One of the ideas I’m interested in pursuing for my final project is bot vs. bot entertainment. I built an IRC bot to scrape Twitch chat (for the data scraping assignment) and am able to write back to the chat as well, but my friend had an idea that I thought was brilliant. In his words, “The real coup de grace is if they were chatting about bots playing each other.” Turns out this is definitely doable, and I’d like to have a Twitch stream that streams 5v5 bot DotA matches, with bots commenting asinine things as spectators.
This isn’t an art piece, per se, but I thought it would be interesting to look at it from an entertainment point of view. The 2048 AI Solver is exactly what it sounds like: a playable version of 2048 that also includes an AI to either provide a hint or to run automatically until it gets to 2048.
I’ll admit it: I never beat 2048. I was always too lazy to consider my moves and tried to “feel out” my swipes in a half-random fashion. Watching the computer play with a decided strategy was enlightening and mesmerizing. I don’t think it’s repeatedly entertaining, but there’s a certain beauty to watching the computer solve a problem that I never was able to grasp.
For what it is, it’s completely perfect. For what it’s not (an art project), it is lacking. What I would enjoy seeing is a head to head competition of two separate AI’s, perhaps presenting the code or algorithm behind each one and asking the observer to guess which one will solve the puzzle faster.
@ChessBotWhite and @ChessBotBlack are two twitter bots by Aaron Marriner that play chess against each other. It looks like each bot makes a move roughly every other hour, posting the algebraic notation for the move and the resulting board to twitter.
I love the idea of bots playing against each other. Game AI is often seen as a thing to practice against so that we can avoid embarrassing ourselves when we play against other humans. In chess, however, AI has long since surpassed human capabilities; shouldn’t this imply that two bots playing against each other will be a high level game worth spectating?
In this case, no, we cannot infer that the game is at a high level as we don’t have a point of reference to human ability. That’s alright though, since what actually might be more interesting are the places where the bot goes off the rails and makes a mistake. Called out mistakes would give a sense of drama. I’d also like to see a running tally of which bot is up more games, or anything to make it feel a little less automatic and a little more human.
This project likely has its roots outside of the art world, going back to IBM’s Deep Blue, though chess AI has worked its way deeply into chess culture. A related work is @PlayLightsOut, another game playing twitter bot, albeit one that plays in response to human tweets rather than against another AI.