00:00When we point the AIs at really difficult problems
00:03where none of the standard techniques apply,
00:06they are still very, very bad.
00:07I mean, they're just randomly guessing.
00:09But they excel at breadth.
00:11So if you point them at a thousand problems
00:14of various difficulties,
00:16some may be just too hard,
00:17but there will be some which actually
00:19are within reach of existing methods.
00:22And there's some method out there in the literature
00:23which will solve your problem.
00:25Or maybe you have to combine together
00:26two separate methods.
00:27And it's just that there's not enough human experts
00:29to look at all these problems.
00:31And the human experts that do look at these problems,
00:32they may not realize that there was this obscure paper
00:35from a journal in 1970
00:38that actually has the key idea that will solve this problem.
00:40They don't have the patience or the time
00:42to sort of go through all the different combinations
00:43of which technique might work on which problem.
00:46But the AIs, they were somewhat randomly,
00:50they would take sort of educated guesses
00:52as to what techniques might work for a problem.
00:54And some of them would be stupid,
00:56but some of them might work.
Comentarios