Abstract
Recent advances in artificial intelligence (AI) find rapidly increasing applications for research in science, and not the least in physics and its related areas. Ranging from controlling experimental arrangements for advanced measurements to analysis of big and complex data sets to the prediction of molecular structure and interactions, AI techniques quickly emerge as powerful tools in the physical sciences. Going a step further, we cannot rule out the possibility that AI models may soon even be used to identify what may be new laws of physics.
While holding great promise, these advances also raise serious questions. Can established principles of transparency, replicability, falsifiability, and accountability be applied to the use of AI, or do they need to be fundamentally revised? Can we trust AI-generated conclusions from big data sets? What degree of human control, involvement and rational understanding is required for something to be called “physics” or “science”: can something be physics if only a machine “understands” it? Do we need new tools to prevent incorrect science to be used as a basis for decision making in government, business, and society? Can physics continue to work as a self-contained discipline, or is broader interaction with other disciplines such as ethics and behavioural sciences a necessity?
