Progress in mathematical AI models has led to a surge of discussion and excitement regarding the capabilities of these tools and the future of the field. However, we mathematicians have had little influence on the narratives around this technology. The major math AI players are in a constant war of headlines and hype. Every advance in contest mathematics or AI-assisted resolution of a conjecture gets heralded as the advent of superintelligence, and untrained individuals flood arXiv and journals with slop co-written with sycophantic bots.

Unfortunately, laymen have little reference for what mathematical capability “looks like” and most non-mathematicians, even in other technical fields, would struggle to name key milestones of progress between “acing math olympiads” and “solving Millennium prize problems.” In absence of proper incentives, AI companies rush towards low-hanging fruit that generate good headlines with little serious vision or long-term strategy for deep autonomous research.

Now is the time for mathematicians to come together and define for the world what are the potential developments in AI that would be exciting, useful, and valuable to the community. What are the facets of an autonomous math researcher or assistant, and is it possible to measure progress in the relevant capabilities?

Any answer to these questions must emphasize the diversity and multidimensionality of mathematics. This workshop is intended to produce a list of milestones marking levels of progress in each of roughly 8 categories of assistance or reasoning a math AI may provide researchers. These categories or “skillsets” will be determined on the first day, by a consensus of the participants on what would be most valuable to the community, but will likely be broad and fundamental abilities such as “Problem solving,” “Generalization,” “Exposition,” “Formalization,” and “Conjecturing.”

This workshop is co-sponsored by the Institute for Computer-Aided Reasoning in Mathematics and Principia Labs. Principia Labs is graciously providing the workshop venue near its offices in San Francisco, and will handle reimbursements for invited participants, as described in the invitation letter. Representatives of several AI labs have been invited to attend.