On March 13, Google DeepMind announced the latest AI agent “SIMA” (Scalable Instructable Multiworld Agent) which can actively play games with you while following your commands. SIMA has been trained with a range of gaming skills to play more like a human than some typical AI. It can easily follow natural language instructions and perform tasks you assign across different games.
This is the first research of its kind, as Google DeepMind claims.” This research marks the first time an agent has demonstrated it can understand a broad range of gaming worlds, and follow natural-language instructions to carry out tasks within them, as a human might”
Google collaborated with 8 game developers who plugged SIMA into games like No Man’s Sky, Teardown, Valheim, and Goat Simulator 3 to train this AI agent and then test its capability. Google DeepMind mentioned that SIMA is not like other AI models like ChatGPT and Gemini. Although trained on large datasets, these models still require human assistance. While SIMA is trained to operate on its own without any particular human assistance.
See Related: Google Announces DeepMind; Accelerating Its Attempt At Leading The AI Race(Opens in a new browser tab)
SIMA Gaming Skills
“The current version of SIMA is evaluated across 600 basic skills, spanning navigation (e.g. “turn left”), object interaction (“climb the ladder”), and menu use (“open the map”). We’ve trained SIMA to perform simple tasks that can be completed within about 10 seconds” DeepMind mentioned in their blog.
Google has evaluated SIMA’s ability to perform almost 1500 in-game tasks. SIMA consists of a learning system with pre-trained vision models and a memory that supports keyboard and mouse outputs.
SIMA is confidently progressing towards mastering game playing and adapting to new ones, although the prospect of it eventually learning to talk, like AI NPCs, remains a possibility.
