Google’s New AI Agent, SIMA, Demonstrates Ability to Play Video Games and Follow User Instructions across Multiple Virtual Worlds
The Google DeepMind team has developed an AI agent known as SIMA (Scalable Instructable Multiworld Agent) that can perform tasks in various video games. Unlike previous AI agents limited to specific games, SIMA has been trained across different virtual environments, making it a versatile and multi-world agent capable of carrying out tasks like a human player.
Partnering with eight gaming studios, Google DeepMind trained SIMA on nine 3D video games, including popular titles like No Man’s Sky and Valheim. The AI agent can seamlessly navigate various game worlds, mine resources, fly spaceships, and more, all based on user instructions.
Utilizing images and natural language instructions, SIMA can control gameplay using standard keyboard and mouse inputs. With nearly 600 acquired skills, it can accomplish tasks in virtual environments within seconds, ranging from basic movements to complex actions like object interaction and map navigation.
Google DeepMind’s innovative approach with SIMA shows promising results, outperforming specialized agents trained on individual games. The potential applications extend beyond gaming, hinting at significant advancements in robotics and real-world tasks where general AI agents can plan and execute actions efficiently.
The team’s technical report sheds light on SIMA’s capabilities and the underlying mechanisms that make it a groundbreaking development in the field of artificial intelligence. As the boundaries of AI continue to expand, SIMA represents a significant step towards creating more adaptable and versatile AI agents that can navigate multiple virtual worlds with ease.