Cyc is one of the most ambitious projects in the history of AI. Launched in 1984, it was originally envisioned as a three-decade research program that would proceed in three distinct stages (see Figure 4.1):

1. slow hand-coding of a large knowledge base;

2. fast and self-initiated acquisition of more knowledge through reading and assimilation of databases by the machine, leading to a point at which the system, according to its creators, having gone “beyond the frontiers of human knowledge,” would be capable of:

3. conducting its own research-and-development projects through learning by discovery, in order to continue ever expanding its knowledge base.

The first stage was envisaged as ending in 1994, the second by the late 1990s, and, as Doug Lenat, the main figure behind the project, put it, “Soon after, say by 2001, we planned to have it learning on its own, by automated discovery methods” (Lenat 1997b). By going through these stages, Cyc was intended to break out of AI's so-called knowledge acquisition bottleneck. According to initial estimates, the critical mass for passing through the first stage of hand-coding was on the order of one million facts. Putting such a large number of facts together with the heuristics necessary to apply them under the right circumstances was a daunting task that called for a “large, high-risk, high-payoff, decade-sized project” (Lenat and Guha 1990: Preface). Cyc was supposed to be exactly such a system – a $50 million project that in its first decade would require two person-centuries for data entry.