- Google DeepMind introduced SIMA, an artificial intelligence agent trained to learn gaming skills. SIMA, which stands for Scalable, Teachable, Multi-World Agent, is currently only in the research phase.
- Google DeepMind trained its video game-playing AI agent on games like Valheim, No Man’s Sky and Goat Simulator.
- While it’s not intended to replace the current game AI, you can consider SIMA as another player that would fit well into your party.
Google DeepMind introduced SIMA, an artificial intelligence agent trained to learn gaming skills. SIMA plays more like a human rather than an overpowered AI. SIMA, which stands for Scalable, Teachable, Multi-World Agent, is currently only in the research phase.
SIMA will learn how to play any video game, even games that do not have a linear path to finish the game and open world games. While it’s not intended to replace the current game AI, you can consider SIMA as another player that would fit well into your party. It blends natural language education with understanding 3D worlds and image recognition.
“SIMA is not trained to win a game; he was trained to work it and do what he was told.” he states.
Google worked with eight game developers, including Hello Games, Embracer, Tuxedo Labs, Coffee Stain, to train and test SIMA. Researchers placed SIMA in games such as No Man’s Sky, Teardown, Valheim and Goat Simulator 3, teaching the AI agent the basics of playing the games. In a blog post, Google said that SIMA does not require a special API to play games or access their source code.
Harley said the team chose games that focused on open gameplay rather than narrative to help SIMA learn general gaming skills. If you’ve played or watched Goat Simulator, you know that doing random, spontaneous things is the point of the game, and Harley said that kind of spontaneity is what they hope SIMA learns.
SIMA currently has around 600 basic skills, such as turning left, climbing a ladder, and opening the menu to use a map. Eventually, SIMA could be instructed to perform more complex functions within a game, Harley said. Tasks like “finding resources and setting up camp” remain difficult because AI agents cannot perform actions for humans.
SIMA is designed to be another player in the game who affects the outcome, not an AI-powered NPC like those from Nvidia and Convai. Frederic Besse, co-leader of the SIMA project, said it was too early to tell what kinds of uses AI agents like these might bring to games outside the field of research.
Like AI NPCs, SIMA may eventually learn to speak, but it’s a long way off yet. SIMA is still learning to play games and adapt to games he hasn’t played before. Google states that with more advanced AI models, SIMA will eventually be able to handle more complex tasks and become the perfect AI playmate to lead you to victory.
Compiled by: Alp Eren Gümüş