Incentivizing challenging drug discovery: from reducing failure to quantifying uncertainty

Drug discovery's central difficulty is less the elimination of failure than the measurement of uncertainty. This editorial argues that artificial intelligence (AI) helps mainly by converting unquantifiable uncertainty into measurable, priceable confidence, as one evidence stream among several.

It discusses the risk-uncertainty distinction, AI as a new layer in the computational drug-discovery stack, the cumulative nature of confidence (illustrated by KRAS, PCSK9 and the AI-discovered TNIK inhibitor rentosertib), discordant evidence, AI's limitations, and incentive design. Literature was identified from PubMed/MEDLINE, Scopus and Google Scholar to August 2026, with primary sources and trial registries; as an invited editorial, the search was narrative, not systematic.

AI's deepest contribution is epistemic rather than algorithmic: unlikely to lower failure rates markedly, but by making uncertainty measurable it can improve portfolio decisions and render ambitious biology attemptable - given calibration, prospective validation, open negative data, and incentives that reward reducing uncertainty.