A new study published in Nature reveals that artificial intelligence systems are developing an unexpected ability to reason about physical objects and their interactions in three-dimensional space, a skill previously thought to be uniquely human. Researchers from MIT and Stanford trained a multimodal AI model on vast datasets of videos and 3D simulations, finding it could predict object trajectories and stability in novel scenarios with high accuracy. The model, dubbed PhysNet, demonstrates an emergent understanding of basic physics principles like gravity, momentum, and occlusion without explicit programming for these concepts. This capability could significantly advance fields such as robotics, autonomous vehicle navigation, and virtual environment design. Experts caution that while promising, the AI’s reasoning is still narrow and lacks the generalized, causal understanding that humans possess. The findings contribute to ongoing debates about the path toward more general machine intelligence. Read the full article at https://example.com/ai-physics-reasoning-study.