A new study from MIT and Google DeepMind demonstrates a significant advancement in AI’s ability to understand and reason about the physical world. Researchers developed a system that can infer the 3D structure of objects from a single 2D image and predict how they will behave when interacted with, such as how a stack of blocks might topple. This capability, known as “visual common sense,” is a fundamental human skill that has been challenging for AI to master. The model was trained on massive datasets of simulated scenes, learning the underlying physics and material properties without explicit programming. This research represents a crucial step toward AI that can interact more safely and effectively in real-world environments, with potential applications in robotics, augmented reality, and autonomous systems. Read the full article for details on the methodology and implications: https://technologyreview.com/2024/07/15/1094750/ai-learns-physical-intuition-like-a-human/