What Clinical AI Is Doing Well... and Where It Still Needs Work With: Dr. Peter Brodeur
As clinical AI adoption accelerates, healthcare leaders face a critical question: what is truly ready for practice versus what remains experimental? In this conversation, Dr. Peter Brodeur, a physician-researcher at Beth Israel Deaconess Medical Center and contributor to the ARISE Network report on clinical AI, breaks down where the field stands today. Drawing from a comprehensive review of emerging evidence, he highlights both promising advances and persistent limitations in AI-driven clinical decision support, patient-facing tools, and workflow integration.
The discussion explores the concept of the “jagged frontier,” where AI systems demonstrate superhuman performance in controlled settings yet fail unpredictably in real-world scenarios. Dr. Brodeur also emphasizes the growing importance of safety evaluation, human-computer interaction design, and outcomes-based benchmarking. For healthcare executives, clinicians, and digital leaders, this interview provides a grounded, evidence-based perspective on how to responsibly implement AI while improving patient outcomes and operational efficiency.
Episode Contents:
About the Guest
Dr. Peter Brodeur is an internal medicine resident and incoming cardiology fellow at Beth Israel Deaconess Medical Center, affiliated with Harvard Medical School. He is a member of the Harvard-Stanford ARISE Network and serves as a reviewer for Nature Medicine and NEJM AI. His research focuses on human-computer interaction and large language model clinical reasoning. Connect with him on his LinkedIn profile of Peter Brodeur.
Key Takeaways
- Clinical AI shows promise but remains inconsistent in real-world scenarios
- Human-AI collaboration design is as critical as model performance
- Patient outcomes, not benchmarks, should define AI success
Transcript Summary
What is the current state of clinical AI?
Clinical AI is advancing rapidly, but performance varies widely depending on context. While models excel in controlled environments, real-world deployment reveals inconsistencies and limitations.
What does the “jagged frontier” mean in clinical AI?
The “jagged frontier” describes how AI systems can outperform clinicians in some tasks while failing in simple variations. This highlights gaps in true reasoning capability and raises concerns about reliability in clinical settings.
Why is safety a growing concern?
Most studies have focused on accuracy rather than patient harm. Emerging research shows that AI can introduce risks through omission or incorrect recommendations, underscoring the need for safety-first evaluation frameworks.
How should clinical AI be evaluated moving forward?
Traditional benchmarks like multiple-choice exams are reaching saturation. The field must shift toward outcome-based evaluation, focusing on whether AI improves real patient care rather than test performance.
What role does human-computer interaction play?
Even high-performing models may fail if poorly integrated into workflows. Effective design must ensure that clinicians can interpret, trust, and appropriately act on AI recommendations.
What are the risks of patient-facing AI?
Patients often cannot distinguish between high- and low-quality advice. Studies show they are equally likely to follow incorrect guidance, making it unsafe to rely on patients as oversight mechanisms.
Where is AI already showing real-world impact?
Subspecialty applications, especially in imaging and diagnostics, demonstrate measurable improvements. Examples include increased cancer detection rates and faster stroke triage times.
What is holding back broader clinical AI adoption?
Barriers include regulatory friction, safety concerns, and the challenge of conducting large-scale randomized trials. However, healthcare systems now have an opportunity to lead implementation science.
What did the field get right recently?
Healthcare organizations are increasingly adopting AI for workflow improvements, such as documentation and operational efficiency. This signals growing trust and readiness for broader integration.
More Topics
- AI in Patient Care
- AI in the Healthcare Industry
- AI and Medical Innovation
- Healthcare Ethics and Policy
Keep Exploring
About the Series
Leading oncology AI thought leaders Drs. Sanjay Juneja, Debra Patt, and Doug Flora bring you conversations at the intersection of medicine, data, and innovation. Each episode explores both the big picture and the breaking news in artificial intelligence and healthcare—examining how today’s technology is reshaping the practice and business of oncology.
From industry disruptors to clinical pioneers, guests share insights that bridge the gap between algorithms and the art of patient care.
To learn more about the mission and upcoming initiatives, visit tensorblack.ai and subscribe to stay ahead of the curve.