Watson beats humans in Jeopardy! dry run If you were betting on humanity in the upcoming grand challenge clash between humanity and IBM's Watson question-answer (QA) supercomputer, you might want to start reconsidering your wager. As El Reg previously reported, IBM's Watson QA box took the same Jeopardy! entrance exam a month ago and qualified to take on past …
Great question. No, Watson doesn't have to perform SR here, it gets the text of the message. When? I think at the start of the reading, which gives a substantial advantage to the machine. That's an important distinction and a fair criticism of the process. I expect we will see SR in later matches to make it fairer for fleshies.
This is just an extension of their chess-player with natural language parsing added on. There's no actual intelligence* involved, and if you limited its resources to those available to its fleshy competitors, it would fail miserably.
AI won't be created by chucking massive databases into enormous supercomputers. It will be created by an elegant system that chooses how to process inputs on its own. Let me know when someone's got that sorted.
*on the part of the computer. Its developers appear quite intelligent indeed.
That "elegant" system may not exist. As far as we know, anything able to navigate in the real world is a horrible kludge with bolt-ons left and right and something called the "limbic system" acting up regularly. Bletcherous!
I fear the real world is too complex and real-time for elegance in processing if all what you have got is polynomial time and constant space processing. Deal with it.
When it becomes more cost effective to have this thingy around than an actual specialist, that may be a bit off... They seem to have a similar shelflife, say 15years, I'd hazard.
Well, supposedly you could have a lot of "interpreters" around who each feed their questions to one centrally located QA machine --- however, it then purely depends on the quality of the interpreters, on them accurately understanding the situation and correctly gauging the importance of each factor present.. almost like a specialist would do. Hm...
The Turing test is only a test on whether or not a computer is able to carry a conversation that, to a human, is indistinguisable from a human conversation. It's not a measure for being a proper AI at all. This is most certainly an example of an excellent implementation of a specific type of A.I. (namely an AI algorithm that reduces an enormous search space to singleton solutions). It's just hard to see - right now - how portable it is for other applications
(yes, I have a degree in A.I.)
Remember that Turing was a mathematician and loved the idea of using mental "tricks" to answer hard questions - something he did by using the Turing machine to prove the Entscheidungsproblem.
When you are faced with the question "can a computer be intelligent" it is very hard to answer directly without answering sub-problems like "what is intelligence".
Turing used reductio ad absurdum to turn that on its head: If a computer cannot be told apart from an intelligent person then it would have to be considered intelligent.
That sets some upper bound to what one might consider a test of intelligence. It does not make the Turing Test the one and only test of intelligence.
Answering some general knowledge questions faster than people is not necessarily intelligence. There are many highly intelligent people that would not know the answers. Rather, intelligence involves original problem solving. When dropped on an island could this computer figure out how to make a shelter from banana leaves?
Mimicking human intelligence is far too much of a goal. AI can't even replicate the survival instincts of a fly.
You may choose to define intelligence so that the Turing test doesn't measure it, but it measures what *I* mean by intelligence. (The sad thing is just how many human beings would fail.) In fact, it's probably the only test of intelligence that we know of.
Had the machine built its own database (and written its own rules) by listening to the unstructured raw input that we call "real life", I'd be impressed. As it is, this system appears to have had a "small" set of rules pre-programmed in and then merely demonstrated how far you can go with just a small set of rules. That is usually surprising to a lay audience, but to anyone with a background in the physical sciences, it isn't telling us anything we didn't already know.
You have a degree in AI great, what is proper AI or even just I?
-
For my money Turing's test(s) at least acknowledges the fact that we know very little about cognition so would do better to accept that (at least short term) and judge appearance of intelligence. If current AI thinking has something better, I'd be very interested though..
So the machine gets the entire answer at the start, as a whole blob of text? And the humans get the answer as successive words at the speech rate of a human quizmaster? And the fastest to buzz in with the answer wins?
Anyone else think this gives an unfair thinking time advantage to the computer? You want a fair comparison, post the question on a screen too, and make sure the contestants are all speed-readers.
Not that it's not impressive a computer can do this. But still, if you're going to play a game then at least you need to have everyone going on the same rules.
These conversations are going to look rather quaint in a hundred years' time, when A.I.s are built from trillions of self-assembling nanoscale quantum processors, or maybe something even more spooky and wonderful.
@Charles Manning: the goal of mimicking human intelligence will, in the context of history, seem like rather a short blip. As of 2011, we're still banging rocks together, really.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
