Google DeepMind has unveiled SIMA, an artificial intelligence that learns gaming skills to play like a human, The Verge reports.

SIMA, which stands for Scalable, Instructable, Multiworld Agent, is currently only at the research stage.

Over time, SIMA will learn to play any video game, even those that don’t have a linear path to the end of the game and open-world games. Although it is not intended to replace existing game AIs, think of it more as another player.

“SIMA isn’t trained to win a game; it’s trained to run it and do what it’s told,” said Google DeepMind researcher and SIMA co-lead Tim Harley during a briefing with reporters.

Google collaborated with eight game developers, including Hello Games, Embracer, Tuxedo Labs, Coffee Stain, and others, to train and test SIMA.

The researchers connected SIMA to games such as No Man’s Sky, Teardown, Valheim, and Goat Simulator 3 to teach the AI agent the basics of playing them. In its blog post, Google said that SIMA does not require a special API to play games or access to the source code.

According to Harley, the team chose games that are more focused on free-play than narrative to help SIMA learn general gaming skills.

If you’ve played or watched Goat Simulator walkthroughs, you know that the game is all about doing random, spontaneous things, and Harley said they hope SIMA will learn that kind of spontaneity.

To do this, the team first created a new environment in the Unity engine where the AI had to create sculptures to test its understanding of object manipulation.

Google then recorded pairs of human players, one of whom played the game and the other gave instructions on what to do next to capture the speech instructions. After that, the players played the game on their own to show what their actions in the game led to. All of this was passed on to the SIMA agents so that they could learn to predict what would happen next on the screen.

Currently, SIMA has about 600 basic skills, such as turning left, climbing stairs, opening a menu to use a map. Over time, Harley says, SIMA can be trained to perform more complex functions in the game. Tasks like finding resources and building a camp are still challenging.

SIMA is not an AI NPC like NVIDIA and Convai, but another player in the game that influences the outcome. SIMA project co-leader Frederic Besse says that it is too early to say what applications such AI agents can find in games outside of the research sphere.

However, like AI NPCs, SIMA may eventually learn to speak, but it is still a long way off. SIMA is still learning to play games and adapt to those it hasn’t played before.