The Layered Cognition — Summary

A nurse in a district hospital in Kerala asks a question in Malayalam about a drug interaction, and the answer comes back in two seconds from a model running on a server in the same building. The model is small and would lose badly to the frontier on general reasoning. For the question she actually asked, it is better, because it answered in her language, at the moment she needed it, without leaving the building.

Most thinking does not need the largest mind. It needs the nearest one that knows the work. A sane system distributes cognition the way a person does: most questions handled locally, fast and cheap; some escalated to a domain layer with deeper knowledge; the frontier reserved for the genuinely rare and hard. The frontier is the neurosurgeon. Indispensable for what only it can do, ruinous as the default for everything.

The domain layer is where the most striking gap lives, the capability the frontier will never build because the return does not justify it. A model deep in Malayalam medicine, or in Telangana cotton agronomy, is enormously valuable to the people it serves and worth nothing to a company optimizing for the global market. But it is cheap to build if you are not trying to build the frontier. The thing the market will not make is the thing most buildable by someone who is not the market.

There is a real dependency: the small model at the edge is often distilled from a larger one. The nurse does not know or care. She asked eleven questions during her shift. Ten never left the building. One climbed to the domain layer and came back. None reached the frontier, because none needed to.