Google DeepMind’s New AI Agent Learns, Adapts and Plays Games Like a Human

Google DeepMind has launched SIMA 2, an advanced reasoning AI agent designed for 3D virtual environments, which represents a significant step toward artificial general intelligence (AGI). SIMA 2 enhances its predecessor by allowing for self-directed learning, where it learns new skills through play and adapts to unfamiliar game worlds generated by another DeepMind project. This new AI can interpret high-level goals, explain its actions, and collaborate within games, marking an improvement in its reasoning capabilities. SIMA 2 achieved a task completion rate of 65%, a substantial increase from the previous 31%. The system can effectively tackle complex tasks and transfer knowledge across different games. However, challenges remain, such as handling very long tasks and limited memory. Overall, DeepMind considers SIMA 2 a promising platform for future robotics applications and a significant move towards AGI.

Source 🔗