3 ms·
Ask HN: Go, chess, but can an AI win at backgammon?
I just played my first game and got in a discussion about it. My opponent said that computers can’t do it yet. I was surprised since it’s a game of perfect information.
I used a couple of search engines to skim some articles but I get the feeling that some people in the HN crowd can give a better perspective on it.
Why can’t an AI win at backgammon? Or can they and did I skim a bit too much?
- slowdog 7y agoDepends on your definition of win. Do you mean perfect game? Be the best? Beat the average player? I think there’s a bunch of vagueness but let’s take “win” to be generally better than humans, I’m not an AI expert but I suspect it’s got more to do with the popularity of the game There’s no reason why backgammon can’t be played by machines, and definitely no reason why they would well. There’s just not as much as a marketing pop compared to chess, go or Starcraft all well regarded complex games. In a lot of ways, googles investment in AI is a marketing opportunity, and in an unpopular game there’s not as much incentive
- yorwba 7y agohttps://en.wikipedia.org/wiki/TD-Gammon https://en.wikipedia.org/wiki/TD-Gammon (1992)
- mettamage 7y ago1992, but not total domination. Couldn’t OpenAI solve this, or a similar research group? I am surprised it is still out in the open. Even poker seems to be more “solved” than backgammon.
- yorwba 7y agoIf it was possible to achieve a level close to top human players with computing resources available in 1992, hardware of equivalent cost can probably handily beat any human in 2020. So no big lab is going to spend any time on hammering home that point, because they're looking for interesting research problems, like games where no program plays at human level yet. If you want to see how well computers play using consumer hardware, maybe try https://www.gnu.org/software/gnubg/ https://www.gnu.org/software/gnubg/ I have no idea whether a rating of 2000 is superhuman or not.