Strong AI and the Imminent Revolution In Robotics 242
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.'"
Re:but handling uncertainty isn't easy (Score:4, Interesting)
Re:unintended consequences (Score:5, Interesting)
Weren't we all supposed to be enjoying 5 months of vacation by now....
Well, we are, even more that 5 moths. Except.. it is called unemployement.
by that measure the advancement of robotics probably won't benefit human lifestyle either. Somehow we'll all end up as slaves to the machines.. if we aren't already!
Slave yes.. not to the machines, but to the banks... and, quite frequent, this include the machines/robots owners.
They solved the failure problem? (Score:5, Interesting)
Re:How long will it be (Score:5, Interesting)
Re:Getting real with AI (Score:4, Interesting)
On the other hand, the technology for underground delivery and transportation networks does exist. It would be expensive to build? So what? Let us pay.
Such a system would not only be able to use AI, it will be AI in itself, an embedded intelligence. Besides, from ecological point of view it would be at least 10 - 100 times more safe.
Re:How long will it be (Score:5, Interesting)
Of course, AI robots by themselves do not solve the problem of resource scarcity.
Asteroids do.
And people will continue to grow exponentially
Don't worry, they don't and aren't.
so on a fixed earth, you'll always reach a point where not everybody gets a basic standard of living.
Its quite doable. And if you look at population trends in societies nearing post scarcity status, like Western Europe, it becomes clear that the best way to head off any such hypothesised difficulties is to provide a decent standard of living to as many people as possible as quickly as possible.
Re:but handling uncertainty isn't easy (Score:4, Interesting)
Engineers and mathematicians have developed partial solutions to the sensing and data extraction problems over the last 10-15 years, so things look good in terms of rate of development. It doesn't mean that there will be a robot than can perform task X by 2022, but it does mean that robots of 2022 will be able to perform a number of tasks that today's robots aren't able to perform.
My gut feeling is that by 2022 there will be experimental robots that will do about half of all household work poorly, but they will be the price of a luxury car and they will cause more trouble than they solve. I'm more optimistic about guide robots as a gimmick to impress and entertain visitors in places like museums, theme parks and corporate headquarters. All they have to do is navigate without crashing into anything (a largely solved problem) and say scripted things at certain instances (a compeltely solved problem) and respond with facts to verbal questions (another largely solved problem).
Re:Ray Kurzweil (Score:4, Interesting)
So you see, with each technological innovation, the capabilities of robots don't increment slightly; they jump to a new height altogether. I don't know if anything like a "sigularity" will happen in the next 50 years, but I suspect the difference capabilities of robots from 2012 to 2062 will be much greater than the difference between robots in 2012 and 1962.
Disclaimer: I am also someone working to implement "the singularity"
Robotics is getting there. Money works now. (Score:5, Interesting)
Robots are starting to work in unstructured situations. I was there at the moment when this was recognized - the 2005 DARPA Grand Challenge at the California Motor Speedway in Fontana, CA. That's when everything changed.
The 2004 Grand Challenge, remember, was a pathetic joke. No vehicle got further than 7 miles, and that was CMU's. The CMU approach at the time wasn't even really autonomous. Entrants got the route on a CD an hour or so before the start. CMU had imagery of the whole area and tried to plan obstacle avoidance manually just before the start, using a huge team of people in a semitrailer full of workstations. Didn't work; the DoD people in charge had moved some obstacles during the night. And that was the best result. One vehicle came out of the gate, turned hard, and ran back into the starting gate. One flipped over. The big Oskosh entry demolished a SUV parked as an obstacle to be avoided. The whole thing was embarrassing.
DARPA was very displeased with the performance by the universities that had long been receiving DARPA funding for robotics. It was quietly made clear to some major CS departments that their performance had to improve or funding would be cut off. That's why entire CS departments were suddenly devoted to the DARPA Grand Challenge in 2005.
In 2005, things were completely different. Everybody who got that far had already been through an elimination, and every vehicle at the 2005 challenge was better than any of the 2004 entries. There was considerable press coverage, and at first, the press treated it as a joke. But suddenly there were over 20 vehicles running around autonomously, and they weren't crashing into stuff. When multiple vehicles finished the course, it was viewed as a triumph.
Finally, the state of the art had reached the point that money and determination would get problems solved. That wasn't true in the 1980s. NASA threw over $100 million at the Flight Telerobotic Servicer project, and got nothing that worked.
Now check out the DARPA Humanoid Challenge. [fbo.gov] (There's much dreck about this on blogs and in the popular press. Read the DARPA announcement instead.) They have an approach that's likely to work, and demand simulated demos (in their simulator) in 9 months, with demos on real hardware in 18 months. I personally think they'll get something able to do most of the mobility tasks and some of the manipulation tasks in that time. Useful humanoid robots will be a lot closer in two years.
Price will still be a problem. But not an unsolveable one. These things could be brought down to the price of an SUV, if not lower, through production economies alone. The parts count is probably lower than that for an SUV.
Re:Ridiculous (Score:4, Interesting)
I disagree wholeheartedly with most of what you wrote.
The thing you get right is that it no longer is possible to know every fact about everything. The last known person to have done so was Pic de la Morandière and that was over 150 years ago.
With respect to fields involving increasing specialized knowledge nowadays, however, I simply beg to differ. The real issue is an inflation of know-how that adds little if anything to the pool of relevant knowledge. It occurs because, for all of history since the ancient Greeks including today, there have always been more scientists alive in any given year than there have been in recorded history. Chew on this fact for a moment, and consider how to train their higher level peers, we require them to come up with an original research thesis.
Most published work and research are simply rehashing obvious consequences of things long known. Rare indeed, is the study that pops out because it identifies an edge case where the results contradict what is expected. Recall, as an example, the study that suggested neutrinos might go faster than light. Physicists the world over instantly heard of it. Subsequent refinements eventually debunked the initial results as a measurement error. Sum of additional knowledge? Big fat zero: nothing goes faster than light. The same, boring and century old theory of relativity.
It's not all bad, mind you: something interesting occasionally does comes out of this farce. For instance, a study on how an erection works can lead to insights in how to engineer structures [ted.com]. This makes the whole process tolerable and, in a sense, interesting for the curious.
To argue that every little fact counts, however, is lunacy. You need to discriminate, synthesize, retain key elements, and off you go. You're a specialist. And to hell with the bozo who is so neck deep studying eye retina that he forgets it is a brain outlet. He has nothing interesting to tell you beyond implementation details.
Now, I've absolutely no clue whether the next 10 years will yield a strong AI. I haven't followed AI in a while, preferring good old history. I do know two things, however. Firstly, that a strong AI is around the corner since about 1950. Secondly, that mathematicians stormed the field of cognitive science and linguistics roughly 20 years ago, ignoring the established quacks such as Chomsky and turning the field upside down. Fast forward 10 years, and we were training robots to train other robots to do tasks. This was inconceivable 10 years earlier. Who knows... Not you, nor I.