Google recently announced the development of SIMA, a novel AI agent poised to become an in-game companion. Currently in the developmental phase, SIMA is honing its foundational skills, marking a significant stride in Google’s ongoing efforts to advance AI-powered products. Unlike conventional platforms geared towards outperforming humans in gaming, Google’s latest innovation, SIMA (Scalable, Instructable, Multiworld Agent), is designed to emulate human-like gameplay.
Reports indicate that SIMA is undergoing training to navigate various types of video games, from open-world adventures to those lacking a linear progression. Its objective? To seamlessly integrate into gaming scenarios as a cooperative participant rather than overtaking the gameplay entirely. Leveraging cutting-edge technologies such as image recognition, 3D environment comprehension, and natural language processing, SIMA is poised to revolutionize the gaming landscape.
Tim Harley, co-lead for SIMA and a researcher at Google DeepMind, emphasized during a press briefing that SIMA’s primary focus lies in executing tasks rather than winning games. Google collaborated with numerous game developers, including Hello Games, Embracer, Tuxedo Labs, and Coffee Stain, to refine and validate the concept. Through rigorous testing involving titles like Goat Simulator 3, Valheim, Teardown, and No Man’s Sky, researchers aimed to imbue SIMA with a comprehensive understanding of gaming mechanics.
In a blog post, Google highlighted SIMA’s versatility, noting its ability to parse source codes and operate games without requiring a specialized API. The selection of less narrative-driven games was intentional, enabling SIMA to develop broad gaming proficiencies centered around open-ended gameplay. To date, SIMA has acquired over 600 fundamental abilities, encompassing tasks such as navigation, item interaction, and menu navigation.
Harley acknowledged that while SIMA continues to evolve, certain challenges persist, particularly in tasks requiring human-like intuition, such as resource gathering and base building. However, as SIMA undergoes further training, it holds the potential to tackle increasingly complex gaming scenarios. Moreover, there’s speculation that with time, SIMA may even acquire the ability to communicate verbally, further blurring the lines between AI and human interaction in gaming environments.