Yesterday I watched a bit of a documentary about Garry Kasparov and how he lost a match to IBM’s Deep Blue. As the story goes, Deep Blue pulled an untypical move in the second game, causing Kasparov to lose. After that, he lost his concentration and failed to win the match. Kasparov, naturally, accused IBM of cheating, using human intervention. I’m not going into all that, except I don’t understand why Kasparov had such a strict idea of how a computer plays. He’s not a Chess program designer, after all.
However, it’s the implications of the match that are of interest. For example, Kasparov’s loss inspired Omar Syed to create a game called Arimaa (Arimaa at BoardGameGeek). Arimaa was intentionally designed to be a difficult challenge for a computer to play. The approach is, however, boring: the game’s difficulty arises from a very wide space of possible moves, which can’t be pruned easily. Free initial setup renders the use of opening libraries impossible. Well, that’s pretty much why Go is difficult for computers.
Another question comes up: what’s different after Kasparov lost the match? Chess programs have for a long time been able to beat most players. Even Go programs can probably beat most Go players, even if their level is lower. I’m not even sure if Deep Blue’s victory is impressive, especially if such high attributes as intelligence are considered. Certainly Deep Blue isn’t even nearly as intelligent as Garry Kasparov. It just plays better Chess (and even that might not be true), and that’s because playing Chess is it’s sole purpose. It was built to play Chess, it wasn’t built to be intelligent.
Really, if someone wants to create a computer that would impress me, I have better suggestions. How about really good natural language processing? Something that passes the Turing test? Or to keep it game-related, a computer opponent that could play any game, given the rules and perhaps some records of past games to study strategy from. Now that would be impressive. I wouldn’t still be worried about the superiority of humans over computers, though. I’ll be worried when the computers learn to procreate without the help of humans…
2 responses to “Games and artificial intelligence”
Couple points –
Deep Blue wasn’t even particularly built to play chess intelligently. It followed a brute-force approach to the game, embellished with tweaks the programmers gave it between games. Not much more than a very (very) clever heuristic search algorithm.
Deep Blue’s successors, such as Deep Fritz, are supposed to be much more interesting in how they approach the game. Apparently, they attempt to emulate how humans see the game (typically in patterns). Don’t know much about them, though, so I could be wrong.
Chess IM David Levy, who won his famous bets about beating all chess computers for (n) years, went on to producing software used in chess computers, then to produce some natural language processing software which won the Loebner Prize ( http://www.loebner.net/Prizef/loebner-prize.html ) in 1997. He’s found commercial applications for this work ( http://www.intelligent-speech.com/ ).