Peter McEwen

Many latches start as useful, then outlive this usefulness (cf @techgnostic's talk/discussion); The body may respond to a strong prediction error by reflexively latching; under dangerous uncertainty the body locks a "known-safe" pattern into tissue. This is a very useful capacity because it's very quick & targeted, and can massively shift behavior/computation in a reliable way. Even after the acute danger passes, the body may keep the latch depending on how serious the danger was and whether there could be similar situations in the future. Slowly, this pattern is consolidated into neural weights (LTP) and the latch is less needed to maintain safety. But the body may hang on to it regardless, because (1) memory is seen as default-valuable, and (2) it's easy to offer evidence of danger but hard to offer evidence of safety. Sort of like a bonsai tree that was tied into a specific shape, which then slowly grew to be that shape naturally, but still has its ties on. After you build yourself (with tension), the work becomes relaxing into your creation (by releasing tension). They're each hard in a very different way. Latches reduce nervous system dimensionality / degrees of freedom, and that can actually be good, because we start with extremely high dimensionality (similar to an unconfigured FPGA). Babies are very unlatched creatures and it's probably a fantastic thing to be a baby. But they're not very predictive or predictable. Latches are necessary to make nervous systems useful. It's just the epistemological & affordance structure of the situation often leads us to go overboard. There are also pathological latches, like in Reynaud's syndrome. Strict malfunctions of this system. This can be e.g. a matter of energy production (it takes energy to relax muscle) or cellular stability (muscle spasms). One big hope of vasocomp is that many cognitive, emotional, & phenomenological problems can be grounded in muscle physiology and addressed directly. "What sorts of phenomenological unpleasantness are due to muscle microspasm?" -- hard to say at this point, but not zero. In the future, we'll need a science of "latch interpretability" -- something like LLM mechanistic interpretability but for latches. Not only a map of where your latches are, but what each one is doing -- what patterns it's stabilizing vs blocking, what classes of events might have triggered it, which latches are (still) doing important things vs which are safe to open. teapawt.net/johnsonmxe/status/1873…