Google working on AI agent that will be your human-like companion in video games

Google DeepMind recently introduced SIMA, a new AI agent designed to bring a more human-like approach to playing video games. Unlike typical AI that may overpower or diverge from human gameplay, SIMA (Scalable, Instructable, Multiworld Agent) is crafted to complement human players as a cooperative teammate. Currently in the research phase, SIMA will be able to master any video game, from linear narratives to open-world adventures.

Human-Like Gameplay: SIMA is trained not to win but to participate and follow directions, aiming to seamlessly integrate with human players rather than replacing the game's AI.

Learning from Multiple Games: Through collaboration with eight game developers, including notable names like Hello Games and Coffee Stain, SIMA has been tested in various games, such as No Man's Sky and Goat Simulator 3, learning the basics without needing a custom API or source code access.

Future Potential: Starting with 600 basic skills, SIMA is evolving to undertake more complex tasks. Although it's still early days, the vision is for SIMA to become an AI companion in gaming, potentially even learning to communicate with players.

SIMA has been taught to interpret human instructions and predict in-game outcomes. This unique training approach, including the recording of human gameplay and instruction, enables SIMA to grasp a variety of gaming skills and adapt to different game styles. For instance, the spontaneity seen in games like Goat Simulator serves as a learning ground for SIMA to understand and replicate the unpredictability of human play.

No AI NPCs for now

Despite its capabilities, SIMA is not intended to replace traditional AI non-player characters (NPCs) but to act as a dynamic participant within the gaming environment. While it's still early to predict the full extent of SIMA's application in gaming, its development hints at a future where AI can offer strategic partnership in video games.