Trusting AI in Pharmacovigilance: From Proving It Works to Scaling Responsibly

In this AI Exchange podcast episode, Jason Bryant (ArisGlobal) and Sam Wallace (BMS) discuss how the conversation in life sciences has shifted from whether AI works to whether organizations are ready to trust it as it moves beyond workflow automation into judgment-heavy pharmacovigilance work. They describe progress at BMS in adverse event processing, including strong and improving F1 accuracy scores, expanded data extraction coverage, and the challenge of reducing 100% QC as trust, governance, and change management lag behind rapidly advancing technology. They explore perception gaps about AI maturity, the limits of pursuing "perfection" in probabilistic systems, and the promise of agentic AI, evaluators, and human-in/on-the-loop models to enable risk-based, regulator-friendly oversight, aiming toward touchless, explainable processing while prioritizing people, process simplification, and durable customer value over model novelty. 00:00 Welcome and Introductions 00:18 From Works to Trust 01:37 Agentic AI vs Humans 02:46 18 Months of Change 04:52 Perception vs Reality 07:13 Value Over Perfection 10:34 Why BMS Started 11:58 Scaling and Agentic QC 15:18 Trust in Practice 17:09 Designing for Trust 19:18 Agentic Work Delegation 19:56 Benchmarking Long Tasks 21:09 Redesigning Complex Workflows 22:52 Separating Hype From Reality 23:57 Vision Execution Governance 25:32 Touchless Safety Processing 27:08 Ask Me Anything Reflections 28:46 Fear Doubt And Trajectory 30:49 Predictability And Hybrid AI 11:17 People Over Technology 34:58 Closing Takeaways Trust