Can AI compress years of biomedical research into months? And if it can speed up literature reviews, coding and data analysis, what happens when the next step still depends on an experiment?I’m joined by Stephan Reichl, a biomedical data scientist at CeMM and the Medical University of Vienna, to explore the promise and practical limits of AI in science. Stephan co-authored a study asking how much general-purpose AI could accelerate biomedical research—and where the bottlenecks would move.We discuss smarter experimental design, predictive readouts, AlphaFold, SimulateGPT, the challenges of adopting AI in research teams, and a question that matters more as these systems gain autonomy: who decides what “good science” looks like?GUEST AND RESOURCESStephan’s paper, “What are the limits to biomedical research acceleration through general-purpose AI?”:https://www.nature.com/articles/s41598-025-32583-w“GPT-4 as a biomedical simulator” (SimulateGPT):https://pubmed.ncbi.nlm.nih.gov/38909448/MrBiomics, the bioinformatics workflow project mentioned in the conversation:https://github.com/epigen/MrBiomicsFit For Science, Stephan’s podcast with Rob ter Horst:https://www.youtube.com/@FitForScienceAccelerate Europe:https://accelerate-europe.org/Subscribe to Tech & Drugs for conversations about data, AI and the future of drug R&D:https://www.youtube.com/@TechandDrugsWhat part of scientific research would you trust an AI co-scientist to handle—and where would you insist on a human decision? Tell us in the comments.#AIforScience #DrugDiscovery #TechAndDrugs