Please see the Wikipedia page on Cyc so that you will have some idea about the background for this project. In fact, the goal of "enabling AI applications to perform human-like reasoning" attribuded there there is actually somewhat more modestly stated than the goal of the Cyc project (as (I) understand it), which is to build something with the same common sense knowledge of an arbitrary human from the USA. This (fuzzier) goal is really monumental in size.
The words are all quite fuzzy, but the point of the Cyc/FC projects is (would be) to build a huge knowledge base and an AI system that is capable of using the knowledge base.
FreeCyc ("FC") is meant to be a free implementation of the basic program being worked on by Cycorp. There are a number of reasons to believe that even given the 20 year "head start" that Cycorp has, FC might actually have better chances of success.
One reason (also recognized by Cycorp) is that a project of this magnitude should have the involvement of lots of people. There have been thousands of person-hours (lots and lots of time, probably something between 200 and 500 person-years) poured into Cyc so far, but this has only been done by hundreds of people. Cycorp's current response is to provide "OpenCyc", and produce tools that will let people who aren't working for Cycorp build knowledge bases that Cyc can read. The idea seems good – it decentralizes the effort. However, there are various reasons that the effort may not really work out as hoped. (I'll go into these later.)
The second reason is that FC would be, explicitly, and from its inception, free in the sense of freedom. For the generic reasons, this gives the project an overwhelming "strength" advantage.
FC has notable application (and interrelationship) to HDM, and to other similar domain-specific artificial intelligence.
I figure that if Wikipedia can collaboratively build a huge encyclopedia for humans, we should be able to collaborativly build a huge encyclopedia for an AI system (and, as should go without saying, the AI system itself).
The KM project is working on tools that make it possible for anyone to easily add knowledge to their system.
There currently is a project called Open Mind Common Sense that is being worked on at MIT. Maybe the results/methods of this project would be useful to look into. [[Mindpixel?]] is a similar effort (with some interesting collaborative filtering tools). (I haven't checked out the licenses for these projects, but if they qualify, we should add them to the project list.)
But (forgive me if I am mistaken!) it seems to me that these projects are lacking the AI part.
And this is interesting, because that is precisely the part that Cycorp is hanging onto most dearly. I'm not saying that the Cycorp AI would be capable of doing all the things that these other projects are missing. I'm just saying that no one seems to have great AI right now - or if they do, they are keeping it to themselves.
I had some discussions with [[kschalm?]] in 2004 about building an AI system that would play (and learn about) any game you could present to it. As a limited example, it should be possible to build an AI system that will play Go on a board of any size. (Perhaps such a system exists, but to the best of my knowledge, competition between computer Go programs mainly focuses on the standard 19X19 board.) Or, more generally, on any graph. These programs would themselves be interesting - but a program that could play chess and go and othello and so on, using some kind of coherent representation for all of the games: that would be pretty cool. Incidentally, the title of this section is sort of a pun: this system is a meta-(game player), because it operates at a level "meta" to the games it plays; however, it is also a metagame player, insofar as hacking lots of games is itself a game, and insofar as the system is self-writing. --jcorneli