Google DeepMind has released SIMA, a well-trained AI agent designed to learn gaming skills and play more like a human player rather than a dominant AI. SIMA stands for Scalable, Instructable, Multi-world Agents and is currently in the research stage. SIMA will eventually learn to play any video game, even those without linear progression paths and open worlds. While it is not intended to replace existing game AI, you can think of it as another player that works seamlessly with your team. It combines natural language instructions with an understanding of the 3D world and image recognition. "SIMA is not trained to win games; it is trained to run games and act according to instructions," said Tim Harley, a researcher at Google DeepMind and co-lead of SIMA, during a press briefing. Google has collaborated with eight game developers, including Hello Games, Embracer, Tuxedo Labs, and Coffee Stain, to train and test SIMA. Researchers have inserted SIMA into games such as "No Man's Sky," "Teardown," "Valheim," and "Goat Simulator 3" to teach the AI agent the basics of playing these games. Google stated in a blog post that SIMA does not require custom APIs to play games or access source code. Harley explained that the team chose games that focus more on open gameplay rather than narrative to help SIMA learn general gaming skills. If you have played or watched the gameplay of "Goat Simulator," you would know that randomness and spontaneity are the key aspects of the game. Harley expressed the hope that SIMA would learn this kind of spontaneity. To achieve this, the team first created a new environment in the Unity engine where the agents had to create sculptures to test their understanding of object manipulation. Google then recorded pairs of human players, with one controlling the game and the other giving instructions for the next move, to capture language instructions. Players then independently played the game to demonstrate what prompted them to take action. All of this was inputted into the SIMA agent for learning to predict what would happen next on the screen. Currently, SIMA has around 600 basic skills, such as turning left, climbing ladders, and opening menus or using maps. Harley said that eventually, SIMA can be instructed to perform more complex functions in games. Tasks like "finding resources and building a camp" are still challenging to accomplish because AI agents cannot perform actions for humans. SIMA is not an AI-driven NPC like those from NVIDIA and Conway Corporation but rather another player in the game that influences the game outcome. Frederick Bessé, co-lead of the SIMA project, stated that it is still difficult to say what purposes AI agents like this could serve outside of research in the gaming field. However, similar to AI NPCs, SIMA may eventually learn to speak, but that is still a long way off. SIMA is still learning how to play games and adapt to games it has never played before. Google stated that as more advanced AI models develop, SIMA may eventually perform more complex tasks and become the best AI teammate to lead you to victory.