CONCEPT
The Knowledge Principle
Edward Feigenbaum's foundational claim that machine competence resides overwhelmingly in accumulated domain knowledge rather than in sophisticated reasoning—a thesis vindicated by large language models through a method Feigenbaum spent his career opposing.
The knowledge principle states, in Feigenbaum's own words, that in the knowledge lies the power. It inverts the founding assumption of artificial intelligence, which located the seat of machine intelligence in reasoning—in the cleverness of the inference machinery, the generality of the problem-solving methods, the elegance of the logical architecture.
Feigenbaum's experience building DENDRAL taught him that the reasoning engine needed to be only minimally sophisticated: modus ponens plus organized search was nearly all the inference that a knowledge-rich system required. What determined a machine's competence was the depth and specificity of what it knew. This was a productive inversion. Expert systems built on it—MYCIN, PROSPECTOR, XCON—demonstrated machine performance at the level of human specialists in narrow domains. And then
large language models arrived and confirmed the principle at a scale that makes the expert systems era look like a footnote: systems trained on the vast majority of everything humanity has written down, whose competence across an extraordinary range of domains is precisely