An artificial intelligence system may have beaten the world’s best Go player, but you shouldn’t expect that same system, no matter how smart it is, to be able to play other games. That’s a limitation of what some call “narrow AI”—an inability to expand outside of known domains. The folks behind OpenAI, a billion-dollar Silicon Valley artificial intelligence nonprofit started by tech superstars like Elon Musk and Y Combinator president Sam Altman, want to help change that dynamic in order to develop AI systems with broader capabilities. And games, because they present many complex tasks for players to complete, may well be the ideal platform for doing so.
— Mike Kuniavsky (@mikekuniavsky) December 5, 2016
That’s why they’ve launched Universe, a new AI training and proving ground that mines a thousand video games from some of the world’s biggest game companies—Electronic Arts, Valve, Microsoft, and others—in order to attempt to “develop a single AI agent that can flexibly apply its past experience on Universe environments to quickly master unfamiliar, difficult environments, which would be a major step towards general intelligence.”
In the early going, AI agents developed using the system will likely only be able to handle simple tasks like navigating Web buttons, interfaces, and sliders, but over time, OpenAI hopes, they will be able to tackle more complex actions.