IOL @ Speinshart Scientific Center for AI and SuperTech

Methods of artificial intelligence are moving from being an object of mathematical research to being an instrument of it. Using large language models to develop ideas and generate code has become part of our daily work. Two further uses are developing quickly and are much less settled. Optimization and learning methods can produce mathematical structures, examples and counterexamples that give a researcher the idea for a formal proof, and agent systems can draft solution strategies on their own and validate them against computer algebra systems and numerical experiments.

This retreat brings the IOL Lab at the Zuse Institute Berlin to the Speinshart Scientific Center for three days devoted to those two uses, which are the subject of a great deal of current work and where the methodology is still open. We want to get past first experiments and work out a coherent research strategy, one that says where these tools already carry weight, where they do not, and how to integrate them into our research in mathematics and beyond.