DeepMind AI plays Goat Simulator

Answer
Google DeepMind has developed a new AI agent named SIMA (Scalable Instructable Multiworld Agent) that has learned to play "Goat Simulator 3" among other video games. This advancement is notable because SIMA can adapt to games it has never played before by leveraging what it learned from playing other games. The AI agent was trained using a combination of keyboard and mouse data from human players across ten different games with 3D environments, including "No Man's Sky," "Teardown," "Hydroneer," and "Satisfactory." This data was then labeled and fed into a language model, enabling SIMA to carry out actions in response to typed commands from human players12345. SIMA represents a significant step towards more generalized AI, capable of transferring skills across multiple environments. This is a departure from earlier game-playing AI systems, which were typically designed to master a single game or follow specific goals or commands. The ability of SIMA to play a variety of games, including "Valheim" and "No Man's Sky," and to follow around 600 basic instructions, such as "Turn left" or "Climb the ladder," showcases its versatility and potential for broader applications beyond gaming2. The development of SIMA is part of Google DeepMind's ongoing efforts to explore the capabilities of AI in gaming environments, which are considered good proxies for real-world tasks. This research could lead to AI agents that are not only capable of playing games alongside human players but also performing useful tasks in other domains. However, despite its impressive capabilities, SIMA is still in the research phase and is not yet at human-level performance in games like "No Man's Sky," where it could only complete 60% of the tasks that humans could do23. The project involved collaboration with several game studios to collect the necessary data for training SIMA, and it did not require custom APIs or access to game source codes. This approach, focusing on open play rather than narrative-driven games, was chosen to help SIMA learn general gaming skills. The ultimate goal for SIMA and similar AI agents is to be able to quickly pick up games they have never seen before, much like a human player would34. This development hints at the potential for AI systems to do more than just chat and generate images or videos, by taking control of computers and performing complex commands. It also raises questions about the future role of AI in gaming and other applications, as researchers continue to work on improving the performance and capabilities of agents like SIMA15.
Related
what other games has deepmind's ai agent been trained on
how does deepmind's ai agent learn to play games
what are the potential applications of deepmind's ai agent technology
i.ibb.co
i.ibb.co
An AI that can play Goat Simulator is a step towards more useful ...
Google develops an AI that will actually play games with you
Google DeepMind trains a video game-playing AI to be your co-op ...
Google DeepMind's SIMA Is an AI Agent That Can Play 3D Video Games ...
Alex Wilkins on X: "The latest project from Google DeepMind is an ...
View More