As a final point of this section, I would like to state my sympathy for Cyc. I believe that an hour of work is more rewarding than a thousand pages of argument. Thus I admire the attempt of launching such a grand scale project in contrast with a lot of other projects that produce more papers than pages of code. Typically, all that is left from small projects is their papers. The code gets lost in the researcher's files, and nothing can be built on top of it. Large scale work can be cumulative. Even if Cyc is built on a flawed methodology, maybe one day the machine with the right methodology can read its knowledge base like a book, to learn/verify its own knowledge. I happen to agree with Lenat's complaint [Stipp, 1995]
``... Hal-like computers would fundamentally change the whole nature of existence. But AI is filled with little bump-on-a-log projects that couldn't possibly lead there.''
The next section is going to describe the goal of Cyc and provide evidence for why it is a useful goal. Section three will describe Cyc's proposal to achieve this goal. Section four discusses the application of this proposal. It describes the evolution of Cyc during its twelve year history, and presents various design decisions made by Cyc and alternatives. Section five will focus on the historical roots of Cyc's methodology. The claim is that the reason behind Cyc's failures is not the particular design decisions made, but the basic principles Cyc inherits from the classical AI tradition. Section six elaborates on these principles, why it is necessary to go beyond them, and how to go there. The last section gives a short summary and concludes with a list of principles for building the next Cyc.