News on Medial

Google Deepmind trains a video game-playing AI to be your co-op companion

TechCrunch · 1y ago

Medial

Google Deepmind researchers have developed an AI model called SIMA (scalable instructable multiworld agent) that can play multiple 3D games by learning from human gameplay videos and verbal instructions. The model associates visual representations of actions, objects, and interactions with certain instructions. Although AI agents trained on multiple games performed better on unfamiliar games, specific mechanics and terms unique to each game can still pose challenges. The researchers aim to create a more natural game-playing companion that can adapt and produce emergent behaviors. This approach differs from traditional simulator-based agent training, as SIMA uses imitation learning from human behavior instead of a reward signal.