A Tel Aviv start-up called QED Science has used an AI system to rank more than 57,000 bioRxiv preprints and crown a 'top 1%' — a list chosen, the company insists, on scientific merit alone. The exercise, described in an interview published by Nature on 10 August, has reignited debate over whether machines should grade science before humans ever review it.
QED's tool reads manuscripts before publication and scores the originality and validity of their claims, training partly on negative results, failed replication studies and other evidence that does not support hypotheses — the counterfactuals the published literature rarely showcases. The company says its metrics ignore author identity, institution and publication venue by design, precisely to counter the prestige bias of the current system.
In its June analysis, QED selected 574 of 57,455 bioRxiv preprints as the top 1%. A validation study of 2,879 preprints that later appeared in peer-reviewed journals found QED rated 12.9% of them more highly than their eventual journals did — 'hidden gems' the company says the publishing system missed. In a blinded panel test, experts preferred the QED-favoured paper in 75% of decisive comparisons.
Critics worry the tool creates a new badge of prestige and reinforces metric-driven academia, and question whether AI can transparently judge scientific quality. QED's tool is free for researchers and has been used by more than 10,000 laboratories across 1,500 institutions in over 70 countries; the non-profit openRxiv has been piloting it on bioRxiv since November. The company says it has no plans to sell the system to publishers.
