How AI is en route to create commercially viable video games without human programmers

By Brian Santo, contributing writer

From the moment that game developers began using artificial intelligence (AI) to improve the play of their games, it was just a matter of time before AIs would be able to actually create games. We might be on the verge of seeing that happening.

Game developers employ AI to create characters with specific, individualized behaviors, often to control fighting styles and tendencies but also other behaviors. It’s such common practice that there are plenty of manuals available that teach AI techniques to game programmers. They include Programming Game AI by Example and Artificial Intelligence for Games . Behavioral Mathematics for Game AI dates back to 2009.

AI and machine learning are now familiar to us from instances where chess- and go-playing computers have not only beaten human grandmasters but have occasionally done so with moves that their human opponents have described as unexpected, intuitive, and even creative.

Game developers keep refining AI and the use of AI to improve games. Unity , which provides a popular game engine, recently released machine-learning tools that game developers will be able to use to create agents/characters that can not only interact with the game player but also with each other, potentially becoming more interactive, more skilled, and perhaps even more creative as they do so.

But can an AI be used to actually create a game? At least one already has.

Michael Cook, a professor at the University of Falmouth, has been working for several years on a project to develop an AI that can create games. Cook calls the AI Angelina, which stands for a novel game-evolving lab rat I’ve named Angelina. We like this guy already.

The games that Angelina cranks out may not be very sophisticated, but baby steps . One is a demo (which you can play)  of a Super-Mario-like game in which, serendipitously as we write this, the hero is Santa Claus.

While Angelina has created games, Cook is equally as interested in how his AI can be used as a supplemental tool for human developers.

That line of inquiry is where the action seems to be. 

For instance, the Georgia Institute of Technology recently developed an AI that can learn how to build a game engine in minutes with no other input other than watching the game. A game engine, as GIT describes it, is the basic software of a game that governs everything from character movement to rendering graphics.

Mark Riedl, associate professor of Interactive Computing and co-investigator on the project said, “To our knowledge, this represents the first AI technique to learn a game engine and simulate a game world with gameplay footage.”

The technique demonstrably works for games in which the gameplay is entirely on-screen. The researchers were looking for games to train their AI on where the AI can make best use of predictive analysis. Examples of the games (and types of games) that the researchers used are Super Mario Bros. and Sonic the Hedgehog.

It is less likely — at least for now — that the AI would be able to interpolate the rules for a game engine for a game in which significant action takes place off-screen — an RPG (role-playing game), for example — let alone actually create a game engine from those rules.

Still, the development is an exciting one. It suggests that a graphic artist — not necessarily even a programmer — could illustrate sequences of a game and then show it to an AI, which would automatically generate a game engine.

At this point, AI is still a tool that human programmers would use to develop games.

But it seems as if it’s just a matter of time before an AI creates one that proves commercially popular.