Barry Barish: Accelerating Basic Science with AI. Advancing AI through Fundamental Research.

In these closing remarks, Barry Barish returns to a question at the center of scientific discovery: how do we know when we have truly discovered something? Looking back to the birth of modern science, he traces the development of the scientific method through Copernicus, Galileo, Kepler, and Newton, then explains how the 20th century added statistical confidence as a central tool for evaluating evidence. Barish argues that the AI era creates a new version of this problem. AI may improve simulations, experiments, synthesis, analysis, and discovery workflows, but the scientific community still needs a convincing way to establish confidence in AI-assisted discoveries. Just as five-sigma became a shared standard for many areas of science, he suggests that AI for science now needs an equivalent framework for trust, validation, and statistical meaning. Outline of this video: 00:00 Returning to the question of discovery 00:28 Why discovery has been Barry Barish's lifelong focus 01:08 How do we believe a discovery in the era of AI? 01:33 The historical roots of modern science 01:53 Copernicus, Galileo, Kepler, and Newton 02:31 Newton and the scientific method 03:01 Updating the scientific method for machine learning and AI 03:41 How statistics changed discovery in the 20th century 04:21 Five-sigma and scientific confidence 04:41 Entering the AI era 04:50 The challenge of knowing when AI has helped make a discovery 05:03 Why AI-assisted discovery still faces a validation bottleneck 05:32 AI scores and why they are not yet enough 05:58 Why this remains an unsolved problem for AI and science 06:08 The need for an equivalent of statistical analysis 06:36 A challenge for the field