John M. Jumper, Anne L’Huillier, William D. Phillips

Details

Inselhalle

Main Hall

Klara Bonneau
Freie Universität Berlin, Germany

Geoffrey E. Hinton
University of Toronto, Canada

John M. Jumper
Google Deep Mind, United Kingdom

Anne L’Huillier
Lund University, Sweden

William D. Phillips
National Institute of Standards and Technology (NIST), United States of America

Moderator: Heiner Linke
Lund University, Schweden

Abstract

Scientific understanding has long implied mechanisms and explanations, not just correct predictions. As black-box AI systems increasingly can generate hypotheses and guide experiments, questions arise. If an AI proposes a hypothesis that experiments confirm, but no human can explain why, has science advanced? Does opacity disqualify AI-driven results from “proper” science, or does it force us to rethink what understanding now means? For something to be “understood“ – who needs to understand it? What are the implications for the training of young researchers?