SIMA currently has around 600 basic skills, such as turning left and opening a menu, and is expected to learn more complex tasks in the future. However, it's not intended to replace existing game AI or perform actions for humans, but to act as another player that impacts the game's outcome. While it's still early to predict the potential applications of AI agents like SIMA in gaming, Google believes that with more advanced AI models, SIMA could eventually perform more complex tasks and become the perfect AI party member.
Key takeaways:
- Google DeepMind has introduced an AI agent called SIMA (Scalable, Instructable, Multiworld Agent) that is being trained to play video games more like a human, rather than an overpowered AI.
- SIMA is designed to learn how to play any video game, including non-linear and open-world games, and is not intended to replace existing game AI but to act as another player.
- The AI agent combines natural language instruction with understanding 3D worlds and image recognition, and is trained to follow instructions rather than to win games.
- Google has collaborated with eight game developers to train and test SIMA, and the AI currently has around 600 basic skills, with the potential to learn more complex tasks in the future.