Mathematics fields invited to help make AI safer and more controllable

Math for AI safety: an invitation for mathematicians

Artificial Intelligence

Summary

AI systems might become too complex for people to understand or control easily. The authors explain how different areas of mathematics—like logic, probability, algebra, and geometry—can help create AI that humans can better understand, guide, and cooperate with. They organize the discussion by math topic and suggest open problems in each area to encourage mathematicians to get involved. This approach aims to build AI systems that are safer and more aligned with human values.

What this means in practice

A position paper. It proposes an approach and reports no results.

Authors

Lionel Levine

Abstract

Artificial intelligence threatens to outrun human understanding and control. New mathematics is needed to design AI that is legible, steerable, and cooperative with humanity. I organize this invitation by mathematical field, so you can turn straight to your own: logic and game theory for cooperation; probability for agency and world-models; algebra and representation theory for learned features; analysis and geometry for generalization and training dynamics. Each section ends with an open problem that is accessible to a working mathematician with no prior experience in AI safety.