Computers that Play Games

“Is there a game that a computer could not eventually beat a person at?”

“It seems unlikely. If the game is at all winnable, a computer could play out all future sub-games and always make an optimal move. Or if that’s unfeasible, it could just implement whatever rules or heuristics human players are using. So at worst, it would be a wash.”

“What about mad-libs? The game where participants are asked for particular parts of speech that an arbiter then places into a pre-defined story with humorous consequences.”

“But that isn’t a game! There is no winning! And besides, a computer could probably play that just as well as a person. Train a machine on a corpus of text with a process like latent semantic analysis. Then, figure out which heuristics for choosing words leads to the funniest results. I would think that words that have very low frequency, or low connectivity to common words would work.”

“That’s true. And I guess even if that doesn’t work, the computer could just give scatological responses.”

“I suppose so. I guess the best computer player of mad-libs would likely have a whole body of experiences, like memories— those which had unexpected components might be where it might derive humor. I guess we can’t really criticize the sense of humor of a machine until we understand the sense of humor of people. Computers, I guess, could just be considered as though they were from a different culture. Have you ever told a completely culturally inappropriate joke?”