Title: AI in Science: Insights from the frontier

 Abstract: Google DeepMind's essay,"A New Golden Age of Scientific Discovery,"  explores how artificial intelligence (AI) is accelerating scientific discovery across various disciplines, and impacting the scientific process itself. The essay draws on interviews with GDM  scientists who developed high-impact models in biology (AlphaFold, AlphaMissense), weather (GraphCast), materials science (GNoME), and mathematics/computer science (AlphaTensor, FunSearch). It also draws on internal research using scientometric data and techniques. While AI offers significant opportunities to tackle complex scientific problems, automate experimentation and extract insights from large bodies of research and data, it also raises critical questions for economists, social scientists and policymakers. These include securing the necessary inputs for AI in science (including problem selection, data, benchmarks, skills, and model access) and addressing challenges like AI's impact on scientific creativity and understanding, research reliability, and equitable access. We summarize the situation in these areas, based on our experience and the literature, and develop a policy agenda to focus AI in science efforts on high-impact areas, provide the required data, benchmarks, and skills, and encourage analysis and experimentation to better understand AI adoption and impact in science, all with the goal of accelerate AI's positive impact while mitigating its risks. 

Bio: Juan Mateos-Garcia is an impact data analyst at Google DeepMind, where he uses data science to track AI adoption and impacts for strategy, communications, and policy. Previously, Juan led the data science team at Nesta, the UK innovation foundation. Juan is an economist with an MSc in Science and Technology Policy from the University of Sussex.