Chess grandmaster Garry Kasparov, who famously beat and lost to supercomputer Deep Blue two decades ago, assesses the arrival of AI for Go's top competitors
MY TWO matches against the chess supercomputer Deep Blue in 1996 and 1997 were called “the brain’s last stand” and compared with everything from the first moon landing to the Terminator movies. I won the first time and, more famously of course, lost the rematch a year later, at which point IBM shut its project down.
Every time a similar challenge hits the headlines my news and social media mentions erupt.
The latest flurry is due to the Google-backed artificial intelligence AlphaGo taking on Go champion Lee Se-dol of South Korea after it beat Europe’s best player 5 to 0. I don’t play this ancient Chinese game so am not qualified to predict the outcome next week, but I do know what the result will probably hinge on and what the future holds for Go.
Computers excel at flawless calculation, our brains at generalities, long-range planning and applying general themes to new circumstances. This contrast makes for interesting battles in the brief windows when humans and machines are evenly matched, as in chess 20 years ago and, apparently, as in Go today.
Early chess machines had blind spots and exploitable weaknesses and the temptation is to target these instead of playing a normal game. I could not resist doing so against Deep Blue. Mind sports like chess and Go require intense concentration and when focus is disrupted by trying to trick a computer you can end up tricking yourself into making objectively dubious moves. As machines get stronger, these are punished.
But the key disparity between flesh and silicon is the mundane machine advantage of relentless consistency. Computers don’t make big blunders, at least not in chess, while a human is only a slip away from catastrophe. No machine suffers complacency, anxiety and exhaustion. When I lost the decisive sixth game to Deep Blue in 1997 I was under huge pressure and played like it. It was the worst game of my career.
Despite that, it was an exciting time, the culmination of interest in mastering the game mechanically dating back to the 18th-century hoax chess machine the Turk. Today AlphaGo represents a machine learning project with real AI implications and deserves wide attention.
Se-dol may be so much stronger than AlphaGo that human fallibilities won’t be decisive yet. Go also has many more possible moves each turn than chess and is less dynamic, factors that work against machine success. But I’m afraid the writing is on the wall. Today, a decent laptop running a free chess program would crush Deep Blue and any human grandmaster. The jump from chess machines being predictable and weak to terrifyingly strong took just a dozen years.
Go, your clock is ticking.
