Unlocking the Potential of AI Researchers in Scientific Discovery: What Is Missing? (2503.05822v2)

Abstract: The potential of AI researchers in scientific discovery remains largely to be unleashed. Over the past decade, the presence of AI for Science (AI4Science) in the 145 Nature Index journals has increased ninefold, yet nearly 90% of AI4Science research remains predominantly led by experimental scientists. Drawing on the Diffusion of Innovation theory, we project that AI4Science's share of total publications will rise from 3.57% in 2024 to approximately 25% by 2050. Unlocking the potential of AI researchers is essential for driving this shift and fostering deeper integration of AI expertise into the research ecosystem. To this end, we propose structured and actionable workflows, alongside key strategies to position AI researchers at the forefront of scientific discovery. Furthermore, we outline three pivotal pathways: equipping experimental scientists with user-friendly AI tools to amplify the impact of AI researchers, bridging cognitive and methodological gaps to enable more direct participation in scientific discovery, and proactively cultivating a thriving AI-driven scientific ecosystem. By addressing these challenges, this work aims to empower AI researchers as a driving force in shaping the future of scientific discovery.