Faster research tools may produce more scientific papers without improving the quality of each one, according to a theoretical economics study. The authors modelled how researchers allocate limited time when language models make parts of a project quicker.
The analysis deliberately assumes ideal tools: the language models save time, introduce no errors, and cost almost nothing. That strips away familiar concerns such as hallucinations and lets the researchers isolate a different effect. When routine work becomes cheaper, a scientist’s remaining time becomes more valuable, increasing the incentive to start another project rather than polish the current one.
The model adapts “optimal foraging” theory, a framework used to study how organisms divide effort among opportunities. A simulated research project begins with checking whether an idea is viable, followed by a decision to abandon it or invest in completing it. When AI shortens parts of that cycle, researchers can explore more ideas, but in two of the three scenarios modelled, effort and quality per publication decline.
This is a theoretical result, not evidence that real laboratories have already become less rigorous. Institutions can also change incentives, review standards, and funding rules. The paper’s useful warning is that productivity tools do not determine where saved time goes. If academic rewards continue to favor output volume, AI efficiency may amplify that incentive instead of giving every project deeper attention.