MIT Researchers Say AI Is Moving Beyond Math Into Economics and Physics

Artificial intelligence is starting to take over theoretical work in fields well outside mathematics, according to researchers at MIT who spoke with WIRED. The group says AI is on track to become the standard method for attacking open problems in economics, high-energy physics and adjacent disciplines, a shift that would change how academic research gets done. The claim moves the conversation past recent demonstrations in pure math, where models have already produced proofs and conjectures that drew serious attention from working mathematicians.

The mechanism behind the expansion is not a single breakthrough but the fit between how these fields operate and what large models do well. Economics and high-energy physics both run on formal structures, symbolic manipulation and large bodies of published results, the same raw material that made mathematics a productive target for AI systems. Once a model can generate candidate derivations and check them against known constraints, the cost of exploring a theoretical idea drops sharply, and researchers can test more hypotheses than a single human career allows.

That prospect cuts both ways for the people who do this work. On one side, AI could compress years of grinding calculation into hours and surface connections across subfields that no individual researcher has read in full. On the other, it raises hard questions about credit, verification and whether a machine-generated derivation in economics or physics carries the same weight as one a human can defend line by line. The MIT researchers frame the change as inevitable rather than optional, which puts pressure on departments to decide how they will evaluate AI-assisted results.

The economics angle is the one most likely to draw scrutiny, because theoretical results in that field feed directly into policy models and forecasting. A physics derivation can be checked against experiment eventually, but an economic model often cannot, so an AI-generated argument that looks rigorous may still rest on assumptions nobody can test. That makes verification tooling, not raw generation, the bottleneck the field will have to solve first.

What to watch next is whether the major research institutions treat AI-assisted theory as a normal part of the pipeline or as a separate category requiring disclosure. Conferences, journals and funding bodies have already started writing rules for AI use in submitted work, and those rules will determine how quickly the shift the MIT group describes actually lands. The open question is who validates a machine's theoretical claim when no human has walked the full derivation.

For now, the signal from MIT is directional: mathematics was the proving ground, and the methods that worked there are being pointed at fields with messier data and higher stakes.