AI boosts scientific creativity by up to 10.2% in major study of 1M papers

A comprehensive study by Ding, Lawson, and Shapira analyzed over 1 million scientific publications from OpenAlex to measure how AI affects research creativity. They found AI-adopting papers are significantly more likely to achieve top-decile creativity, with a 5.5 to 10.2 percentage point higher probability. However, the mechanism matters: 'tool-oriented' AI research—applying existing models to domain problems—yields the largest gains in recombinant novelty (recombining existing ideas). In contrast, 'adaptation-oriented' AI—modifying models for specific tasks—produces higher object novelty (entirely new concepts).

This suggests AI does not advance science through a single pathway but through structurally distinct creative routes. The findings challenge one-size-fits-all science policy, arguing that evaluation frameworks must differentiate between recombinant and conceptual creativity. For practitioners, the implication is clear: the mode of AI integration matters more than AI adoption itself. The study calls for nuanced assessments that recognize how different AI adoption strategies produce fundamentally different scientific contributions.