Goat Simulator 3: Exploring AI through Learning to Do Random Things in Games, Videos, and Animated Ungulates

To do this, the team first built a new environment in the Unity engine where the agents needed to create sculptures to test their understanding of object manipulation. Then, Google recorded pairs of human players — one controlling the game and the other giving instructions on what to do next — to capture language instructions. Afterward, players played independently to show what led to their actions in the game. All of this was fed to the SIMA agents to learn to predict what would happen next on the screen.

The team chose games with open play that helped SimA learn general gaming skills. When you play or watch a video game, the point is that you do random things, and Harley said it was a chance forSIMA to learn.

A video game called Goat Simulator 3 is a game in which players can take their domesticated ungulates on wild adventures that include jetpacks.

That might seem odd for a place with a chance of the next big leap in artificial intelligence.

Gaining an edge in building on the recent generativeAI boom depends on broadening out the kind of data that is useful for learning from.