wired.com — Computers can beat the world's best chess players, but we still can't get them to think like a 4-year-old. Artificial Intelligence has managed to exceed, and fall short of, our grandest expectations. A look back at what's been accomplished in 50 years and a look ahead at where the field is going.
Jul 17, 2006 View in Crawl 4
revengJul 17, 2006
Not to mention, it's still a common belief amongst AI researchers that we cannot create anything which is smarter than ourselves. The best we could hope to do is create something that can learn and recall faster and more accurately. But to make something truly more 'intelligent' would require knowing what it is that makes one intelligent.
darthtrevinoJul 17, 2006
wrong level, my bad
twangoJul 17, 2006
Alice is great at answering fuzzy questions, but start pushing for precise answers and it's back to Eliza again.Specialist systems are just glorified databases. IMHO what's needed for broader intelligence is general problem-solving heuristics (learning without being told the answers) with pattern-sorting (dreamtime) in the background and GIGO analytics. Whatever all that means.
kurtu5Jul 17, 2006
And RevEng, how do some dumbass parents have smart kids?Answer: kids can generalize and learn, thats what AI is shooting for.
cyberdactylJul 18, 2006
"That assumes that passing the Turing test has any significance with regard to intelligence. Simply because Turing thought so, doesn't make it so."Hey screw you, many of the bosom babes at HOOTERS can pass the Turing Test for close to 20 minutes at a stretch.
craniumJul 18, 2006
@inkyblueSorry, but you're incorrect. Take a theory of computation class. When non-computability is proven with respect to a Turing machine, it applies in the abstract, which means there is no combination of turing instructions that will satisfy the requirement, regardless of how complex the program is.<a class="user" href="http://en.wikipedia.org/wiki/Computability_theory_%28computer_science%29">http://en.wikipedia.org/wiki/Computability_theory_%28computer_science%29</a>
dubwaiJul 19, 2006
Brain reseach has created neural networks. It's easily concievable that we can create a computer based model of the human brain. This brain would not only have more resources and have faster switching, it would be able to keep reorganizing itself indefinitely. Our brains start to stop doing so are around 35. In addition such an electronic brain would not be doomed to one day die as all human brains are. The electronic brain could start 'life' dumber than we and in a few years or decades be smarter than any human ever born.
inkyblue2Jul 27, 2006
@craniumagain, that proof only applies to programs which try to solve the problem formally according to the system of rules used when setting up the proof. brains don't do this. AI will not do this. AI will look at the problem, recognize relevant features, compare it to similar problems, try solutions that have worked in the past, etc. AI will use heuristics which do not rely on formal correctness.bury a quarter on the beach and ask a turing machine to "solve" for its location. you can't do it formally without digging every conceivable hole. tell an old guy with a metal detector to find the quarter and he'll start by cutting down the problem space using reverse psychology and looking for places where the sand looks like it's been disturbed. two entirely different characterizations of the problem."Take a theory of computation class."don't be condescending.