 Man Vs Machine
Davey Winder explores the moves that brought artificial intelligence out of the labs and into the black and white world of popular science.
Chess and computers have a long and, if you'll excuse the pun, chequered history. Perhaps the first chess-playing machine was The Turk back in 1769, invented by Baron Wolfgang von Kempelen. The Hungarian engineer built his mechanical chess player for an Austrian Empress who was said to be most impressed, as the moustached life-sized model beat all challengers. So successful was the machine, and those that followed, that they enjoyed fame into the 19th century, despite the fact that it was a fraud - a chess master hid in the box base and controlled all the movement.
In more recent years, and deserving perhaps a little more respect, was the memorable match between Garry Kasparov and the IBM-developed Deep Blue computer. This was the first time a computer had beaten a world champion, a grand master, and some might even say the greatest exponent of the modern game.
But was Deep Blue so different from The Turk? Sure, IBM invested big bucks in developing the parallel-processing monster with 400 chess-specific chips and the ability to search 200 million positions per second. Yet there's good ground to argue that it was ultimately still more a case of smoke and mirrors than a true reflection of computerised thought - that Holy Grail of AI that is both craved and feared. Back in May 1997, Kasparov even described Deep Blue as 'playing like a God' in some situations, but the truth was more down to Earth.
The year before, Kasparov had convincingly defeated Deep Blue and so went into the contract negotiations and the match a little over-confident, which is borne out by a closer look at the contractual side of the contest. This agreed that Kasparov would have access to all the IBM machine's public games, essential at this level of chess, where grand masters will often study every tournament match in the opponent's history in order to best grasp their strategies and reveal their weaknesses.
Unfortunately, apart from the one match against Kasparov in 1996, Deep Blue hadn't played any games in public, with all its training and learning going on behind closed doors at the IBM research labs. Kasparov was told there was nothing to provide when he asked for this important information. In effect, he was entering into one of the highest-profile chess matches ever, blind. Deep Blue, on the other hand, had not only seen all of Kasparov's games, but had them sitting in its memory. If this wasn't bad enough, the negotiated match terms allowed IBM to adjust the Deep Blue program between games. The importance of this became apparent during the second game when it chose an unexpected defensive move instead of capturing an exposed pawn. Kasparov and the entire higher echelons of chess were shocked as to how the computer could have had such strategic foresight to defend while ahead, reducing the chance of counter moves by the opponent.
The truth is that IBM tweaked Deep Blue to allow for its failings in the first game. The in-between game tweaking continued and the rest is history, but it left a big question mark over the actual importance of Deep Blue in the field of AI.
Indeed, it may well be the case that the Russian mathematician Alexander Kronrod was right in 1965 when he predicted that computer chess would be the Drosophila (the much-experimented-upon fruit fly) of AI. 'Computer chess has developed much as genetics might have if the geneticists had concentrated their efforts starting in 1910 on breeding racing Drosophila. We would have some science, but mainly we would have very fast fruit flies.'
|