Computer Go Player Defeats Top Human Player

For about a year in graduate school, I was a serious Go player. I read a few books and even played in a tournament. It's a beautiful game, no question about it, but after wasting so much time just to become a mediocre chess player, I eventually decided not to repeat the process with Go.

One thing I noticed, though, was that Go players were constantly comparing their game to chess. In particular, they were really keen on the idea that Go was more complex than chess. They liked to point out that computers were competitive with the top human chess players (and are now superior to them), but the top Go programs were no challenge for the top human Go players.

Now, as far as I'm concerned both games are plenty complex and beautiful. But it still gave me a little satisfaction to read this:

In a major breakthrough for artificial intelligence, a computing system developed by Google researchers in Great Britain has beaten a top human player at the game of Go, the ancient Eastern contest of strategy and intuition that has bedeviled AI experts for decades.

Machines have topped the best humans at most games held up as measures of human intellect, including chess, Scrabble, Othello, even Jeopardy!. But with Go—a 2,500-year-old game that’s exponentially more complex than chess—human grandmasters have maintained an edge over even the most agile computing systems. Earlier this month, top AI experts outside of Google questioned whether a breakthrough could occur anytime soon, and as recently as last year, many believed another decade would pass before a machine could beat the top humans.

But Google has done just that. “It happened faster than I thought,” says Rémi Coulom, the French researcher behind what was previously the world's top artificially intelligent Go player.

Ha! Take that Go snobs!

Incidentally, top Go players are not usually referred to as grandmasters. That's a chess term. Go typically uses a different system of ranks and titles.

I'm also skeptical about claims that Go is exponentially more complex than chess. That's based on combinatorial calculations of the total number of possible games in Go versus chess. But a large percentage of legal games are ones that would be dismissed as absurd by even amateur players. So the raw number of possibilities is not necessarily a good measure of complexity.

Which is why I am happy to say they are both plenty complex, and just leave it at that!


More like this

"It happened faster than I thought." Always does. I swear I'm beginning to think Ray Kurzweil is right about the rapidly approaching singularity -- " the merging of our biology with the staggering achievements of "GNR" (genetics, nanotechnology and robotics)."

I wonder what kind of literature we'll like when that happens. What type of films & music & paintings & suchlike.

By Bill McNeal (not verified) on 27 Jan 2016 #permalink

Machines have topped the best humans at most games held up as measures of human intellect, including chess, Scrabble, Othello, even Jeopardy!.

Maybe the reason these games are so hard for humans to play, so much so that only the greatest geniuses can keep up with machines, is that they are measures of human ineptitude rather than intellect. That's why it takes the smartest people in the world to play chess competently against a computer.

Compare this to the difficulty engineers are having creating a self-driving car, or even a robot arm that can open a bottle and pour a drink into a glass, an action so easy (for humans) that 99% of children have mastered it by the time they are 5 years old.

By Valhar2000 (not verified) on 28 Jan 2016 #permalink

Maybe the reason these games are so hard for humans to play, so much so that only the greatest geniuses can keep up with machines, is that they are measures of human ineptitude rather than intellect

For scrabble, at least, it's mostly a matter of memory. There is always an objectively highest scoring play, based on the letters you have and what's on the board. The computer can identify it, most people can't, because they simply don't have every possible English word memorized. If you become a semi-serious player, one of the first things teachers will tell you to do is memorize every two letter word (easy, there's only about 30) and three letter word (harder; there's about a thousand). Some of the world's best scrabble players don't even speak English, they just memorize which combinations are legal plays and which aren't.

There is deeper strategy, of course (like thinking about what sort of openings you leave for your opponent; maybe you want to "lose" a few points in order to deny them a juicy triple). But those sorts of strategic concerns only really matter if you're playing someone with an approximately equivalent vocabulary; someone who has memorized (most of) the dictionary and can recall those words as needed will pretty much always wipe the floor with someone with a 'normal' vocabulary just by maximizing their score on every play, no other strategy needed.

For scrabble, it is not just knowing the word, but how well your "content addressable memory works" to find words that fit the constraints.

But a large percentage of legal games are ones that would be dismissed as absurd by even amateur players.

True, but you can say the same thing about chess. For instance, the Fool's Mate: 1 g4 e5 2 f3 Qh4++. That's an extreme case, but I'd say only about half of White's 20 possible opening moves get any serious consideration from anybody even moderately familiar with the game, and most effort focuses on about half of that half. Further down the opening tree, many theoretically possible moves are also obviously dismissed. Things do get complicated in the midgame, but in many positions, particularly if you are defending, there is one obvious best move.

By Eric Lund (not verified) on 28 Jan 2016 #permalink

I agree, but that's precisely my point. I don't think the raw number of possibilities is a good measure of complexity for either chess or go. At some point the analysis tree becomes so vast that it is pointless to compare the relative complexity of the games. I think it's reasonable to say that chess and go are both more complex than checkers or Othello, but to compare one to the other just gets a little pointless.

Those who like to are welcome to find a source of either satisfaction with the sophistication of the software program or disappointment in the capacity of the top player to "defeat" the "computer" in this development but I see it as neither.

For me, the comparison constitutes a "category error" in logic. In no real sense did the computer "play" the game--as the human player did. The fact that computers can calculate finite data sets and determine, from a pre-existing set of game histories, the most probable best move --again, according to pre-established and programmed criteria, already in the computer's CPU, isn't and shouldn't be a marvel to us.

