Google DeepMind unveiled SIMA, an AI agent designed to learn gaming skills so it plays more like a human rather than an overpowered AI. SIMA stands for Scalable, Instructable, Multiworld Agent. It is intended to eventually learn how to play any video game, including open-world games. SIMA is trained to follow instructions and run games, not to necessarily win them. Researchers want it to mesh well with human players. Google worked with several game developers to train and test SIMA in games like No Man's Sky, Teardown, Valheim and Goat Simulator 3. SIMA does not need custom APIs or source codes to play games. It can learn general gaming skills from open-ended games. Researchers first trained SIMA to understand object manipulation using a Unity environment. SIMA currently has around 600 basic gaming skills but will hopefully learn more complex tasks over time. SIMA is meant to be another player controlled by language instructions, not an autonomous NPC. Its abilities may eventually include talking, but it still has more to learn about playing new games. With advanced AI, SIMA could become the "perfect party member" to help humans achieve victory.
Google’s new AI will play video games with you — but not to win
Source:
www.theverge.com