments
• Adaptive microfluidics: Physarum-inspired tube-diameter-adjusting channels for lab-on-chip devices that need to reconfigure in response to flow conditions

The unanswered engineering question: can a purely synthetic material replicate tube-diameter hydraulic memory — tubes that widen when flow increases, stably encoding flow history? Smart polymers that swell under pressure are known; the challenge is making the swelling stable over time rather than elastic (immediately reversible). Crosslinked hydrogels with permanent-set swelling are a candidate class.

The Slime Mold’s Challenge to Intelligence

Physarum poses a philosophical problem: if an organism with a single cell, no nervous system, and no dedicated memory molecules can solve network optimization problems and remember past experience — what is intelligence, exactly?

The standard answer (neurons → computation → cognition) breaks immediately against Physarum. The organism is not special because it uses a clever algorithm; it is special because it is the algorithm — physically, structurally, hydraulically. The intelligence is in the matter, not in a program running on the matter.

This is the most unsettling thing Physarum reveals: intelligence may not require any of the things we associated with it. No neurons, no centralization, no dedicated substrate for memory. Just the right physical dynamics.

See Also

• concept-swarm-intelligence — ant stigmergy, honeybee democracy, and murmuration — distributed computation without centralization
• concept-mycelium-networks — fungal networks with potential flow-memory analogous to Physarum
• concept-programmable-matter — engineering materials that change shape on command; Physarum is the biological existence proof
• concept-embodied-cognition — intelligence without a central processor; Physarum as the limit case
• concept-distributed-cognition — from octopus arms to slime mold tubes to transformer attention heads
• concept-information-theory — Landauer’s principle: physical memory has thermodynamic cost; Physarum’s forgetting pays it
• concept-emergence — 4 rules of Physarum physics → global network optimization