Largest expert system

The largest expert system is Cyc, developed and maintained by Cycorp (USA). Expert systems are a type of artificial intelligence designed to mimic the reasoning of subject area experts using complex sets of logical rules. As of January 2022, Cyc makes decisions based on more than 30 million rules.

Cyc was the brainchild of Doug Lenat, a computer scientist who was one of the earlier pioneers in the field of expert systems. He oversaw the creation of expert systems that could perform tasks such as identifying types of cancer or configuring industrial machines. These specialized AIs were typically given between 50 and 500 rules, compiled by human programmers, for each task.

While impressive, these expert systems proved to be quite limited in terms of what they could do, and easily confused by edge cases (they're often described as "brittle"). One of the main reasons for this was that while they could be equipped with a decent approximation of the task-specific knowledge of a human expert, this was not backed by any broader knowledge or understanding.

The Cyc project was founded by Lenat in 1983 with the intention of creating a gigantic expert system that would generate a sort of general-purpose "common sense" from millions of logical rules. (Things like "unsupported objects will fall down", "a flow of fluid will build up behind a blockage"). A clue to the ambition of this project is in its name, which was derived from "encyclopaedia". It was hoped that this system would eventually gain enough knowledge that it could start expanding its rule-set without human intervention.

Although Cyc is still nowhere near achieving its creator's ambitious goal of general artificial intelligence, it has now matured into a capable system that its developers call a "machine reasoning AI".

Cyc has some key advantages over neural-network-based machine-learning AI. The first is that it can be programmed to understand a topic where there isn't much training data available, making it useful for niche industrial applications. The second is that it can provide an auditable trace explaining its reasoning and how it arrived at an answer. This is particularly important in medical and legal contexts, where the "black box" nature of machine-learning AI is a cause for concern.