So a Computer Beat Lee Sedol and I Watched the Whole Thing

I had every intention of going to bed at a normal hour last night. Instead I was on the couch at close to 1 AM with a bag of pretzels watching a YouTube livestream of a Go match, because AlphaGo played Lee Sedol today (well, today in Seoul, which was the middle of the night here) and I could not physically make myself close the laptop.

For anyone who somehow missed it: AlphaGo is the Go-playing program built by DeepMind, the AI outfit Google bought a couple years back. Lee Sedol is one of the best Go players alive, 9-dan professional, the guy people figured would mop the floor with a machine the way Go players have always assumed machines can't really do Go. Chess got "solved" by computers back with Deep Blue in 97, sure, but Go has this reputation of being too big, too intuitive, too much a game of feel rather than brute-force calculation for a computer to be competitive at the top level. There are more possible board positions in Go than there are atoms in the observable universe. I've seen that stat thrown around so many times this week it's basically become a meme at this point, but it's true and it's a genuinely wild thing to sit with.

So this is game one of five, best of five, million dollar prize pool (which, if AlphaGo wins the match, DeepMind said is going to charity and STEM programs, since obviously the software doesn't need spending money). And AlphaGo won. Lee Sedol resigned after 186 moves, something like three and a half hours in. The English broadcast had Michael Redmond, an actual professional player, doing commentary, and you could hear him getting visibly thrown off partway through, going quiet in that specific way where you can tell someone is recalculating everything they assumed going in.

I don't play Go. I want to be upfront about that because I think a lot of the instant hot-takes today are coming from people who also don't play Go, myself included, confidently explaining why this is either the biggest deal in AI history or totally overblown. My actual Go board is a cheap wooden one I bought at a going-out-of-business bookstore sale in like 2013 and it has lived under my bed since approximately two weeks after I bought it. I know the rules. I am bad at it. I have never once beaten anyone at it who also knew the rules. So take my Go opinions with that grain of salt.

What got me watching live wasn't really the Go, honestly, it was the uncertainty. Nobody I follow online seemed to actually know what was going to happen, which almost never happens with these AI demo events anymore. Usually you can smell a stacked deck from a mile off. This one felt like an actual unscripted contest, and Lee Sedol himself said before the match he expected to win comfortably, maybe even a clean sweep. That's not the kind of thing you say if you think it's rigged.

The part that stuck with me watching the stream wasn't a specific move (I wouldn't know a brilliant Go move from a mediocre one if it bit me) but the reaction shots. There's a camera on Lee Sedol during his long think times and at one point he's just sitting there with his hand over his mouth for what felt like a full minute of dead air on the stream. That's the moment that made this feel real rather than a press release with a scoreboard attached.

There are four more games this week, Thursday, Saturday, Sunday, next Tuesday I think, and I'm fully aware I'm going to lose more sleep over this than is reasonable for someone who cannot tell you what a good opening move looks like. I keep thinking about how Kasparov described losing to Deep Blue, how there was this specific dread of feeling like something on the other side of the board was thinking, actually thinking, not just calculating. I wonder if Lee Sedol felt any of that today or if it's too early to say and he's just annoyed he blew an endgame.

Either way I've dug my Go board out from under the bed. Doesn't mean I'll be any less bad at it by Thursday.