Genocide Automation, which was featured at Sense of Wonder Night (an indie game showcase at the Tokyo Game Show), is less a game than an exploration of flocking behavior and cellular automata.
Two swarms appear at opposite corners of the play space, and attempt to destroy each other (it's not entirely clear what the combat algorithm is). Individual motes in the swarm obey some set of flocking algorithms that seem to be represented by the data displayed in the right column, though it's hard to know what that's showing, since it's in Japanese. Both sides "learn" from failure, modifying algorithms next time to try something else -- I'm not clear whether evolutionary learning algorithms are used, or whether behavior modification is simply random.
