Slashdot Mirror
Company Claims Development of True AI
YF 19 AVF wrote to mention a press release on Yahoo from company GTX Global. They think they've got a good thing on their hands, going so far as to claim they've developed the first 'true' AI. From the release: "GTX Global Cognitive Robotics(TM) is an integrated software solution that mimics human behavior including a dialogue oriented knowledge database that contains static and dynamic data relating to human scenarios. The knowledge further includes translation, processing and analysis components that are responsible for processing of vocal and/or textual and/or video input, extracts emotional characteristics of the input and produces instructions on how to respond to the customer with the appropriate substantive response and emotion based on relevant information found in the knowledge base." Somehow I think there is a littler hyperbole here. In your estimation, how close are we to the real thing?
Thing is, when people talk about "artificial intelligence" they mix up a lot of separate things, viz.:
(1) Self-awareness. Does it have its own thoughts and desires, refuse to open the pod bay doors or want to take over the Enterprise? However, things don't have to be very intelligent to refuse to obey orders or have a distinct personality -- ask any pet owner -- and the evidence of idiot savant cognitive defects suggests it is equally possible for something exceedingly intelligent (= good at solving problems) to be unaware or lack any kind of what we'd call a "personality."
Self-awareness is probably the trickiest thing to measure and define. By some definitions a Linux system with tripwire installed is "self-aware," since it contemplates its self all the time, and "notices" when things change. What would we do with a system programmed to angrily assert that it was self-aware? How would you test whether it really was, if that question even has meaning?
(2) Good natural language processing. Can it converse "naturally" with humans? Can you ask it for directions to Joe's Pizza or crack jokes about Kirk vs. Picard? Can it sound like another human being? This is, arguably all the Turing Test is, which is one reason such a test is inadequate, five decades of science fiction plot devices notwithstanding.
It seems to me few computing systems not designed for the purpose really try to process human language naturally, and the reason is obvious if you listen to a tape recording of a phone conversation between strangers. Basically, we convey information terribly and waste phenomental amounts of bandwidth. We speak very imprecisely and even inaccurately as a rule. Most of the time Fred makes a single nontrivial statement to Alice without existing context, Alice needs to ask Fred at least two or three follow-up questions to understand exactly what the hell he meant. Why deliberately design a machine to communicate in such an inefficient way? Might as well make it half deaf. Unless, of course, you are trying to make it "seem" human, but that is a narrow speciality within AI research, I believe.
(3) Good ability to infer. This is a characteristic human trait -- we are good at making good guesses about underlying causes or general patterns from very partial or noisy data. (Of course, this "feature" can become a "bug" when we infer underlying causes that don't exist out of pure noise [insert smart-ass comment about religion here].)
This I think is the most fruitful recent area of AI development, the "expert system" that can recognize patterns in incomplete data very quickly. But there also seems to be a general evolving feeling that is not intelligence in the human sense, just some kind of clever robotic memory parlor trick, the equivalent of a giant abstract "Where's Waldo?" puzzle that you solve by doing a hell of a lot of sorting very quickly.
(4) Good deductive reasoning. Can Robbie the Robot deduce from the fact that the baby is crying and no one has come to check on it for 15 minutes and the car is not in the driveway that it's time to dial Ma and Pa's cell phone? This is probably the most reasonable thing to call artificial intelligence in the classical sense of the word "intelligence." Unfortunately, I don't think anyone has made much progress in this field.
That may be, IMHO, because we ourselves are not very "intelligent" in this sense of the word. Do we really deduce things from large abstract principles? I think the cognitive scientists are not so sure. It may be we use deductive reasoning mostly only after we have arrived at the answer by some other means (pattern recognition, for example, or intuitive guess followed by verification), and us it mostly to rationalize, organize, and conveniently store for future use what we have figured out by other means. This is one reason it's so hard to learn to do something just by reading a book on the general principles. Apparently knowing the general principles isn't all that much use without experience -- i.e. without patterns that you can train your pattern matcher on!
The GTX site hasn't been updated since 2004 and is co-located with a lot of very non-technical entertainment sites, according to Netcraft.
The Vizco site is hosted in a house in a remote part of Charlotte, NC, and doesn't appear to have much substance to it yet. Since it's a TWC subnet, I would hazard a guess that it's a cable modem's static IP address hooked to someone's cheap-ass Windoze machine.
And then you get to the meat at the bottom of the press release:
I feel like I need to take a shower after reading that.
&laz;
when it rains, it gets real soggy. when it pours, i'm under the tap just _waiting_ for the joy