Fishbowl Panel: What Is Truth, Anyway？ (And Does It Matter？)

This fishbowl panel, moderated by Addison Snell of Intersect360 Research, rotates audience members through the stage to debate what ground truth means for AI-driven science. Participants include Martin Schulz (Technical University of Munich), Tara Murphy (Quantum Motion), Andrew Shea (HPE), Ryo Yokota (Institute of Science Tokyo and RIKEN), Jeanette Lorenz (Fraunhofer), Niklas Roemer (ETH Zurich, Next Generation Committee), and Steve Frank (Sofa Compute). The discussion opens with the tension between simulation, which encodes assumed physical truths, and AI models trained to produce the most likely rather than the correct answer, touching on uncertainty quantification, science skepticism in climate and weather forecasting, and how reinforcement learning phases nudge language models toward truthful responses. Panelists argue that hypothesis testing and property testing can ground AI-for-science outputs much as physics validated unobservable entities like electrons. The conversation then turns to national sovereignty following a US executive order on quantum and a new Chinese system topping the Top500, with discussion of international foundation-model collaborations, weather data as a strategic resource, and resiliency as the practical meaning of sovereignty. Further threads cover how to communicate quantum computing accessibly, including a children's coloring book on qubits, whether gigawatt-scale AI data centers built today will be obsolete within a few years, and how digital twins combine multiple domains with heterogeneous computing. The panel closes by emphasizing intergenerational learning and questioning AI rather than merely using it.