What is intelligence?
The cuts
Every aspect of learning or intelligence can in principle be so precisely described that a machine can be made to simulate it.
The Dartmouth framing: intelligence as whatever can be specified. It founded a field and begged the question.
A Proposal for the Dartmouth Summer Research Project on Artificial Intelligence1955
Intelligence is not skill. It is the efficiency with which you acquire new skills on problems you have never seen.
Severed intelligence from performance — a system that memorizes a million tasks is skilled, not smart.
On the Measure of Intelligence2019
Intelligence is relevance realization — the standing capacity to zero in on what matters here and now without checking everything. Not information, not skill: the art of ignoring almost all of it.
Named the missing middle between McCarthy's 'anything specifiable' and Chollet's 'skill-acquisition efficiency' — the ability to find what's relevant is what has to be acquired first.
Relevance Realization and the Emerging Framework in Cognitive Science2012The tensions
Dartmouth 1956 framed it operationally: intelligence is whatever can be described precisely enough to simulate. The framing founded a field, and quietly assumed the answer to the question it was asking.
Chollet’s 2019 cut severed skill from intelligence: performance you already have versus the efficiency of acquiring performance you don’t. By that measure, most of what impresses us about machines is skill. This site keeps one of its founding convictions here: intelligence is not one thing.
Vervaeke points beneath both. Before you can acquire a skill or simulate a behavior, you have to solve a prior problem no one hands you: out of the infinite features of a situation, which few are worth attending to at all? That sorting — relevance realization — is the work most invisible when it succeeds, and the first thing that fails in a system with no stake in its world.
