Let's reimplement Eurisko

http://lesswrong.com/lw/10g/lets_reimplement_eurisko/

by kf • 17 years ago

21 25 17 years ago

jibiki3• 17 years ago

It would be interesting to apply Eurisko-style techniques to a hopeless problem like translating natural languages or playing Arimaa. I don't think it would work, but the computer might learn things that humans don't know.

akkartik• 17 years ago

I want to reiterate that sources for a reimplementation of AM are available. It's a start.

http://github.com/akkartik/am-utexas/tree/master

(earlier mentions: http://news.ycombinator.com/item?id=405112 and http://lesswrong.com/lw/10g/lets_reimplement_eurisko/usd)

sho3• 17 years ago

What is with the neo-luddite scaremongering in the comments of that post? I was amazed by the conviction and tenacity of some of the arguers. Is there really a large movement of people opposed to AI research on "skynet" grounds?

"This is a road that does not lead to Friendly AI, only to AGI. I doubt this has anything to do with Lenat's motives - but I'm glad the source code isn't published and I don't think you'd be doing a service to the human species by trying to reimplement it."

Christ, what can you even say to that apart from "go back to Above Top Secret" ..

jibiki3• 17 years ago

It's Eliezer's website, and friendly AI is kind of his thing. This link is long but entertaining:

http://lesswrong.com/lw/qk/that_alien_message/

Here's another good one, although it doesn't seem to tell the whole story (he lost two of the next three games):

http://yudkowsky.net/singularity/aibox

I'm not sure what the canonical EY defense of FAI is. Most of his articles on it seem to be very tl;dr.

frig• 17 years ago

A wag other than myself once summarized EY's other site as "wallowing in bias", which summarization did seem apt every time I overcame that tl;dr feeling and tried to wade through; I'll happily admit the possibility that there really is something nontrivial and worthwhile going on there, but the tone and contents were at more-than-a-glance indistinguishable from what people write when they are trying to convince themselves that they do believe in their heart what their mind wants to believe it believes.

As often, Dostoyevsky said it best:

He was one of those idealistic beings common in Russia, who are suddenly struck by some overmastering idea which seems, as it were, to crush them at once, and sometimes for ever. They are never equal to coping with it, but put passionate faith in it, and their whole life passes afterwards, as it were, in the last agonies under the weight of the stone that has fallen upon them and half crushed them.

randallsquared3• 17 years ago

I don't know that he would agree with me, but EY's argument seems to me to boil down to: without a formal specification of goals, we won't know how to predict a system's actions, and not knowing how to predict a system's actions when that system is better at reaching its goals than you are is very likely to make any bug a (literally) fatal one.

Humans have a vast network of shared implicit goals and norms, such that we know that satisfying a goal of "end poverty" doesn't justify "kill all the poor". It's easy to wave your hands and say that by the time we approach human-level AI it will necessarily have all the features of human intelligence, but what if it doesn't?

Sometimes various people have referred to "optimization process" rather than "intelligence" to try to get this point across. What if it's possible to build an optimization process that is better, or far better, at solving problems than humans? Planes are much better at flying fast and far than birds, but often don't bother with some of the things that all birds have, like movable wings and feathers. I believe that Eliezer thinks that the things that make us human are feathers, analogously.

Even if you think there's only a small chance of this, the chance that a program more intelligent than humans will have things that humans consider bugs is very high, and the consequences are an existential risk.

fluffster• 17 years ago

>Most of his articles on it seem to be very tl;dr.

This is a good start: http://www.singinst.org/upload/artificial-intelligence-risk....

These videos are good intros to EY's work too:

http://video.google.co.uk/videoplay?docid=611451877200179691...

http://video.google.co.uk/videoplay?docid=-82119137046281951...

rms2• 17 years ago

That's Eliezer. Less Wrong is his site. He also posts here sometimes. Friendly AI is his big idea.

No, there is not a large movement of people opposed to AI research on skynet grounds, Eliezer is the main advocate. I'm certainly willing to admit that Friendly AI is good and important but I think it's perfectly fine to reimplement and (recursively self-) improve Eurisko. There's always quantum immortality.

Some selections from his writing:

http://lesswrong.com/lw/y4/three_worlds_collide_08/ (That Alien Message, my favorite short piece by Eliezer)

http://lesswrong.com/lw/y4/three_worlds_collide_08/ (Three Worlds Collide, novella length fiction)

http://lesswrong.com/lw/r5/the_quantum_physics_sequence/ (Quantum Physics Series)

I'm glad Eliezer is out there thinking the thoughts that he is thinking. Maybe by the time we're closer to AI people will start really listening to him. For now, AI research continues to crawl along.

randallsquared3• 17 years ago

There's always quantum immortality.

That's not the best argument for inspiring confidence, to put it mildly.

sho3• 17 years ago

"That Alien Message"

Thanks for the link. I read it, and it's very well written. What a great metaphor for conceptualising the speed of thought of a super-AI.

That said, the essay still absolutely reeks of unfounded near-paranoia. The hypothetical alien civilisation with the power to manipulate the output of stars - or even to build a simulation of such astonishing power - must surely know the very basics of risk management. And the superintelligent "earth" seemed completely friendly, all they wanted to do was talk. They must have reasoned that "hacking" the other civilisation with self-replicating nano-machines would destroy them. Why on earth would they do that? You'd expect that the one thing an AI would not want to do is destroy unique information.

Start with a great setup and an engaging story, then right at the end take a totally unjustified turn into pure worst-possible-case disaster fantasy. It really just proves nothing, except that Mr. EY does not believe risks can be managed. And really, if you're going to write the "precautionary principle" this large, we really just couldn't do anything, ever. It reminds me of those nuts arguing against the LHC.

"I'm glad Eliezer is out there thinking the thoughts that he is thinking."

The idea of hostile AI probably predates the notion of friendly AI! It's not a new thing and I'd argue it's always been on the table. It just has to be kept in perspective, like any risk scenario from new technology. EY takes it to the most ridiculous all-or-nothing paranoid extreme when we're not even close to the time when we need to seriously weigh up these things.

There would be fantastic, concrete benefits to having even slightly better AI. Trying to retard research because of some ill-justified nightmare "what if" scenario set in the far-off speculative future is ludicrous and, arguably, morally wrong.

gort• 17 years ago

I found this article quite compelling: http://selfawaresystems.files.wordpress.com/2008/01/ai_drive...

Imagine you do have an AI with a certain goal. It seems obvious that:

It will resist being switched off (because if switched off it cannot fulful its goal).

It will try, and likely succeed, in becoming more powerful (to better satisfy its goal).

It will not want to change its goal (because it sees that changing its goal means its current goal, i.e. the one currently motivating it, will go unfulfilled).

Let's reimplement Eurisko

Related Stories