AI is flooding science funders with proposals

As AI drives down the time & effort cost required to produce scientific grant proposals, funders are likely to be flooded with eligible, reasonable-seeming submissions that strain an already overburdened review system. It seems like we are starting to see the effects. Last week, the European Research Council (ERC) announced stricter application rules for 2027, not mentioning AI specifically but citing a “rapidly increasing number of applications.” The new policy increases the amount of time that unsuccessful applicants must wait before submitting another proposal (you cannot submit in 2027 if you received a “B” score on a proposal in the past two years, or a “C” score in the past three) and limits researchers to one application per year. Last summer, the NIH instituted a less strict but similarly motivated rule capping annual applications at six per investigator. Caps are one option for re-imposing application costs on researchers and encouraging self-screening, but they are a blunt instrument. Some other proposals for dealing with the influx include distributed peer review (which is being piloted by the UK Metascience Unit, and can produce more targeted application costs by scaling an applicant’s review requirements to the number of applications they submitted), tiered review (which could be a valuable tool, especially if combined with soft caps, but also runs the risk of increasing the number of applications by lowering the first-stage requirements), and AI-driven review (which is interesting and should be studied, though I remain wary of turning science into this cartoon). Figuring out the optimal solution is an open, and important, metascientific question.