First, apologies for not writing very many posts this month. I have minor hardware issues. The power cable for my laptop is not working any more and I have ordered a new one, but it hasn’t arrived yet. So for the last week I have been turning on my laptop very sparingly, to conserve the charge it has on it. I can still get on the Internet, write my blog, etc., from my Mac, but I don’t have Chessbase (because Chessbase isn’t made for the Mac) and that means less ability to upload game positions. I could use Shredder but the appearance would be different from usual, so I prefer not to do that.
So partly due to my own stubbornness, I am limited to topics that don’t require any graphics. But I have a good one today!
As some of you may have heard, Google has developed a go-playing program called Alpha Go, which has challenged one of the world’s top professional go players, Lee Sedol, to a five-game match beginning on March 8. To chess players, there is a certain amount of deja vu here. Will this be like the Garry Kasparov – Deep Blue match in 1997? Or will it be more like the 1996 match, which Kasparov won fairly easily?
One thing seems sure: the computers are coming. Go has long been considered the most difficult human game for a computer to master, and until recently people were saying that it would still be at least ten years before a computer could challenge a human professional (let alone one of the top players) on even terms. But AlphaGo has radically altered the battlefield, with a 5-0 victory in October over the European go champion, Fan Hui.
To me, the AlphaGo-Fan Hui match was perhaps most reminiscent of the Deep Thought-David Levy match in 1989. Levy was an International Master who had predicted in 1968 that no computer would beat him within ten years. He won that bet, but by 1989 it was clear that computers were better than International Masters and the match between him and Deep Thought was almost a formality. I think that Fan Hui is probably comparable to an International Master; on a world rating list I found online he is #370. By comparison, Lee Sedol is #4. So the question is: can AlphaGo progress from beating an IM-type player to beating a Levon Aronian-type player in five months time?
I don’t think so. But it should be interesting.
I’ve been commissioned to write an article about the upcoming match for Science magazine, which is exciting because it’s the closest I’ve come to combining my “real” career (science writing) with my avocation (chess). If any of you know of people whom I ought to interview — especially people who are “fluent” in both go and chess — I would welcome your suggestions!
{ 4 comments… read them below or add one }
We have a Chess Master Go Master in Colorado named Larry Leeper.
Tiger Hillarp Persson is a GM and a 1 dan Go player (last I checked), which I guess corresponds to about FIDE 2000 in chess. I don’t know if that’s high enough for your needs. He has a blog with a contact form.
If you want readers, I am 2000 USCF and 4 kyu AGA, which means I’m not incredibly skilled at either game, but know a reasonable amount about both and could confirm that everything you said about Go makes sense.
A few weeks ago on a Facebook thread I saw (and participated in) a debate about this topic that probably would interest you. I don’t recall right now whether it was on Alex King’s personal page (he’s a teacher on-staff at the Marshall so you probably at least know of him), or within a FB group probably Chess Connections.
The debate involved relative degrees of complexity (for algorithmic solution purposes) among chess, Go, and poker. (I brought in bridge, as well.)
The only reason I mention it now is that someone who either had professional-level knowledge of the mathematical theories of complexity, or did a decent job of faking/sounding like he had such knowledge (this debate played out on the lanes of the Information Sewagehighway, after all), made a few posts in which he stated that:
1. Poker is proven to be mathematically more complex than either chess or Go – not in the obvious sense of how much math is required to identify the objectively best “play” in any given situation, but in the probabilistic sense that incorporates human behavior and psychological strategy into the decision mix. He used a phrase such as “game-theory complexity” versus “computational complexity,” or something like that. That’s why I thought he might possibly have real academic/research knowledge in this field.
2. Computers have fully solved checkers (only about 10 years ago, despite most of the public including the chess-playing public wrongly believing that checkers was “solved” by an algorithm in the 1970s or earlier), have mastered chess, and are getting closer to mastering Go, but are very far indeed from mastering poker.
3. He had what sounded like a good comeback when I pointed out that of course chess fell before Go or poker, but not for the reason he had stated. I noted that pros of all stripes have been attacking chess with algorithms since at least the 1940s (Turing), and that endeavor gained high-level government support (Defense) in the ’50s and then commercial support in the ’70s, all the way up to today’s thriving mass consumer market for the strongest possible chess-playing software. But he responded that Go and even poker in fact have had massive amounts of programming effort devoted to them – more even than chess has, he said.
So, Dana, since I know that you are a professional mathematician who also examines these kinds of issues for public consumption, I thought you would be the one to confirm or refute those claims. Email me if you’d like to know more.
Jon Jacobs
Tom Bartell is an IM from Philly who is also 5 dan in Go. You should be able to reach him on Facebook, or through the chess shop he runs: http://mainlinechessandgames.com/