go pro beaten by supercomputer – so what?
I’ve just read on slashdot (here is the article):
“…the go program MoGo, running on an 800-core supercomputer, beat 8-dan professional go player Myungwan Kim in a 9-stone handicap game. Most in the audience were shocked at the computer’s performance; it was naturally assumed that the computer would be slaughtered, as usual. Go is often seen as the last bastion of human superiority over computers in the domain of board games. But if Moore’s law continues to hold up, today’s result suggests that the days of human superiority may be numbered…”
Well, I don’t think so. First of all, it isn’t about supremacy. I don’t care a bit if a computer engine beats me at any game. I play go because I like it, because it makes me think in a very unique and complex way. I’m not playing for winning, or the feeling of supremacy over a machine. Second, it’s very important no note that it wasn’t the program’s intelligence or knowledge about the internals of go that beat the human player, it was it’s brute computing capacity. There are basically (a very shallow categorization) two approaches to programming game algorithms: first is simulating real thinking, which is way too far to be realized, if ever(!!), while the second approach is finding profitable moves (in this case) without any deep understanding of the game’s nature. For chess, having the variations of next moves from a certain state on the board much less than of go, this evaluation-enumeration can be done in a fairly deep level, so very good moves can be calculated at low (nowdays, with fast computers) costs. For this go playing engine to require 800 cores – well this shows how much more deep go is. And of course there’s the 9-stone handicap. Those who play go know how enormous advantage 9 stones can be – it means a difference of many levels. That’s one thing that’s so beautiful about go – with a proper handicap, an amateur can have a fairly fun and “even” game with a top-ranking pro. In chess, this is impossible. The go playing engine wasn’t beating an 8-dan pro – it was more like playing against a strong amateur. For pro ranks, the estimated difference between a 1-dan pro and a 9-dan pro is about 3-4 stones only, which is roughly equivalent to the number of advantage stones to be given for an equal play.
The article is correct in one thing, though – as faster and faster computers/clusters appear, go engines will get ’stronger’ – but this means nothing. There’s no go engine yet which is able to play go – only ones that can search effectively over huge decision trees. In this meaning a go engine is not a bit different from a chess engine or a tic-tac-toe engine. Those who have played against computer go engines can tell the difference of playing level, style from a human – there’s much elegance in human go, a lingering deep meaning, there’s such beauty to the game which cannot be expressed or simulated by any program.
One last sentence: I’m saying these as a professional programmer and as a former AI-enthusiast.
Additional comments powered by BackType
