A.I. to compete on Jeopardy

http://www.pcmag.com/article2/0,2817,2346066,00.asp


04.27.09

IBM said Monday that it plans to produce "Watson," a computing system designed to parse questions intelligently and respond in the manner in which it was intended. The goal? To knock off human contestants in "Jeopardy," in much the same way IBM's "Deep Blue" computer beat Garry Kasparov in chess. Watson will be forced to interpret the question, process puns and other word games, search through its database and determine the correct answer, all within less than a second -- the reaction time of "Jeopardy" players. The machine will not be connected to the Internet, but will have to parse its own database of content. While IBM has apparently convinced Sony Pictures to stage a match against a Watson system, details of the taping (or if it will be taped) have yet to be disclosed. A Sony spokesman could not be reached for comment. "The essence of making decisions is recognizing patterns in vast amounts of data, sorting through choices and options, and responding quickly and accurately," said Samuel J. Palmisano, IBM's chairman, president and chief executive officer, in a statement. "Watson is a compelling example of how the planet - companies, industries, cities - is becoming smarter. With advanced computing power and deep analytics, we can infuse business and societal systems with intelligence. This project is the latest example of IBM's longstanding commitment to fundamental research and to overcoming 'grand challenges' in science and technology." Like Deep Blue, IBM plans to take a massively parallel approach toward the problem, considering solutions, prioritizing some and discarding others. It's a sort of holistic solution to a problem: part hardware, part software, brute force plus intuition. It's also an interesting test of what search is evolving towards: not just matching keywords, but trying to intuit what the questioner is also asking. Getting to the heart of the question is also a part of what's known as "business intelligence," or trying to find the interesting needles in a haystack of seemingly random data. One can imagine similarities between these two phrases: "This highly prized form of corundum gets its red color from minute amounts of chromium." "This product commanded the highest margin during the month of June from our "'Lifestyle' product line." In this case, the answer could be the same for both: "rubies". But could "Watson " also parse this clue: "Apes might enjoy eating these legumes." On its own, no. But under the heading of "anagrams," possibly. (The answer? "Peas".) The New York Times reported that the Watson prototype for some reason interpreted a "sheet" as a fruit. Small glitch or fatal flaw? We'll find out when the show airs.

How much longer until we have a Turing Test passer?

How much longer 'til I can watch people compete against a computer, that is damn interesting!

How much longer 'til I can watch people compete against a computer, that is damn interesting!

Ever seen anyone play a video game?

On a more serious note: unless we can tell absolutely no difference between a human and a particular AI implementation, I doubt that it will ever gain social acceptance as a strong ("true") AI. A purpose-built Jeopardy-playing machine may be a spectacle to behold. However, it's not like anyone is going to congratulate such a "narrow", weak AI after its potential victory. Remember Deep Blue? Would you ascribe credit for its victory over chess grandmaster Garry Kasparov to Deep Blue itself, or to the engineers at IBM that built it?

In general, I think part of the problem with gaining social acceptance for strong AI is that any program achieves exactly what it's told to achieve. For instance, look at Diablo 2. It's a fun game to play, and the average player will find it very unpredictable. It's not like you can ever experience two identical games in it with the exact same items being dropped when killing Mephisto, right? Wrong. The "unpredictability" is usually based on something as simple as letting the program select one of a limited number of outcomes depending on what time it is. Since there is a discrete number of possible games, the exact same game conditions can be found again after a finite amount of time. Thus, if the same time-based game conditions are found and the game play is automated by a bot, the outcome and item drops will be exactly the same. Similarly, a soccer match between robots will have exactly the same outcome if all of its initial conditions are met. This is based on the assumption that there are no external influences (such as wind gusts) that can alter the trajectory of the ball, et cetera.

Of course, with more effort, you can use seemingly true randomness based on the white noise of a sound card, entropy of molecules, quantum decoherence and such. However, it can still be argued that a program can never escape its order of complexity to become something truly unexpected. That being said, Diablo 2 can still be very fun to play and it certainly appears truly random despite its limitations. If the technological singularity ever occurs, it will be mathematically beautiful but most likely still predictable. However, it will be predictable in such a complex way that that an unaided human is not likely to understand it.

Edit: Also, I find it preferable if news-containing posts don't start off with a link and a huge quoted segment. It doesn't look good on the portal as it only displays the first 750 characters.

Ever seen anyone play a video game?

Jeopardy > Madden football

Zelaron > Jeopardy.

Zelaron > Jeopardy.

Pre Skurai of course.