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 Google DeepMind trained a neural network on vast datasets of simulated physical interactions, such as stacking blocks or predicting how a rolling ball will bounce. The AI, dubbed ‘PhysNet’, demonstrated the capacity to make accurate predictions about novel scenarios it had never encountered during training, suggesting the emergence of a form of intuitive physics. This development could significantly accelerate robotics and autonomous systems, allowing machines to better navigate and manipulate the real world. However, the researchers caution that the AI’s reasoning remains narrow and task-specific, lacking the broad, causal understanding that humans possess. The findings open new avenues for research into how machines can learn fundamental concepts about the world. Read the full article at https://example.com/ai-physics-reasoning-study.