kasku.net/blog/2022/11/25/

Game AI

Chess

I was linked to this 1991 documentary, "The Chip vs. The Chess Master" [vimeo]. It's pretty good, but not must-watch or anything. It's about Gary Kasparov vs Deep Thought.

It went through some of the usual bits, like commentary on the way people seperate themselves from machines by "unique" characteristics. At some point in history, someone probably said something like "computers will never play chess well because they aren't intelligent". Obviously no one says this now.

This moving target for what counts as intelligence is interesting. Computers can solve equations, read text from images, and they can beat chess grandmasters. All of these were "intelligent behaviours" that seperated us from computers, at one point.

From Ken Thompson at 13:45:

You'd think that the human mind is something special, it's more than just a machine, and uh, when you're beat by a machine, at something that's considered to be the epitome of intellectual activity, something happens, either the notion of intelligence or chess or something has to change when this happens.

and it turns out people no longer say things like "we will have true artifical intelligence when computers can play chess", so I think it's safe to say that our definition of intelligent behaviour has changed.

It's also interesting how long it took for chess engines to get really good. I remember seeing a grandmaster talking about trusting a chess engine's evaluation of a certain line in an opening in 2004, but after playing it, realising that the evaluation was incorrect. This basically couldn't happen now. Chess engines are extremely refined and accurate, but obviously don't base their evaluations on how well a human player could continue from that position.

Go

I don't know much about the game, but this documentary about alphago is also entertaining and worth watching. I don't really know how strong go engines are now, but I'm going to tentatively assume they don't have have the crystalline refinedness of modern chess engines, simply because the world champion was only beaten so recently.

Othello

I don't know much about this game either, but Moore's Law, Progress, and Displacement by Arnold Kling is about how he lost to a new computer running the same program.

Starcraft 2

This is another game where the world champion was defeated by a computer, though I'm not sure if he was world champion at the time. There's a playlist of Serral's games against alphastar, and I think it's interesting to watch, because it really doesn't seem like the program exhibits mastery of the game, in comparison to e.g. lastest stockfish. I'd go as far as to say it seems kind of bad, lots of the time.

There's a difference between someone who can win (e.g. against the world champion), and someone who's so good that it's impossible for them to lose. This second category is someone you'd call a "machine". For several periods of time, Serral has exhibited an apparent inability to lose, but this particular version of the program doesn't feel like that at all. It feels relatively imprecise.

I'm not making any judgement on whether this is a meaningful achievement or not, just I don't feel much excitement about it. A game between stockfish 15 and another chess engine sounds like it would be a lot more interesting than a game between alphastar and itself, even if that's just because of style or an aesthetic preference. I like watching Serral more than I like watching other top zergs, for example.

Brute Force

This is a phrase that came up a little in the first documentary. They talked about how the computer analysed whatever large number of positions, and how human players analysed only a few of the ones they judged to be more important.

I think the same principle can be applied to some other fields too. On a higher level, it's possible to intuit a reasonable strategy for improving at a particular activity and spend your time that way. I did this for a while when playing osu, and was able to set 2 scores that were better than scores of people who had played in osu world cup, despite being ranked around 50,000 at my peak, and playing relatively little.

But when you take a brute force approach, you unlock new capabilities. When I was playing osu, I made sure the time I did spend playing didn't go to waste. If I'd played more, I could have experimented with different things, and played a wider variety of maps, and found a way to improve skills other than the one I focused on (accuracy/reading). Instead, I stuck to a relatively narrow path, and had relatively narrow success.

Another example is from LS (whose Man vs Machine video seem somehow relevant to this article). I remember him talking about someone he had coached in league of legends, but who hadn't improved. LS said that the client wasn't seeing results because he hadn't been playing enough games. If you're playing 3 games a day, about 20 a week, and it takes 100 games to see improvement from a particular approach, it would take over a month for that to happen. In comparison, if you play 10 games a day (7-8 hours) then it takes just over a week to reach 100.

My strategy would have been to think my way out of this problem, but that doesn't really work. You can get a lot better at osu by thinking about it, but not league of legends. Don't ask me why, I don't understand this process.

Doing Lots of Stuff

I'll substitute "brute force" with "doing lots of stuff" to be a bit more neutral about it, as well as a bit more general. Brute force sounds inefficient, but being more fast is another way to do more stuff.

I don't consider myself a particularly fast typist, but that's because I know people who can type over 200 WPM. I took a single typeracer test while writing this, and got 124, then 134 on the "prove you're not cheating" test. Statistically, this is relatively fast. If I couldn't type as fast, then it would take me longer to write stuff. I probably wouldn't bother writing about random stuff like this, because my lowered capacity to write would make it take longer, to the point where it probably wouldn't be worth it.

I have more to say on this topic, but I can't put it into words yet.