Strong AI and the Imminent Revolution In Robotics

An anonymous reader writes "Google director of research Peter Norvig and AI pioneer Judea Pearl give their view on the prospects of developing a strong AI and how progress in the field is about to usher in a new age of household robotics to rival the explosion of home computing in the 1980s. Norvig says, 'In terms of robotics we’re probably where the world of PCs were in the early 1970s, where you could buy a PC kit and if you were an enthusiast you could have a lot of fun with that. But it wasn’t a worthwhile investment for the average person. There wasn’t enough you could do that was useful. Within a decade that changed, your grandmother needed word processing or email and we rapidly went from a very small number of hobbyists to pervasive technology throughout society in one or two decades. I expect a similar sort of timescale for robotic technology to take off, starting roughly now.' Pearl thinks that once breakthroughs are made in handling uncertainty, AIs will quickly gain 'a far greater understanding of context, for instance providing with the next generation of virtual assistants with the ability to recognise speech in noisy environments and to understand how the position of a phrase in a sentence can change its meaning.'"

Comments

but handling uncertainty isn't easy

[rmstar]

Pearl thinks that once breakthroughs are made in handling uncertainty, AIs will quickly gain 'a far greater understanding of context, for instance providing with the next generation of virtual assistants with the ability to recognise speech in noisy environments and to understand how the position of a phrase in a sentence can change its meaning.

Oh, of course. But pretending that these "breakthroughs in handling uncertainty" are just a minor stumbling block is somewhat silly. These are some of the hardest problems in maths right now, and there are no easy solutions on the horizon.

Re:but handling uncertainty isn't easy

[aaaaaaargh!]

Mod parent up.

I'm working in that field and know Pearl's work very well. The problem with uncertainty and current framework is the complexity. Probability theory, possibility measures, ranking theory, plausibility measures, Dempster-Shafer and all these slight variations of the same theme are altogether computationally intractable. Strongly heuristic shortcuts based on implausible assumptions are used (like stipulating independence between random variables for purely technical reasons), and much better ones need to be developed. Human cognition takes amazing shortcuts and AI methods are much too combinatorial in contrast to that.

Moreover, the problem of knowledge representation is still not solved adequately. Yes, there are a few large ontologies like Cyc, but they do not suffice. Basically, a lot of tools are there, but they are disconnected and there is no unifying framework or representation at all. To give you an example from NLP, the kind of tools used by computer scientists (e.g. description logic, event calculus) are practically worthless for doing real-world semantics, and of course logic has the same combinatorial complexity issues.

Breakthroughs will come by combining symbolic AI with connectionist and geometric representations, but only few people work on that (e.g. Smolensky), the math is complicated and not what your average AI/CS guy or computational linguist can handle.

I think what Norvig should have said is that robots with convincible, but ultimately non-intelligent soft AI will enter the consumer market within the next few decades - which is true, but something else entirely.

Getting real with AI

[Max_W]

My new automatic washing machine is an extremely useful robot, even though it does not have legs or hands.

There is only embedded intelligence. A pure intelligence does not even exist and cannot exist.

Why built an AI, which drives a car, if it is quite possible to build an underground transportation network and automate it with AI. This robust technology already exists.

It is easier to send an AI robot to another planet than to a local supermarket. And the problems are not mathematical, but social. The AI is already here and it is bigger than the current society's setup. The social setup and the infrastructure of society are to be changed in order to use it.

Re:Getting real with AI

[Ostracus]

So in plain English instead of making an AI more capable of dealing with greater complexity (like say animals), you artificially constrain the problem set till you get something that present systems (don't need to be AI) can handle then?

Re:How long will it be

[c0lo]

Personally, I'm more concerned about whetever we get space communism or resource contentration at the hands of 0.01% after 99.9% of the workforce getting laid off due to machines doing everything better for cheaper.

With nobody buying (being sacked, can't afford), what's the point of producing? Everything would be relatively too expensive no matter how absolutely cheap.

Re:unintended consequences

[TheRaven64]

Weren't we all supposed to be enjoying 5 months of vacation by now.

Most people who post on Slashdot probably can, as long as you're willing to accept a lower income than you would if you worked full time. I did for several years. I made enough to live comfortably, but not extravagantly, and had a very high quality of living. I'm now 'working' full time back in academia, because now I get paid to work on things I was doing as a hobby before. The standard of living for someone with the same inflation-adjusted income as me now is far higher than when my parents were my age.

by that measure the advancement of robotics probably won't benefit human lifestyle either

Really? I suggest that you try living in a house that contains no technology developed in the last 100 years for a while if you honestly think that...