SCI-AI · Scientist in the loop - Automation of scientific discovery

Where do new scientific ideas come from? That is the question at the root of research on scientific discovery. In the past decade, AI-augmented methods have shaken the practices of discovery across a range of research fields: AlphaFold solved the long-standing protein folding problem; automated discovery methods are being used to decipher ancient scripts once considered unreadable, and to synthesize drugs for previously untreatable diseases. According to the most enthusiastic advocates, in less than a decade, AI will transform practices of research beyond recognition.Should the novel scientific findings generated by AI-powered systems like AlphaFold be considered as instances of machine creativity? When does human-AI collaboration lead to successful scientific problem solving, and when does it fail? Are AI systems reconfiguring the scientific method? Currently we cannot answer such questions, because systematic empirical studies of problem-solving activities in AI-augmented labs have not been conducted. Nor do we have the conceptual tools needed to evaluate the rationality of problem solving in AI-augmented labs. That is the gap the SCI-AI project addresses. The main aim of the project is to generate understanding of the AI-augmented discovery transformation and its consequences for the development of human knowledge. The main objectives are:(A) to uncover and analyze problem-solving processes in AI-augmented labs, (B) to examine the implications of AI-augmented discovery for the epistemology of science, and (C) to use automated discovery as a mirror for exploring human creativity and intellect.To reach these objectives, the project combines philosophical analysis with computational models of problem solving and ethnographic fieldwork that we conduct in three frontier AI labs (a self-driving lab in biology, a closed-loop behavioural science lab, and a virtual lab using digital twin technology in drug development and atmospheric research).