The Next Great Scientific Theory is Hiding Inside a Neural Network

Explore groundbreaking applications of neural networks in scientific discovery through this Presidential Lecture at the Simons Foundation. Discover how machine learning tools traditionally used for tasks like text generation are being repurposed to uncover fundamental physics principles. Learn about an innovative methodology that focuses on interpreting trained neural networks rather than direct data modeling, allowing scientists to extract and improve theoretical understanding from the networks' learned patterns. Gain insights into the Polymathic AI initiative, a collaborative effort between the Flatiron Institute and global researchers that aims to harness ChatGPT-like technology for modeling diverse physical systems across multiple scales. Understand how this revolutionary approach is reshaping the scientific discovery process and opening new pathways for theoretical breakthroughs in physics and related fields.