Aletheia: A New Approach to Mathematical Research

A recent study introduced Aletheia, an artificial intelligence (AI)-based mathematical research agent designed to automate the process of discovering and verifying mathematical solutions. Aletheia uses an advanced Gemini Deep Think model, enhanced by inference-time scaling techniques, to tackle complex problems that go beyond the level of the Mathematical Olympiads.

The agent is capable of generating, verifying, and revising solutions end-to-end in natural language, leveraging advanced tools to navigate the vast mathematical literature. Aletheia's deliveries include:

  • The autonomous generation of a research paper (Feng26) regarding the calculation of structural constants in arithmetic geometry.
  • Human-AI collaboration in demonstrating bounds on systems of interacting particles (LeeSeo26).
  • A semi-autonomous evaluation of 700 open problems on the Bloom's Erdos Conjectures database, with autonomous solutions to four open questions (Feng et al., 2026a).

The researchers suggest codifying standard levels to quantify the autonomy and novelty of AI-assisted results, in order to improve public understanding of developments in the field of AI and mathematics. The study concludes with reflections on human-AI collaboration in mathematical research.