The computer doesn't know anything--neither that it is playing a game or even what a "game" is.

By proximity1 (not verified) on 01 Feb 2016 #permalink

@ #7
"In no real sense did the computer “play” the game–as the human player did."

I'm not so sure about that. Of course it had a backlog of previous games in mind, just as human players do, but then, according to the article in Nature ,which broke the news, "DeepMind then combined this search approach with the ability to pick moves and interpret Go boards — giving AlphaGo a better idea of which strategies are likely to be successful." Sounds a lot like a human player to me. Also interesting, is that unlike the chess program Deep Blue, AlphaGo was not preprogrammed to play go, it learned using a general-purpose algorithm.

As both a chess and go player I agree with Jason that it's silly to call one more complex than the other. The numbers are much larger in go, but I don't think that qualifies it as more complex. Maybe I don't understand what they mean by complex.

My nitpick objection is that, for no reason at all, go is almost always capitalized. Even the Nature article, (and Jason), consistently capitalizes go. You don't capitalize chess, why would you capitalize go?

I capitalize “Go” only because I see everyone else capitalizing it. You'll have to ask everyone else why they do it!

There's no rhyme nor reason for it. You might as well capitalize chess and baseball.

Maybe 'chess' and 'baseball' should be capitalised when they are used as proper names, same as 'God'.

By Paul Murray (not verified) on 02 Feb 2016 #permalink

@ 8: Baseball ought to be capitalized. ;^ )

Tom, How many calculations per second can the computer in this case perform? Tens i of thousands? Hundreds of thousands? In such a case does it make any difference to the kind of calculating which is being operated mechanically that no game histories or set strategies were preprogrammed? In each case a machine compares and operates binary code with no existential awareness whatsoever.

How is that fact finessed?

Here, in our exchange of opinion, you and I have clear understandings of the nature of a dicussion, of differing opinion, argument and reasoned grounds. If one of us were a macine, that could not be the case and there'd be a qualitative difference which sooner or later should become apparent to the human participant but never so to the machine.

By proximity1 (not verified) on 02 Feb 2016 #permalink

@ #12
"a machine compares and operates binary code with no existential awareness whatsoever."

In other words, the machine is not a sentient being, unable to argue, discuss, or give differing opinions. OK, true enough, though I don't see how that's relevant to the topic. Perhaps you'd be happier with a headline that said, "machine solves a puzzle," rather than, "computer wins a game."

"Perhaps you’d be happier with a headline that said, “machine solves a puzzle,” rather than, “computer wins a game.”

No, I really wouldn't. The computer also doesn't really "solve" the puzzle, "sing" the tune, "tell" the story. These are anthropocentric projections of ours onto an inanimate object.

See: "Richard Feynman Computer Heuristics Lecture" @ Youtube. It's both instructive and humorous in Feynman's wonderful way.

By proximity1 (not verified) on 04 Feb 2016 #permalink

@ 14
So what happened then? Both the computer and the human followed the same rules, both made decisions, the computer result was better. How would you characterize it? Words are necessary to describe things.

"So what happened then?"

A machine's operation--in this case a highly-developed digital computer or tabulator/calculator capable of many thousands of binary data sorts and comparisons--proved, banally, and once more, to much for the very talented Go-player for calculating speed or data storage capacity.

It's about as exciting and unexpected as evaporation.

By proximity1 (not verified) on 05 Feb 2016 #permalink

... many thousands of binary data sorts and comparisons per micro-second.

By proximity1 (not verified) on 05 Feb 2016 #permalink

Yeah, I don't think it's going to catch on.

By tomh (not verified) on 05 Feb 2016 #permalink

In reply to by proximity1 (not verified)

The argument that I have seen for Go being deeper than chess, is the range of Elo scoring among ranked players. The different between the best and worst players in Go is greater than for chess, but NOT exponentially greater.To the extent that Elo puts a linear relationship on the average of nonlinear and non-transitive relationships, Go is several times more complex than chess.

By Dan Eastwood (not verified) on 13 Feb 2016 #permalink

Bill McNeal fell for the propaganda, and Proximity1 is correct.

What Kurzweil and his pals are up to is the search for a super-intelligent conscious machine into which they can upload their minds to achieve immortality. That is nothing more or less than a religious quest for a new deity and eternal life of the soul.

Toward that end, any apparent "breakthrough" is touted as a step toward the hoped-for miracle. This is not science: if anything, it's pseudoscience wrapped in technology to make it look "sciency." If you can reincarnate into a computer, you can also reincarnate into a cat.

A computer simulation of a specific human behavior, such as playing chess or go, is not evidence of "intelligence" much less "consciousness" in the computer: any more than the "automata" clockwork simulations of ducks and suchlike that were built in the 1800s, were real ducks (etc.). The builders of automata gave demonstrations on stage, where the clockwork ducks would flap their wings, eat, quack, and poop, but the clever simulations were in no way alive.

Now consider the fact that Kurzweil has numerous True Believers among the Silicon Valley elite, including Sergey & Larry at Google, who have given him a blank check to proceed with his sillyness. One can't help but think of the popularity of Scientology among certain well-known actors in Hollywood. Hopefully it will turn out to be a transient fad.