Angelina: An AI that makes computer games from scratch

A PhD student at Imperial College London has created a computer AI, called Angelina, that can create computer games autonomously.

In a process called “cooperative co-evolution,” which is an emerging field of evolutionary computation, each aspect of the game is broken down into “species” — level layout, enemy behavior, and power-ups are all species. Angelina takes a random selection of species and throws them together to create a game.

Now, the evolutionary part: Angelina simulates a human playing through this random game 400 times. Levels that are too hard to complete are softened. Maps that are completely impossible are thrown out entirely. Power-ups are added, removed, and tweaked. Enemy behavior is changed. After each of these changes, another 400 iterations are played. Each time, Angelina breeds together fun designs, while bad designs are left out of the gene pool. The end result, which you can experience for yourself on the Games By Angelina website, is surprisingly good. Some of the games require you to visit multiple locations to pick up the right items, in order, to finish the level; not bad, for an evolutionary algorithm.

For now, Angelina’s creator Michael Cook still has to create the art and audio assets, but the rest is completely autonomous. Mind you, it wouldn’t be hard for Angelina to pull down its own textures from Google Images. “In theory there is nothing to stop an artist sitting down with Angelina, creating a game every 12 hours and feeding that into the Apple App Store,” says Cook, speaking to New Scientist.

Looking to the future, though, Cook, perhaps naively, hopes that Angelina “won’t steal anyone’s job.” While I don’t think Angelina will usurp lead designers like Shigeru Miyamoto or Peter Molyneux, I think, at least when it comes to cheap and cheerful Flash and smartphone games, Angelina could really do rather well. After all, if robots have taught us anything it’s that they don’t have to be as good as human workers; they just have to be good enough.

As far as the big picture is concerned, Angelina is an example of the strengths of evolutionary computation, the sphere of computer science that will likely give birth to human-level machine intelligence. Evolutionary computation is primarily concerned with the optimization of processes through survival of the fittest — much in the same way that a human learns through mistakes, and how strong genetic traits are preserved through breeding. Mimicking game designers is one thing, but it’s not too hard to imagine the other human roles that an Angelina-like program could perform.

Angelina: An AI that makes computer games from scratch