(awefaoefiaef;eif had this post all typed up and then I accidentally deleted it -_-)
Google's computer AI AlphaGo has beaten the 4th-ranked player in the world, Korea's 33-year-old Lee Sedol!
I have been staying up late to watch the official Google streams for Games 4 and 5, so I am slightly incoherent from sleep deprivation!
I didn't udnerstand much of the go, but I loved watching the streams anyway, partly because one of the commentators, Michael Redmond (the only 9-dan from North America), has beautiful stone-handling and he was totally sassing the other commentator.
I learned a couple of new go terms from the commentary, although they are somewhat non-sensical to me: "Nice Aji (taste)! There's good Moyou (pattern/shape) over here!" And I guess I kind of re-learned some terms that I should have learned form Hikago (Ko, Knight's move, Atari...) and a bunch of other random bits of go strategy.
I felt pretty sad after Lee Sedol lost Game 3, but then he said "Today is Lee Sedol's defeat, not humanity's defeat." True! Now I am mostly just excited and hyper :D This is amazing technology, and all this hubbub over go is amazing too!
I was pretty psyched when I saw Aja Hwang, the guy who was placing moves for Alpha Go, holding his stones in the amateurish way that Hikaru does at the beginning of HnG :) Saaaaaaiii!
Other random observations and facts:
-AlphaGo won 3 games in a row, clinching the best-of-five series, then lost game 4. Some are saying AlphaGo has a subprogram that tells it to purposely lose when it's starting to look too scary so that the humans won't unplug it :D
-In game 4, move 78 by Lee Sedol was apparently brilliant. “Lee Sedol played the Divine Move.” —Gu Li.
-Other people are saying "AlphaGo has a weakness! You just have to play the Divine Move!" XD
-AlphaGo played some "nonsense moves" (Redmond's comment) after move 78, although I think the jury is out on whether those were actually "bad" moves. AlphaGo doesn't think like a human go pro--it tries to maximize its chances of winning rather than maximizing its point count. It doesn't play greedy and usually wins by small margins.
-After game 4, Lee Sedol said that AlphaGo is better at playing white than black, which is perhaps why Lee won that game. It makes sense that AlphaGo, with its conservative style, would not be as strong at playing aggressively to make up for the komi (which was 7.5).
-Lee Sedol has a very high-pitched voice XD
-My brother said that when he watched the Game 5 stream live, there were shots of Lee Sedol looking really sad, but later on when he showed me the stream it seems like those shots were taken out.
-AlphaGo first learned by playing strong amateurs on the internet, but after that it mainly learned from playing itself. Isn't that slightly alarming! Stephen Hawking is probably not happy.
-There are two versions of AlphaGo: the stronger distributed version, comprising 1200 cpus, and the individual version, which is just one cpu. Lee Sedol played the distributed version.
-Running AlphaGo is too expensive to allow it to play very often.
-AlphaGo has a system just for time management--to decide things like "is this move important? How much time should I spend thinking about it?"
-The next person to play AlphaGo will be the current top-rated pro, China's 18-year-old Ke Jie. After game 2, Ke Jie blogged, “Even if AlphaGo can defeat Lee Se-dol, it can’t beat me." Later, he said his chances of winning have gone down because AlphaGo learns so fast.
-AlphaGo should probably be referred to as an "it," but I've heard people referring to it as "she"? (Can anyone confirm?)
![[personal profile]](https://www.dreamwidth.org/img/silk/identity/user.png){kind=small}
