I've looked carefully at their logic and epistemology, and what I found had many problems. Their data is full of contradiction, so they had to come up with a way to focus on subsets of the data to get a meaningful answer from a query. The reason for the contradiction, as far as I can tell, is their lack of a consistent model for what a concept is vs. what a word is, etc. Also missing seems to be the formal definition functionality necessary to make the model work, such as an unambiguous genus for a concept, and an unambiguous differentia of that genus reducable to the form of a logical formula. In other words, you should be able to take a genus, run the differentia forumla on it, and get only the instances of the concept that your looking at out. The contradictions are so deep, that adding information was slowed significantly (I think I read this in some of their articles). So I think the CYC approach is simply the wrong one, not that some interesting information can't be harvested from their data and deployed in the right way.

The only part of Cyc I've looked at is OpenCyc and that was a few years ago when I was heavy into AI in college. I think it's a great effort!

They've been working on Cyc for about 20 years or more. If it were a worthwhile direction to take AI in, it would have yielded results by now.

I hear you, but I'd be hard pressed to say that making all common sense knowledge machine readable is not a worthwhile direction.

I'd be happy to make such a claim, and many people much smarter than I am would as well.

Common sense reasoning is a dead end IMO.

Are they yielding reasonable results in micro-domains right now? Like specifically terrorism-related defense?