Her name is Angelina: she runs on a heavy-duty Mac server and she’s building some addictive computer games for you.
Angelina (a tail-recursive acronym for “A Novel Game-Evolving Labrat I’ve Named ANGELINA”) is a project in evolutionary computing by Michael Cook, a PhD candidate at Imperial College in the UK. Angelina generates computer games from scratch. It arrives at a final product by breaking down the important elements of a game into sub-tasks called “species,” which together form a whole game. While auto-generating portions of videogames is nothing new to the medium, Angelina expands on the concept to almost fully-automate game development.
Cook says his “labrat” has created many games (sometimes at the rate of one game every ten minutes), and you can play some “Games by Angelina” here. While not all of Angelina’s spawn have been worthwhile or playable, the few that you can play on Cook’s site are pretty clever (especially for a machine).
According to a 2012 paper by Cook (PDF), the Angelina system draws on three species that evolve independently to create games. The first species, maps, determines “passable and impassable areas.” Layouts specifies the different entities in the world and the game player, and then rulesets defines the way the player’s obstacles will move. Angelina puts those categories together, then simulates a human playing the game 400 times to find glitches and problem areas that don’t work to cast aside. During these trials, Angelina also finds retaining levels that are considered “fun”: things that pose a challenge but then get easier. The most successful species are the ones that cooperate best with other species, so when those get grafted together you have a game by Angelina.
