Google's AI Agent Learns from Video Games for Future Collaboration

Google's latest AI agent, SIMA, is trained to carry out tasks in video games and play games it has never seen before. The research aims to understand how AI systems can become more cooperative and helpful. SIMA is trained to transfer skills between games and behaves in unplayed games. Researchers used imitation learning techniques with human player inputs to teach SIMA to play different games, and find that SIMA outperforms game specialists. However, the AI agent is not perfect, and the goal is to improve its performance, including making it able to follow more detailed instructions and to develop AI systems that can act in diverse environments and interact with users.