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 trained a large language model on vast datasets of text and 3D object simulations, finding it could accurately predict outcomes like stability, collisions, and support relationships between objects. The AI’s performance on benchmark tests rivaled that of humans, suggesting foundational models are acquiring a form of intuitive physics. This emergent capability could significantly advance fields like robotics and autonomous systems, though the researchers caution that the AI’s understanding remains narrow and tied to its training data. The full findings are detailed in the article ‘AI Models Show Surprising Ability to Understand Basic Physics’ at https://technologyreview.com/2024/05/15/ai-physics-reasoning.