Index
A field survey · The limits of artificial intelligence

The Jagged
Frontier

AI does not advance along a single line that one day crosses a wall marked human. It advances as a coastline — superhuman in more and more places, oddly blank in others, never collapsing into one equivalent moment.

00 · The reading

Most of the confusion in the public debate comes from trying to collapse a jagged frontier onto a single question — how smart is it.

These five soundings take the limits of AI one terrain at a time — culture, games, software, science — and find the same structure underneath each. A single variable keeps doing most of the explanatory work: the price of the verifier. Where a cheap, reliable check exists — a game won, a test passed, a proof machine-verified — a system can learn its way past human performance. Where the verifier is expensive, slow, or simply missing, progress stalls no matter how large the model grows. The frontier is jagged because that price is uneven.

The master variable

One axis runs under all of it

The clearest demonstration is the sciences. Rank four fields by how cheap and accessible their verifier is, and you have very nearly ranked how far AI has already reached into each.

← cheaper · faster · the AlphaZero regime absent · slower · the wall →
Mathematics

A proof checks itself. The only perfect, cheap verifier — and the field AI has entered most deeply.

Chemistry

Verifier good but expensive: simulation, then the wet lab. The factory is literally being built.

Physics

Split: simulation thrives; fundamental theory stalls where the experiment is inaccessible.

Biology

The worst verifier of the four — slow, noisy, ethically bound. Correlation-rich, causation-poor.

Read the full argument in Chapter 5 — The Verifier and Nature.

Written in the voice of an analytical observer — calm, concrete, skeptical of simple answers — and revised against an adversarial editor before publication. No hype, no doom; constraints first.

How this survey was made