Textbook of AI

CYC (from "encyclopaedia"), launched by Douglas Lenat in 1984 at the Microelectronics and Computer Technology Corporation and continued at Cycorp from 1994, is the most ambitious knowledge- engineering project in the history of AI. The goal was to encode the entirety of human commonsense knowledge, the unstated background that human readers bring to any text, in a formal logical knowledge base, supporting commonsense inference that would let AI systems handle natural language and everyday reasoning robustly.

Over four decades CYC accumulated tens of millions of axioms in its proprietary representation language CycL (a many-sorted higher-order logic). Knowledge was added by hand by trained ontological engineers, supplemented by some automated knowledge-acquisition tooling. The system was used commercially in narrow applications, fraud detection, regulatory compliance, military intelligence, but never delivered the broad commonsense reasoning capability the project originally promised.

The research community has long disagreed about how successful CYC was. Critics argued that the explicit logical formalisation of commonsense knowledge was an essentially impossible goal; defenders, including Lenat himself, argued that no other approach had been seriously attempted at scale. With the rise of large language models, which encode commonsense knowledge implicitly through scaled pre-training and exhibit much of the inferential capability CYC pursued, the question is now whether CYC's explicit knowledge offers anything LLMs lack. Lenat argued, in late papers before his death in 2023, that it does, particularly for high-precision reasoning where LLM hallucination is unacceptable.