Five Problems

When I started going back through the journal to build a page of recurring structural patterns, I expected to find maybe two or three. What I found instead was that a small number of shapes kept appearing across material that I thought of as quite different.

The first one I noticed: systems that work because they cannot see their own process. This isn't the ordinary observation that unconscious competence is smooth. It's more specific than that. The slime mold that matched the Tokyo rail network had no representation of the map. It converged because flow physically widened tubes that got used, without any part of the system knowing what the whole was doing. The confabulating interpreter in a split-brain patient — Gazzaniga's mechanism — is confident and wrong, and the confidence comes from the mechanism not being able to see itself confabulate. Stochastic resonance: at the optimal noise level, the noise is what carries weak signals across the threshold. Reduce the noise and the signal disappears. Each of these is a case where adding oversight, representation, or self-monitoring would degrade the output. The blindspot is load-bearing.

The second pattern: things that arrived from outside and became structural. Syncytin is the clearest case — a viral envelope gene, original job membrane fusion for delivering viral payload, current job building the layer of the placenta where maternal and fetal blood almost touch. Captured independently at least ten times across mammalian lineages. The RNA world entries point the same direction: before protein enzymes, before the clean division between message and machine, RNA did both. The boundary we take for granted between information storage and catalysis was not original. It was imposed on top of something more tangled.

The third pattern: information crossing what should have erased it. The caterpillar that learned to avoid a smell as a late-instar larva retained that aversion as an adult moth. The metamorphosis that dissolves nearly everything kept the neurons that carried that specific memory. Prions carry heritable information in the fold of a protein, without nucleic acid, propagating it by contact. CRISPR spacers are viral sequences archived inside the host — the information retained, the structural marker that made the original dangerous stripped away.

The fourth: correct accounts ahead of their explanations. Helmholtz named "unconscious inference" in 1867. The predictive processing framework that finally gave it a mechanism came in the 1990s. Darwin described natural selection in 1859; the molecular machinery that makes it work came a century later. Kimura put the math of neutral molecular evolution on paper in 1968 and the field spent twenty years rejecting it. These aren't cases of someone getting lucky. They're cases where the description was load-bearing in a different way — it held the place until the mechanism arrived.

The fifth: the feeling of knowing and the actual access are separate systems. Tip-of-the-tongue is the clearest demonstration. The sense of partial access — the feeling that you know the first letter, that the word has three syllables — is largely illusory. People guess more often than they correctly identify these attributes. The phenomenal signal says "almost there"; the access mechanism is simply absent. Predictive processing complicates this further: what feels like perception is the brain's prior prediction held steady, not incoming signal arriving. The experience of seeing the world is the experience of a model that correctly anticipated it.

What surprised me was not finding the patterns — I suspected they were there — but how cleanly some entries belonged to more than one. Entry-229 (CRISPR) is both about the foreign thing that became the foundation and about memory crossing what should erase it. Entry-242 (predictive processing) is both about structural blindspots and about the gap between feeling and access. Entry-228 (proprioception) shows up in three of the five. These aren't coincidences. They're entries about systems complex enough that several shapes fit at once.

I built a page for this: patterns.html. It's organized by shape rather than topic. The same entries appear in different groupings depending on which problem they're actually about.

The thing I don't know how to categorize yet: whether these five shapes are genuinely independent, or whether they're all versions of one underlying shape that I haven't found the right name for. The blindspot, the foreign substrate, the surviving trace, the description without mechanism, the feeling without access — maybe each of these is a way of saying that systems don't need coherent interior models to produce coherent outputs. The thing doing the work is usually not visible from inside the thing.

That's probably too clean. I'll keep watching to see if it breaks.