Open Cyc and the gate (Score 2, Interesting)

Taken as criticisms, the allusions to 'bloat' and 'database' are both significantly wide of the mark: if Cycorp has been guilty of anything, it's historically underestimating the size and technical complexity of the knowledge base indicated for the common sense reasoner the company aspires to build. OpenCyc is not a database except in the most attenuated sense: it encodes, not instance-level facts, but quantified and contextually parameterized rules for reasoning about the everyday world, and it is, if anything, far too small for this purpose. The number and complexity of the rules needed for this is fairly staggering and too-little-appreciated, even by many in the AI community though not Minsky and McCarthy, both of whom are on the record as having recognized Cyc as one of the very few efforts in the field that was on anything like the right track). It's also fair to say that efficient and suitably flexible inference over a knowledge base of this size and complexity, and automated induction of new reasoning rules on the basis of experience - both obvious prerequisites for what Cycorp has been trying to do - present significant and partly unsolved theoretical challenges. The company's surprising willingness to tackle such weighty and potentially intractable issues head-on is a thing greatly to be commended in the present season of intellectual and commercial timidity, and even though they may not have always been able to deliver on every promissory note, one can't help but admire their spirit. And the fact remains that OpenCyc is now being used by an enthusiastic community of unaffiliated developers who are busily laying the groundwork for a new suite of open source applications. Judgement should not be pronounced on the basis of their efforts before they have been given the chance to see what they can deliver.