Google DeepMind has taken a step forward in the evolution of artificial intelligence with the launch of CHASM, an AI agent designed to learn gaming skills and play more like a human than a conventional AI. SIMA, which means Scalable, Instructable, Multiworld Agentat the moment is in the research phasebut promises to revolutionize the way we interact with video games.
Unlike traditional AIs that seek to win the game at all costs, SIMA focuses on running the game in a more human and adaptive way, following instructions rather than simply following a set of predefined algorithms. This approach makes it ideal for integrating as another player into a group, rather than competing against it.
SIMA was tested in some popular games
To train SIMA, Google worked in collaboration with eight developers from leading games, including Hello Games, Embracer, Tuxedo Labs, and Coffee Stain, among others. With games like No Man's Sky, Teardown, Valheim and Goat Simulator 3, SIMA learned basic gaming concepts, from manipulating objects to navigating 3D environments.
Google's research team focused on games with open gameplay, such as Goat Simulator, to teach SIMA to adapt to unpredictable situations and perform spontaneous actions. Using a Unity environment, recordings of human players, and machine learning techniques, SIMA has developed around 600 basic skills so far.
Although SIMA cannot yet perform such complex tasks as finding resources and building camps, its potential is promising. With more advanced AI models, it is expected that SIMA can perform more sophisticated actions and become a valuable member in any gaming group.
Although it is still early to predict SIMA's exact impact on the video game industry, its ability to play in a more human and adaptive way could have significant implications on game design and experience. Over time, SIMA could transform the way we interact with games and open new possibilities in artificial intelligence for games.