SIMA, l’IA de Google qui Joue aux Jeux Vidéo comme un Humain

SIMA, short for Scalable Instructable Multiworld Agent, is a generalist AI agent designed for 3D virtual environments. It is developed by Google and is intended for researchers and developers who want to advance AI agent research. SIMA can understand a broad range of gaming worlds and follow natural-language instructions to carry out tasks within them, similar to how a human would.

SIMA works by using a model designed for precise image-language mapping and a video model that predicts what will happen next on-screen. It is trained on a variety of video games, including No Man’s Sky and Teardown, and can learn from different gaming worlds to capture how language ties in with game-play behavior. SIMA can perceive and understand a variety of environments, then take actions to achieve an instructed goal, all without needing access to a game's source code or bespoke APIs.

Researchers and developers who want to create more helpful AI agents for any environment can get the most value from SIMA. By using SIMA to learn from video games, they can unlock more helpful AI agents that can follow instructions in a variety of settings, not just limited to gaming. This can lead to significant advancements in AI research, particularly in areas such as high-level strategic planning and multiple sub-tasks to complete.