Published September 4, 2026
As part of its ongoing exploration of human agency in the era of AI, the IC² Institute conducted in-depth interviews with 20 doctors about the impact of AI on their clinical reasoning skills. The doctors came from leading medical centers and represented a broad range of specialties, including internal medicine, radiology and neurology.
Use of AI is on the rise: A 2026 American Medical Association survey found that 81% of doctors are using AI in their practice — more than double the 38% reported in 2023. While there’s a growing body of research measuring the performance of AI against that of human physicians (and AI is outperforming the humans, especially at diagnosing patients), there’s a need for more research into how AI is impacting doctors’ clinical reasoning skills. The IC² Institute turned to doctors who are at the cutting edge of AI use to learn more about how their clinical reasoning is evolving.
Through the interviews, the researchers took a deep dive into daily workflows, explored the process of making a diagnosis and managing patient care, and probed how AI is changing their practice of medicine and impacting clinical skills.
Here are five key takeaways:
1 – Some doctors are REskilling.
Several observational studies suggest that increased reliance on AI may diminish physicians’ clinical abilities, resulting in physician “deskilling.” But the IC² interviews point to something different, which is that some physicians are experiencing “reskilling.”
Rather than diminishing cognitive engagement, AI can push clinicians to develop skills that enable them to guide, interpret, and evaluate AI outputs. New skills include things such as re-sequencing clinical questions for ambient listening tools, editing AI-generated patient notes, and conducting risk assessments of AI tools.
Further, AI-generated outputs may expand doctors’ own thinking about possible diagnoses or nudge them to consider alternative treatment plans. Often, this leads to a back-and-forth dialogue between clinician and AI, where doctors are using AI as a thought partner — and clinical reasoning takes on a new shape.
2 – Clinical workflows are changing.
The integration of AI into clinical practice is yielding a new, expanded workflow. Even before an AI tool is used, clinicians must assess whether AI is applicable and trustworthy for a given clinical situation. A doctor might decide, for example, that ambient listening tools may not function well for specific patients due to communication challenges.
Once tools are vetted and selected, doctors are interacting with them across the workflow — refining prompts, adjusting settings, adding contextual details to responses, probing uncertainty. Doctors are also using their judgment to critically evaluate outputs. Is the AI-generated diagnosis accurate? Is the AI-generated treatment plan plausible? If the outputs are deemed inadequate, doctors may restructure patient questions, refine AI prompts, apply constraints to the model, or reject an AI output altogether.
Finally, the physician must decide how to translate AI outputs into clinical action — deciding how to incorporate outputs to refine diagnoses, create treatment plans, and support shared decision-making between doctor and patient.
3 – If human touch/empathy/connection is the #1 uniquely human skill within health care, then “contextualizing” is a close second.
Among the interviewees, there was consensus that the human handholding (physical and emotional) is the most obvious and primary element that sets a physician apart from AI in the delivery of health care. But many doctors identified contextualization as a second, uniquely human domain. While AI can rapidly retrieve and synthesize data and guidelines, it’s up to the human to layer in contextual elements, especially in the outlining of treatment plans. These contextual elements include things like patient preferences, values, and living conditions; co-morbidities; care setting; hospital/clinic resources; and insurance constraints.
4 – “Responsible AI” demands considerable pushback from doctors across all phases of the workflow.
A common theme among the interviewees was the importance of pushing back against AI. This “pushback” must begin even before an AI tool is applied, with questions about how the model actually works, how it was trained, what data is behind it. The pushback continues as clinicians fill holes in patient encounter summaries and challenge AI-generated diagnoses and treatment recommendations. Only as clinicians learn to effectively spar with the large language models can they leverage AI’s potential as a thought partner and maintain human agency within changing workflows.
5 – Medical education needs to transform — and fast.
Medical schools are scrambling to stay on the front edge of AI technology and determine how and when to introduce it to their students. Several of the interviewed doctors noted the benefits of integrating AI into medical education. They are especially excited about using AI to generate vignettes for their students, exposing their students to a greater variety of scenarios than is possible with actual patients. But should medical students be given immediate access to AI? Will it expedite and/or deepen their learning? Or will early access and easy shortcuts stunt their ability to synthesize information and reason?
What’s next for the Institute?
Questions about AI and medical education have prompted IC² researchers to team up with Dell Medical School on an investigation scheduled to launch this fall, the AI Learning Lab. Researchers will use simulated patient environments to examine the clinical reasoning of students, residents and faculty.
S. Craig Watkins, executive director of the IC² Institute, laid out the goals for the lab: “We hope to understand, empirically, the benefits and risks of integrating AI technologies into the clinical decision-making process. Moreover, how can the findings be used to design a curriculum that incorporates AI but doesn’t compromise students’ ability to develop the clinical and analytical skills that they need?”
Future investigations like the AI Learning Lab will expand the Institute’s understanding of human-AI collaboration, the potential of agentic AI, and the preservation of human agency in an increasingly AI-impacted world.
NOTE: The IC² Institute research team for this project included TuQuyen Dao, Matt Kammer-Kerwick, Sookja Kang, Emily Spandikow and S. Craig Watkins.