An essay proposes using AI to find connections between established methods
An essay proposes using AI to find connections between established methods
On October 9, John Rom Brink published an essay describing two kinds of scientific problems. Some require new data, concepts, or theory. For others, everything needed is already available in papers, software, and measurements, but the pieces remain scattered.
Brink sees the second kind as a distinct set of problems for AI to work on. Researchers may know the methods in their own field well yet miss how a technique from another discipline could answer their question. In his account, a model reads a broad body of work, connects its parts, and proposes applying a method from one field to a question in another.
“Everything needed to solve the problem already existed in the body of knowledge, but no one had connected the pieces.”
This framework was prompted by OpenAI’s mathematics release on October 6. The company published results, a repository, manuscripts, and proofs formalized in Lean, a language in which software checks the logical steps. Brink suggests that many of the mathematical results in this collection may have emerged as the model combined established techniques.
Claude’s work with ecological equations and genomic data has already provided a concrete example of this kind of search. In his account of the project, Matthew Schwartz, a Harvard physicist and visiting researcher at Anthropic, explained why a technically correct calculation still needs to be assessed by a scientist with expertise in the field.
“In almost every case, Claude was technically right, but the result became truly interesting only after an expert helped us choose a direction.”
In biology, this process depends on laboratory experiments. An agent can propose a connection or a calculation, and an experiment tests it in a living system. An acquaintance of Brink’s who works in virology and immunology considers this testing a bottleneck.
The practical task for an agent begins with three decisions: which scattered materials to bring together, which method to transfer, and which measurement to use to test the result.