AutoDiscovery

AutoDiscovery is software that reads papers, analyzes data, and helps generate and test hypotheses on behalf of researchers.

Think of it as an extremely diligent research assistant joining your lab. This assistant scans thousands of papers overnight, digs through data, finds statistically notable patterns, and produces multiple hypotheses. But this assistant is good with numbers while still lacking common sense in medicine or biology, so it sometimes brings back ideas that look statistically plausible but don't actually make sense. That's why a final expert review step is essential.

Built by the Allen Institute for AI, this tool was actually used in cancer research through a collaboration with the Providence Swedish Cancer Institute, automating much of the process from literature search to hypothesis generation and validation. However, it was emphasized that results only become useful when researchers keep applying their own expertise and judgment.

Articles describe it like: "AutoDiscovery often finds statistically surprising hypotheses. The problem is that some of them turn out to be biologically nonsensical." A common misunderstanding is thinking this tool automatically validates hypotheses end-to-end. In reality, it's an assistive tool that always requires human expert judgment to follow.

AutoDiscovery publishes its run results on the web. Visiting the demo page lets you look at recently run research cases and see what questions were asked and what hypotheses were produced. It doesn't let you enter your own prompts directly, but it's useful for getting a feel for how AI generates hypotheses in real scientific research.