Problems in 22 scientific fields had solutions hiding in plain sight. An AI has found them | Science | AAAS
Short summary
Researchers used BootLoops, an AI tool built with Claude, to scan scientific literature across 22 fields and identify problems that could be solved by borrowing techniques from unrelated disciplines. In 3 months, the team produced 36 papers solving problems ranging from calculating evolutionary tree probabilities to creating an 8x larger database of language sound systems.
- BootLoops discovered that many unsolved problems in different fields share identical mathematical structures with known solutions from physics, computer science, or math—but researchers didn't know to look across disciplines.
- The AI solved 15 previously unsolved integrals in high-energy physics and created ACCSTACK, a database of 6,000+ languages' stress and tone patterns that's eight times larger than the previous largest dataset.
- The approach requires heavy human judgment to work: the AI lacks "research taste" and can't determine what scientists actually care about, and it tends to favor older, highly cited work over new problems.
Full summary
A physicist at Harvard named Matthew Schwartz created BootLoops, an AI tool that works with large language models like Claude to spot solutions to scientific problems by recognizing when different fields face the same underlying mathematical structures. In three months, Schwartz and collaborators used this system to produce 36 preprints tackling everything from sunspot dating to health policy analysis to medieval manuscripts. The core insight is simple: many disciplines have unsolved problems that mathematicians, physicists, or computer scientists already know how to solve—they just don't know those solutions exist outside their own fields. Schwartz first developed BootLoops when he tasked Claude with solving a specific type of integral used in particle physics; the AI not only solved 30 of them but created software to do so, including 15 integrals that had never been solved before.
The concrete wins show real value. An evolutionary biologist used BootLoops to prove that two competing evolutionary trees for humans, chimps, gorillas, and orangutans had exactly equal probability—something previous methods couldn't establish. Linguists working with the tool built ACCSTACK, a database of stress and tone patterns across more than 6,000 languages that's eight times larger than the previous best effort. A health policy analysis identified roughly $1 billion in potential taxpayer-funded insurance bonus reallocations. These aren't theoretical demonstrations; they're actual research outputs that experts in their respective fields found valuable enough to publish.
The limitations matter. Schwartz admits that Claude could identify what it could *calculate* but lacked "research taste"—it couldn't judge what scientists actually cared about. The system also tends to favor well-cited, established findings over novel problems, potentially narrowing rather than broadening scientific exploration. Schwartz withheld some insufficiently validated results on colleagues' advice. Still, he argues that once human researchers learn to work effectively with these cross-disciplinary AI tools, the collaboration could unlock progress in fields where solutions were hiding but nobody was looking in the right places.