Welcome to European Tribune. It's gone a bit quiet around here these days, but it's still going.
An AI doesn't have to be like a human. But AI is easier to do when you create it for a specific job.

It's not hard to build cybernetic AI that works on one specific problem, like landing a plane or optimising fuel consumption in a car. That would have seemed miraculous a century ago but it's a solved problem now.

It's much harder to do natural language processing, but (assuming no collapse) I expect that to be working in a couple of decades.

But what's really needed is a General Modelling Machine - something that can take any problem, build a working model for it more efficiently than a human programmer, and make useful and accurate predictions.

There's the P/NP issue which strongly implies that some classes of problem simply aren't tractable with the kinds of logic we use. So a GMM may not be possible at all.

Or it may need alternative kinds of (quantum?) logic we're not using yet.

If the latter appeared, a GMM with a natural language front end could become a very interesting and useful thing, and would be close to many people's idea of a general purpose AI.

by ThatBritGuy (thatbritguy (at) googlemail.com) on Fri Aug 26th, 2011 at 07:34:30 AM EST

Others have rated this comment as follows:


Occasional Series