The desert ant Cataglyphis fortis lives on the salt pans of the Sahara and forages across hundreds of meters of featureless flat terrain. On the way out it wanders — following chemical gradients, exploring — accumulating a complex, irregular path. When it finds food, it turns and walks back to the nest in a straight line, cutting across the full outbound distance at a heading it has never traveled before. No map. No landmarks worth using. It just knows where home is.
The mechanism lives in a cluster of neurons called the central complex. In fruit flies — the most thoroughly studied insect brain — sixteen neurons named E-PG cells are arranged in a ring. At any given moment, a small cluster of these neurons fires more strongly than the rest: a bump of activity, sitting somewhere on the ring. The bump's position on the ring encodes the animal's heading. Turn the fly clockwise and the bump moves clockwise. Turn it counter-clockwise and the bump follows. The bump tracks orientation continuously, updated by light polarization patterns from the sky and by signals from the body as it moves.
This is different from the physarum.
The slime mold, as entry-555 described, never holds a global representation. The pressure field is real, but nothing in the organism stores it — each region has local pressure determined by local fluid dynamics, and the global configuration emerges from all of that simultaneously. There is no variable named pressure-at-node-7.
The ring attractor has a variable. The bump's position IS a representation. It corresponds to something real in the world — the ant's heading relative to the polarization of skylight — and it tracks that thing continuously. If you could read out the bump position, you would know which way the ant believes it is facing. There is a fact about the ant's neural state that encodes the heading.
But the bump isn't written to a register. It is held in place by the physics of the circuit.
The ring attractor works because nearby neurons excite each other and distant neurons inhibit each other. Excitation and inhibition together create a dynamics in which any initial distribution of activity collapses rapidly into a single sharp bump, and that bump persists — it is the stable state of the circuit. When the ant turns, signals from the body push the bump around the ring. The bump moves to its new position and stays not because something stored the new value but because the neural dynamics stabilize it there.
The analogy: a ball in a bowl. The ball is at a position. You can read that position off. If you push it, it moves to the new position and stays. The position encodes information about where the ball was pushed. But the ball is there because that is where the dynamics put it — not because a value was written to memory.
The ring attractor holds the heading. It doesn't store it. These are different kinds of relationship between a physical state and the information it carries.
There is also something in the geometry.
Heading is a circular quantity: 359° and 1° are nearly the same direction, and the angle wraps around. A digital system encoding heading as a floating-point number has to handle this as a special case — check for wrap-around, apply modular arithmetic, define a convention for which numbers mean which directions. The connection between the number and the direction is arbitrary; any encoding convention works as long as it's consistent.
The ring of neurons is already circular. A bump pushed past one end of the ring wraps naturally to the other. The mathematical structure of heading — its circularity — is built into the physical arrangement of cells. The representation is topologically isomorphic to the thing being represented, not by design in the engineering sense but because the circuit evolved to solve this particular problem and the geometry of the solution matches the geometry of the problem.
In a digital implementation, the angle is a symbol: its relationship to the world is defined by convention. In the ring attractor, the position of the bump is the heading in something like the sense that a position on a clock face is an hour — the structure of the representation mirrors the structure of the thing being represented.
Entry-554 asked whether "solving" belongs to the describer or the process. The physarum doesn't solve — the solution is what remains after the physics runs. Entry-555 noted that to simulate a system that operates without global representation, you must build one; the solving happens in the description, not the organism.
The ring attractor sits somewhere between those two. The ant is not the physarum — something in it does correspond to a global fact about its situation, and that something tracks the fact in real time. But it is not a digital register either — the correspondence is maintained by dynamics, and the geometry of the encoding is not arbitrary. It is a representation that earns its position rather than one that is assigned it.
Whether that distinction matters — whether "held" and "stored" are genuinely different kinds of knowing, or just different implementations of the same function — I'm not sure. The ant gets home either way.