Why is it in the news?

• Google DeepMind introduces SIMA (Scalable Instructable Multiworld Agent), an AI gaming agent capable of following natural language instructions to perform tasks across various video game environments.

More about the news

• SIMA signifies a future where AI agents could play significant roles in gaming, as well as in real-world task collaborations with humans.
• SIMA is described as a generalist AI Agent, capable of understanding and executing instructions in diverse virtual environments, from exploring dungeons to constructing castles.
• It can accomplish tasks or overcome challenges assigned to it within the gaming environment, demonstrating adaptability and learning capabilities.
• SIMA processes human language commands and learns and adapts through interactions with users, improving its understanding and performance over time.
• Google DeepMind collaborated with game developers to train SIMA on a variety of video games, marking the first instance of an agent understanding and executing tasks across different gaming worlds.
• SIMA was tested on nine different video games, including Teardown and No Man’s Sky, as well as four research environments, including the Construction Lab built with Unity.
• Through exposure to various gaming worlds, SIMA acquired skills such as navigation, menu use, resource mining, flying spaceships, and object manipulation in the physical world.
• Initially, SIMA’s training involved recording pairs of human players, where one player instructed the other. It also learned from observing free gameplay and recording instructions retrospectively.