Google DeepMind has unveiled an advanced AI agent capable of playing a variety of video games, such as No Man’s Sky, Valheim, and Goat Simulator 3, aiming to serve as an interactive gaming companion.
This innovative tool, known as SIMA 2, is an evolution of its predecessor, SIMA (Scalable Instructable Multiworld Agent), which was introduced in March 2024. Now integrated with Google’s Gemini AI, SIMA 2 transcends simple instruction-following. It can comprehend users’ overarching goals, perform intricate reasoning, and adeptly execute actions to achieve objectives in games, even those it’s encountering for the first time, as detailed in a DeepMind blog post. Currently, it is available to select academics and developers in a limited research preview.
Though SIMA 2 excels in gaming, the team at DeepMind clarified to The Verge that their aim extends beyond merely developing a consumer-oriented gaming assistant. Jane Wang, a senior research scientist at DeepMind, described it as “an excellent training platform” for potentially applying these skills to real-world scenarios in the future.
All of this ties back to the escalating race for Artificial General Intelligence (AGI) among tech giants like Google, Meta, OpenAI, Anthropic, and others. DeepMind’s blog post highlights this development as a critical move towards AGI, carrying significant implications for the future of robotics and AI embodiment.
Joe Marino, another research scientist at DeepMind, emphasized that SIMA 2’s capability to operate in unfamiliar virtual environments represents a “fundamental” advancement toward AGI. This progress might eventually contribute to the creation of versatile, general-purpose robots.
