Artificial intelligence is moving from solving mathematical problems to conducting mathematical research, with an Austrian scientist building a digital laboratory where autonomous AI agents develop theories, construct proofs and challenge each other’s findings.
Tamás Hausel, professor at the Institute of Science and Technology Austria (ISTA), has secured a Project Numina Fellowship to develop Project Sandbox, a computational environment designed to accelerate mathematical discovery.
The platform will function as a virtual research institute, deploying specialised AI agents to formulate conjectures, develop mathematical tools, construct proofs and independently review results.
Crucially, the system will not simply generate answers. Every accepted mathematical result must undergo computational verification and formal proof certification before entering the research record.
A computer algebra oracle will validate computational claims while an independent proof kernel determines whether mathematical arguments are logically correct.
Human researchers will retain control over scientific priorities and decide which discoveries deserve further investigation.
The initial research focuses on connecting representation theory and equivariant topology, two mathematical disciplines spanning algebra and geometry.
Hausel’s team has already developed an interactive visualisation interface called the ‘eyepiece’, with plans to introduce virtual reality capabilities.
The architecture builds on Numina-Lean-Agent, an interactive mathematical proof assistant introduced earlier this year.
Hausel reported that an early prototype became operational within three weeks, with researchers already proving an important mathematical result.
Supported by Paris-based nonprofit Project Numina, the initiative could redefine mathematical research by combining autonomous machine reasoning with independently certified proofs and human scientific oversight.
