How Open Source Artificial Intelligence Democratizes and Improves Science: Evidence from Structural Biology

Abstract

Artificial intelligence (AI) has the potential to advance scientific research by identifying patterns in large datasets and generating predictions. However, there is concern that the application of AI in science has been slow and uneven, limiting its broader impact. In this paper, we study whether open source can democratize the use of AI in science, using structural biology as an empirical context. We identify subfields where AI is applicable using keyword-based methods and apply a difference-in-differences approach to examine how the rate of research in these AI-relevant subfields evolved before and after Google’s release of the open source AI framework TensorFlow. We find that open source AI increases the number of publications in AI-relevant subfields without compromising quality, suggesting it meaningfully accelerates scientific progress. This increase is primarily driven by new researchers entering these subfields for the first time, many of whom are domain experts in structural biology who adopt AI as the barriers to its use decline. These new entrants also contribute to more novel research outputs. Together, our findings highlight the importance of openness in software frameworks for broadening access to AI and advancing science and innovation.

Keywords: artificial intelligence, open source, research tool, economics of science