Thoughts on AlphaGo
1. There is an undeniable poignancy to this moment. But we shouldn’t let ourselves fall for a tragic “man vs. machine” narrative. First of all, AlphaGo is, in many ways, the most human human – it’s a human project developed by a bunch of people with human values not least of which is a genuine fascination with and appreciation of Go. If anything, AlphaGo represents the power of collective action over individual effort. It’s not the story of a machine beating a person, it’s the story of a bunch of people pooling their resources and bringing all of their best problem-solving techniques to bear on one of the world’s most beautiful problems. Looked at in this way, AlphaGo resembles, not so much a mechanical version of an individual player, but a hyperspeed version of the whole history of Go, a process of learning by playing millions and millions of games, recording wins and losses, accumulating data, getting smarter as patterns, heuristics and strategic concepts emerge inexorably over the course of generations. In this sense AlphaGo is like the game of Go itself, a convolutional network in which Lee Sedol is a single node, it is as human, and as alien, as we are.
2. We are going to learn a lot about Go. Consider for example, the surprising 4th-line shoulder hit from game 2. Or even better, take AlphaGo’s infamous “slack” moves, its strange habit of seeming to take its foot off the gas in the end game, passing up bigger plays in favor of smaller ones. What makes these moves look bad to us is that we are used to using the heuristic of “maximize score” as a proxy for the harder problem of “maximize chances of winning”. It’s a perfectly natural thing to do, and we’ve been doing it forever, but AlphaGo demonstrates that it isn’t, in fact, the best way to play. Are we making this mistake in other games? In other non-game situations?
3. From this outsider’s perspective Chess is more interesting than ever. Carlsen seems like a cool world champion, I love watching ChessNetwork stream blitz, and freestyle tournaments demonstrate the potential of human/AI collaboration. Instead of having a chilling effect on the world of Chess, Deep Blue’s victory over Kasparov seems to have injected it with new energy. I don’t see why the same thing wouldn’t happen with Go. I think it’s possible that the next 10 years will be an exciting and eventful time for the game.
4. Elizier Yudkowsky had some very interesting comments regarding what AlphaGo can show us about AI in general, a topic that might be among the most crucially important subjects of the next 100 years. It’s remarkable that something so ancient has secrets to tell us about the future. How strange and beautiful games are.
5. If you want to search for information about Go on the internet you should search for “Baduk” (Go’s Korean name.)
6. I made $100 by betting on AlphaGo. I thought it was close but eventually decided that the AlphaGo team had the best information regarding the program’s strength and wouldn’t have scheduled the match if they didn’t think they had a good chance of winning. Don’t call me Gaius Baltar, remember I was betting on humans, not against them.