Meet Cyc.

June 8, 2002 12:17 PM

Meet Cyc. This endeavor to produce AI has been going on since 1984. In ’86 it asked if it were human; it later asked “if any other computers were engaged in such a project”. It’s strength lies in a database of assumptions and generalities, in the hopes that it will eventually “generalize as much as possible until further generalization would be false”. Is this going to be the breakthrough, or does it just seem really cool? (yes, via Slashdot)
posted by sherman (9 comments total)

Cyc is (very) unlikely to be much of a "breakthrough", in the sense of it emerging into consciousness one fine morning, but it is certainly really cool, and already useful. Lenat's approach is too top-down to create a living, evolving intelligence, but I'm not sure that's really even his goal. Cyc is designed to do, not to be, and it does Cool Stuff.

(side note: Lenat had some earlier Cool Stuff in AM and EURISKO (here, here, and the requisite extrapolation here.)
posted by apostasy at 2:54 PM on June 8, 2002

Well, this is old, obviously. And I've heard about it before.


I always thought it was kinda funny. In the 1970's people realised that it would take 20 years to enter in all the data you would need for any kind of usefull interaction with a computer. Most people gave up but these guys just sat down and started.

Now it's 20 years later...
posted by delmoi at 3:06 PM on June 8, 2002

This spring, the developers' company, Cycorp Inc., sent their 18-year-old creation off for some higher education, creating a Web link to let the public download Cyc's knowledge base and teach it things, too.

ugh! your story is on the web! give the link for the love of god.

here it is
posted by rhyax at 3:38 PM on June 8, 2002

Dayum. And here I thought the wee AliceBot was cool.
posted by gummi at 3:49 PM on June 8, 2002

Jorn Barger's hosted a Cyc FAQ by David Whitten online as long as I can remember; it actually predates Cycorp. (Jorn has worked in the textual AI field. As with his politics, he's a bit of an iconoclast.)

Certainly, it's better to think of Cyc as an inference engine instead of something conscious. The FAQ notes that while you can ask Cyc questions, you can't expect to hold a conversation (which involves, among other things, a "state") with Cyc. In that respect, Cyc per se isn't even a candidate for the Turing Test.

About the OpenCyc project; it looks like you can already download a beta.
posted by dhartung at 3:50 PM on June 8, 2002

While Cyc may not explicitly be trying to win the Turing prize, I imagine that such a database would be immensely useful in the development of a later program that could compete. I imagine that it could be used by a more explicit attempt at true intelligence as a kind of onhand database of human common sense, just as our programs of today draw on assembly-code operations that are only vaguely referenced in the structure of the programming language.

It has long been my imagining that if we do achieve some kind of artificial intelligence on our current models of computing, it would likely be an intelligence that seems very alien to our own. After all, such an intelligence would be expected to conform to the advantages and pitfalls inherent in its architecture, and our current computing model has very little to do with the computing model that our brains run on. Our advantages include powers of abstraction, generalization, and improvisation, while our defects include an imperfect memory and (as often as not) imperfect reasoning. Could we even expect a mind exactly like ours, but with the gift of (say) an absolutely perfect memory to hold that much resemblance to us? (Borges speculates on this in the story The Zahir.)

I speculate that artificial intelligence, at this point of the game, is a slave of the standard architecture, unlikely to rise too far beyond the intellect of insects... To get anything that may qualify for the Turing prize, we will have to begin exploiting new technologies, such as massively parallel computing, quantum computation, what have you. But on the other hand, the Turing Test looks for human-like intelligence, which is in no way the only possible kind.
posted by kaibutsu at 2:18 AM on June 9, 2002

I did an article on Cyc and the current state of AI about a year ago: Machines in the Myths. I have all kinds of stuff from Cyc, including the classified tests that they released to journalists, and I tell ya...it hasn't come much farther in the last year. I think the guys are visionaries, but I'm not sure what they envision is really possible. OpenCyc was slated for release in August of 01, but was just finally released in April of this year...in incredibly simplistic form. I am however really interested in seeing how other people apply the knowledge base.
posted by dejah420 at 8:51 AM on June 9, 2002

> Cyc (pronounced "psych")

Well, maybe to them, but it means "tits" in Polish and it's pronounced "sits". There's a band called "Big Cyc" and it means big tits.

Silly buggers.
posted by pracowity at 9:01 AM on June 9, 2002

I remember someone writing about how this system would prove to be a very good 'shepherd' to a pseudo-evolutionary AI mutation program. Basically use random code mutations to change a codebase and use a system lke Cyc to determine the success or failure of any given mutation, until after trillions of iterations your have code that is AI-like. Or something. But it would be cool for that.
posted by n9 at 9:26 AM on June 9, 2002