I just came back from a fascinating session at ASCPT 2026 titled “From Black Box to Better Health: Debating AI’s Role in Clinical Pharmacology and Drug Development.”
The discussion highlighted the shift from point solutions (single-task AI) toward workflow intelligence, where multimodal foundation models and specialized agents support complex PK/PD workflows. The idea of an “AI co-scientist”—coordinating data retrieval, modeling, simulation, and reporting while keeping a human in the loop—was presented as a possible future direction by the AI champignons.
The counterpoint from the “Black Box Warning” team focused on a familiar tension in our field: predictive performance vs. mechanistic understanding. They argued that true progress in clinical pharmacology requires models that capture biological mechanisms so they can adapt to new contexts and changing systems.
Several polls during the session showed the audience leaning toward the more cautious perspective. Personally, I often agreed. We have heard many sales pitches about AI, but what we really need are generalizable, scalable workflows that are accessible to the broader community.
The takeaway for me: we shouldn’t have to choose between predictive performance and scientific understanding—we should demand both.
Curious to hear from the community: where have you actually seen AI improve PK/PD or clinical pharmacology workflows in a way that is both scalable and scientifically interpretable?