Why AM and Eurisko Appear to Work (1983) [pdf] | Hacker News Reader

Why AM and Eurisko Appear to Work (1983) [pdf]

aaaipress.org

27 pointstonyg5 comments

Eurisko is definitely part of CS and hacker folklore - but it is not really science. It wasn't reproduced (or for that matter, produced again by Lenat) - it's not clear, out of the described Eurisko output, what was really Eurisko and what was Lenat's contribution.

Lenat's "next gen Eurisko", Cyc, hasn't been nearly as impressive as Eurisko is supposed to have been.

It's a big enigma for me. It's like he achieved some big AI advance that ostensibly won a major competition, went dark for two decades, and then resurfaced with a grand project that looks crappy with no tangible results.

It matches the theory "AI advance way ahead of his time that Lenat keeps secret with a cover story" ... but I think Occam favors your explanation now.

> It matches the theory "AI advance way ahead of his time that Lenat keeps secret with a cover story" ...

This might be the case for all I know; As you say, Occam favors a simpler explanation. But that's not what bothers me - what bothers me is that so many references just take this folklore as truth without any critical thought.

I first read about Eurisko in a book[0] (the existence of which is hard to confirm on the Internet - kind of surprising). I was 12, it looked fantastic then, but it was fantastic of the "when I know a little more, I might have a grasp of how that works". But it's been 30 years since - and now, with a little bit of knowledge and scepticism under my belt, I'm almost sure that -- however impressive Eurisko might have been -- none of the praise written about Eurisko was written by someone who had witnessed it working or was even able to critically review its output.

Not much different than any other kind of lore. Definitely not science or technology.

[0] http://www.acornelectron.co.uk/books/info/methuen/The_Hitch_...

The Wikipedia page on Cyc [1] matches my memory of how it was presented at the time. It was being done at MCC with the expectation that it would be a commercial product at some point, I don't remember being shocked that there wasn't much being published about it.

[1] https://en.wikipedia.org/wiki/Cyc

Lenat himself was talking about using some of that tech for the US Intelligence community and policy prediction of other states. Not sure if anything came of it, but it's not far-fetched for these organizations to have shoveled some money in that direction.

Reply on news.ycombinator.com