A universal Turing machine isn't intelligence — it lacks three things: a prior (Solomonoff: favor short programs), a goal (no G; Hume's guillotine), and resource economics (a UTM has infinite time). All three together = Legg–Hutter / AIXI. Internal to this conversation, a sharper definition: intelligence = self-supplied convergence acceleration — a measure-concentration device (UTM = flat measure; intelligence un-flattens it). Equivalently, intelligence = internalization of the eight control-methods (search = internalized selection, System-2 = internalized audit, imagination = internalized isolation — Popper: "let our hypotheses die in our stead").
Simon's key move: intelligence isn't optimization, it's satisficing under scarcity — replace argmax with a feasibility problem ("find any x with u(x) ≥ T"; utility demoted from objective to constraint). "vibe-linter is a satisficing machine: the validator checks u(x) ≥ T, the first artifact to earn the key is accepted — nothing ever asks if it's the best code." That demotion is wirths-law-as-kkt-constraint executed at civilizational scale.
This also resolves prosthetic-vs-constitutive-scaffolding: prosthetic scaffolding is externally-implemented cognition, so the model reclaims its own organs; constitutive scaffolding can't be internalized not functionally but positionally (internal gate = self-discipline, external gate = institution; promise ≠ proof) → institutions-operate-on-the-joint-distribution.