Google and FutureHouse release AI systems that generate and evaluate scientific hypotheses for drug discovery

Nature published two papers on Tuesday describing AI systems designed to help scientists develop and test hypotheses. Google's Co-Scientist operates as a "scientist in the loop," with researchers regularly directing the system, while FutureHouse's Robin can additionally evaluate biological data from standard screening assays like flow cytometry and RNA-seq.

Both systems are agentic (they call out to separate tools in the background) and address the challenge of navigating the explosion of scientific literature. Robin analyzed 551 papers in 30 minutes compared to an estimated time of 540 hours for a human. Both systems form hypotheses by identifying connections across fields and rank them through comparison methods before human experts review and prioritize suggestions.

In tests on disease targets, Co-Scientist identified drugs effective against subsets of acute myeloid leukemia cells, while Robin suggested drugs and assays to test for macular degeneration and proposed that increasing retinal cells' ability to pick up debris outside cells could address the disease mechanism.