May 24th, 2006
|mathemajician||02:41 pm - Hard AI Future Salon with Yudkowsky|
For those of you with fast internet connections, here is a two hour talk given by Eliezer Yudkowsky on the safety of super intelligent machines and related topics.
I felt that in some parts of this talk he dwelt too much on things that aren't very important, and some of the questions from the audience made no sense at all, however I do think that Yudkowsky did have some good points to make.
|Date:||May 24th, 2006 02:31 pm (UTC)|| |
I would be curious to hear what you think of Mr. Yudkowsky and his project in general. I used to be an avid follower of his writings -- subscribed to the extropy mailing list, his own SL4 mailing list, and even hung out on IRC #sl4.
I have in the past couple years drifted away from all this, partly because I don't make time for endless internettery anymore. Opportunity cost of lurking on mailing lists became too great when compared with things like work and women.
To return to my question: how do you regard Mr. Y and the singinst? How feasible do you think the project is?
I think that the work they are doing is important as I think that AI could be dangerous, but I'm not sure in what way. Unfortunately, I'm not convinced that they know either... but then at least they are trying to figure it out.
I hadn't thought about the problem much myself as I'd always figured that without a solid design that I was sure was going to be intelligent, it was impossible to really tackle the problem of safety. I have thought a bit about the problem over the last week, and even thought I might have made significant progress, after Ben Goertzel challenged me to try to solve some of the issues here. Unfortunately I failed... at least for now.
I used to follow sl4 from time to time, but most of the posts there seem to be a load of rubbish in my opinion.
Do I think the singinst will actually develop AI? I doubt it. If I had to put money on one project getting there first, I'd bet on the Blue Brain project, or some off-shoot of that project.