Together with colleagues at the University of Essex, UK, and DeepMind, Togelius runs the General Video Game AI competition – now in its third year – testing AIs across a variety of different arcade games. The best StarCraft bot is worthless at Super Mario Bros.” The AI simply gets good at a specific task, and its skills aren’t transferable.Ī general AI would be able to play many different games, even ones it has never seen before. “Deep Blue beat Kasparov in chess but can’t play checkers. “You can’t just take AlphaGo and apply it to another problem, not even another game,” he says. The problem is that you can’t achieve it by training an AI on one game. Togelius is interested in what’s known as general artificial intelligence – the kind of smarts that can be applied to many different problems. Computers are quickly closing the gap, though. “The best StarCraft-playing programs can barely beat a beginner,” says Togelius. But as computers hone their gaming skills, we will need new ways to make them fun opponents. Games are a great way to test AIs because they offer a range of challenges, says Julian Togelius at New York University.

And there are regular showdowns between bots and professional players, such as the annual human versus AI StarCraft competition. There are bot-on-bot events involving first-person shooters like Unreal Tournament. As well as Go, the DeepMind team have pitted their algorithms against dozens of 80s arcade games and 3D mazes based on 90s shooters. “The program is not like this.”Ĭomputers play lots of different games these days. “The problem is humans sometimes make very big mistakes, because we are human,” he said. That’s how European Go champion Fan Hui described the experience of losing five straight games to AlphaGo, an artificial intelligence built by Google’s DeepMind team.

Spyparty controls