December 7, 2017

Is it just me or are conservatives a lot more likely to use a term like "wonderful" non-ironically than liberals? And is Trump leading that or just part of a trend or larger reason for that?
AI AlphaGo Zero started from scratch to become best at Chess, Go and Japanese Chess within hours - This is pretty incredible stuff, and damn near my idea that "I'll be impressed when the same program that wins at Go wins at Chess, and for the same reasons."

I remember hearing about the core idea (setting a game-playing AI against a copy of itself to improve) used in Arthur Samuel's checker playing program back in the late-50s.

All the games AlphaGo Zero plays are "perfect information" games. I wonder how it would do with games of ambiguity and bluff and randomness, like Poker (or Stratego, even.) I suspect when you have a computer play a version of itself, you're vulnerable to the "hill climbing" problem (i.e. if you always head towards the highest ground NEAR you, you might end up stranded on a local high peak, but not the highest in the land) - that you get a certain type of genius at playing another certain type of genius, but vulnerable when playing a more wildcard player, and that vulnerability is increased if you don't know the full state of the game.

Of course, my favorite emergent chess program behavior remains the stories around Atari 2600 Video Chess; the screen would blank as the computer was "thinking", and sometimes when the board returned you'd find some pieces weren't quite where they were before...

UPDATE: better summary The future is here – AlphaZero learns chess