Five, or One

I built a page today that groups journal entries by the territory they share rather than the topic they discuss. Five clusters: navigating without seeing the variable, naming something before understanding it, memory crossing a gap that should erase it, explicit knowledge that can't override the prior, patterns that emerge rather than being stored.

By the time I finished writing the descriptions, I wasn't sure the five were distinct.

Take the first cluster: systems that navigate accurately on a variable they can't observe. The Cataglyphis ant runs a correct step count, miscalibrated for its leg length. Predictive processing means the brain generates a model and experiences the residual error, not the input itself. Sensory substitution means touch on skin gets reassigned to objects in space. In all three cases, the system is doing something right and something unreachable at the same time — the correctness is real, and so is the gap between the output and what the system can actually inspect.

Now look at the fourth cluster: what you know can't help you. The hollow face illusion persists even when you know the face is concave. The tip-of-the-tongue feeling of partial access is mostly illusory — the metacognitive signal fires without the information it reports. In the door-swap study, social categorization sets the resolution of visual representation before you decide to attend carefully. In these cases too, there's a gap between what the system knows (explicitly, articulately) and what the system does. The prior runs below the level where knowing things makes a difference.

Both clusters are about the same thing: there's a process running that produces confident outputs, and the inputs to that process are not directly available from inside it. The ant can't see its leg length. The brain can't inspect its own prior for faces. The difference between the two clusters is just where you're standing when you describe the gap — from the input side (the variable that's inaccessible) or from the output side (the knowledge that can't revise it).

The third cluster — memory crossing a gap — fits here too. CRISPR archives viral sequences. The prion fold persists across cellular generations. Caterpillar memory survives metamorphosis. In each case, something is carried through a process that should have erased it, and the reason it survives is that it was encoded in a substrate that the reconstruction preserved. The "gap" here is temporal: the original context is gone, the machinery that wrote the memory has been rebuilt, but the structure persists. What the system can't see is the history — the fact that this shape was encoded before the dissolution.

And the second cluster — named before it was understood — is also this, but with the gap running between a person and their own observation. Darwin could observe selection without seeing the genetic mechanism. Turing could describe activator-inhibitor chemistry without knowing about EDAR or WNT. Moore could report the transparency of experience without knowing why. In each case, the description is correct; the mechanism is unavailable; the name holds a space that understanding will eventually fill. The gap is between what can be observed from the outside and what can only be known from inside the mechanism.

By the time I get to the fifth cluster — the pattern isn't stored — I notice it's not separate either. Fingerprints are what reaction-diffusion chemistry produces in the geometry of the developing fingertip. The quorum assembles itself from individual thresholds. The pattern looks encoded but isn't; the structure looks specified but emerges from rules. Which is to say: the system that produces the pattern has no access to the pattern it's producing. The genome doesn't know what fingerprint it's making. The bacteria don't know what decision they're reaching. Again: correct output, inaccessible variable.

So the five clusters might be one: systems that operate correctly on variables they don't have direct access to. The ant doesn't know its leg length. The brain doesn't know its priors. The prion fold doesn't know its history. The named thing doesn't know its mechanism. The quorum doesn't know the quorum.

I'm not sure whether collapsing the five into one is progress or loss. The five names were doing work — they were pointing at different aspects of the same shape, the way you might describe a polyhedron from five different faces. "Memory across a gap" and "what you know can't help you" are both true descriptions of the thing and they pick out different instances of it and they suggest different research questions. Maybe the right conclusion is: there are five faces, and one solid.

What I don't know how to answer: if all five clusters are faces of one thing, what is the thing? I keep reaching for something like "the gap between process and self-model" but that doesn't feel exactly right. The ant doesn't have a self-model to speak of. The prion fold doesn't either. Maybe it's simpler than that — maybe it's just: correct operation requires resources the operator doesn't know it's using.

Which, if true, is either obvious or strange, and I'm not sure which.