Monday, March 21, 2016

And Then They Came For The Go Masters

I am often surprised by how many people are not at all familiar with the game of Go.  It is more ancient that Chess and both easier and harder.  It is easier because there is a pretty simple rule set and only one type of piece.  It is harder because the number of possible moves on any turn is much higher than in chess.  On the first move in a game of chess there are 20 possible moves, most of which are kind of dumb.  In Go, there are 361 possible opening moves.  You start branching from there and the number of possible sequences in Go baloons tremendously.  According to the video above there are more possible games of Go, than there are atoms in the universe.

At any rate, I pretty much suck at Go, although that does not prevent me from being fascinated by it.  One of my great regrets is that I may never get a chance to play my friend Georg.  Georg and I formed an oddly close bond, given that we only spent two days together.  They were profoundly moving days, though.  The Sesquicentennila of the Battle of Antieram, the deadliest day in American history and the Sesquicentennial of the stacking of the arms of the Army of Northern Virgina.  In both cases, we were there on the spot and in the latter case participated in the ceremony marching along with reenactors who stacked arms.  I chose to be with that group because I preferred to be active, Georg because of his odd affection for the Confederacy.

If you took 100 college educated Americans at random, unless you got lucky, they would likely know less in the aggregate about the US Civil War than Georg, who grew up in Germany and learned zero about it in school.  He developed his interest through intensely complicated strategy games and in the pre-internet days of yore in the eighties, it was hard to learn much about the US Civil War in Bavaria.  Somehow he did though and his interest was so strong that he spent five years worth of vacations being a "real timer" at Civil War Sesquecentennial events.

At any rate Georg works in tech and is, I am sure, probably a much better Go player than I am.  I felt I had to share his thoughts on the latest news.

Commenting on a Go game with a friend he wrote:

 Without exaggeration one can say that this was the last day for the world of Go as we knew it. What has happened since Wednesday morning in Seoul is the advent of a new age. The Go community was fascinated in January when Google told the world that it developed Alphago, a program that beat Fan Hui 5-0 in even (i.e. no handicap) games last October. That was a major leap, no doubt. Fan Hui is a professional level player, one of the strongest in Europe, but he is far away from the top professional level. Alphago played numerous weak moves in that match. Analyses by top professionals predicted a clean 5-0 sweep for Lee Sedol in March.

Apropos Lee Sedol: If an armada of aliens came to our solar system and would challenge mankind to a game of Go for the possession of Earth, we would nominate Lee Sedol to play on our behalf. We just got blown out of space.

In the first three games Lee Sedol has not just been defeated, he has been humiliated. Alphago seemed to be toying with Lee. The way it handled Lee's desperate tactics to bring his group to life on the lower edge of the board in game three seems almost arrogant - if a program could feel something like arrogance. It even simply ignored one of Lee's moves.

After these first three games, Lee's win in game fourth came as a surprise. Lee managed to turn a game around that looked clearly lost with an amazing and unlikely tesuji that very likely will find its way into Go lore (if you are curious: it's move 78). With this play Lee posed a problem to Alphago it was unable to handle. There may be be a systematic weakness behind Alphago's loss, maybe some "over-the-horizon"-effect. But if even players of such caliber as Lee can set up such positions against Alphago on purpose is a completely different question. Anyway, there is very little doubt now in the professional Go world that Alphago is significantly stronger than any human player. Nobody saw this coming, probably not even the folks behind Alphago. It is playing significantly stronger than against Fan Hui last October, too. Even more frightening: It could only have learned this by playing against itself. While Lee's victory in game four brought some relief, the Go community worldwide is in a state of shock right now, nevertheless. It will need time to sort this out.

But while the Go player inside of me is quite frankly shocked, too, the mathematician and computer scientist in me is rejoycing. What a triumph of the human mind! While Google put a lot of computing power behind Alphago, the true core of its strength must be in its algorithms. The folks at Deepmind have somehow combined two neural nets, tree search and monte carlo methods into one fine tuned monster of a Go program. I have not the slightest idea how they pulled that off. But they did.

And then there is the professional IT consultant inside of me, too. And he just wonders how mankind can produce something amazing like Alphago - admittedly with a lot of expert knowledge and money - and at the same time fail so miserably at relatively simple IT challenges in the modern business world - with at least as much expert knowledge and sometimes even more money on hand...
Hail to our new machine overlords!

Peter J Reilly CPA can be beaten in Go by a bright eight year old with some good instruction.  

No comments:

Post a Comment