Monday, 27 April 2015


I'm kind of interested in the way a game was created specifically to be difficult for computers but easy enough for a 4-year-old to play with standard chess pieces, called Arimaa. It's an interesting goal to push computer game-playing research further, and it seems to have worked. The game was invented in 2002 and then a computer won the yearly human-vs-computer tournament this year (2015). On the one hand, this might represent massive gains in processing power and algorithmic research since the game's invention, or it might represent an unforeseen weakness in the game design. Most likely, it's a bit of both.

My question is: what comes next? Does someone else design a new iteration of game that is really tricky for computers to play but comparatively easy for humans? Do we modify Arimaa to bump up the computer difficulty? When do we reach the point where "easy for a 4-year-old" also means "trivial for computers"? If it takes decades for computers to beat human grandmasters at chess, but only one decade for computers to beat human Arimaa masters, how much difficulty can we expect to squeeze out of a new iteration of game difficulty? Probably a year or two, I'd think, given the pattern.

Mokalus of Borg

PS - I'd love it if a computer designed a game itself to this goal.
PPS - Which, I guess, would make that game a kind of CAPTCHA.

No comments: