Stefano Bianchini (Univerity of Strasbourg)

Artificial intelligence in science: Promises and perils for creativity

Abstract: Scientists are enthusiastically embracing AI/ML tools to improve their research. While these technologies hold great potential for scientific discovery, recent concerns have been voiced about their possible adverse effects. Can AI/ML truly boost productivity and lead to more innovative science? Thus far, answers to this question remain anecdotal, often limited to a handful of disciplines. In this paper, we study the diffusion of AI/ML across more than 80 scientific fields from 2000 to 2022 and its impact on creativity – measured through novelty and impact. We find that AI/ML adoption has accelerated in nearly all disciplines since the early 2010s, with research activity becoming increasingly concentrated in three major regions: the EU27, the US, and China. Our analysis confirms an overall positive effect of AI/ML on scientific creativity, though with significant heterogeneity across fields: while many benefited, others experienced no gains, and some even encountered negative returns. We propose that the characteristics of the knowledge space – and, by extension, the patterns of knowledge production – within a scientific field may moderate the influence of AI/ML. Specifically, we show that AI/ML has greater transformative potential in “rough” knowledge spaces, where human cognition may struggle to navigate complexity. These findings are contextualized within recent policy initiatives designed to promote AI-powered science.