AI for Scientists: Using Cognitive Task Analysis to Guide Scientific AI

Advancements in artificial intelligence (AI) have the potential to fundamentally change how science is conducted. To ensure scientists can employ AI tools effectively, design should be grounded in a deep understanding of the work context. The current study proposes using a modified cognitive task analysis to map out scientific workflows and decision-making processes. Following this approach, we conducted interviews with scientists from two structural biology case studies, genome annotation and protein crystallisation. Analyses revealed the major challenges facing scientists and the strategies they employed to overcome them. We also identified key opportunities and implications for AI in workflows, capturing findings in a novel workflow diagram. Notable contextual differences emerged across the two case studies, highlighting the need for tailored AI solutions for different problems. Results highlight the utility of cognitive task analysis in understanding scientific workflows in their unique context, paving the way for strategic implementation of AI in science.