Using Generative AI to Bring New Perspectives to Medical Problem Solving: Peter Lee, Microsoft & Brandon Pletsch, Real Chemistry

·45 min·3 clips

Peter Lee says it is “exceptionally dangerous” to let generative AI propose an initial diagnosis today.

Aaron Strout opens the show by setting up a conversation about AI, healthcare, and practical innovation. He introduces Brandon Pletsch as Real Chemistry's Practice Leader of Scientific Visualization and one of the company's leaders in AI. Brandon's guest is Peter Lee from Microsoft, whose role is to incubate new research powered products and move them toward production, working across AI, computing foundations, health, and life sciences. Peter's background also includes senior roles at DARPA and Carnegie Mellon University. The discussion starts with a basic question: what should generative AI actually do in medicine? Peter argues that the wrong way to think about it is as a traditional computer. It makes more sense, he says, to treat it as a reasoning engine. That matters because it changes the expectations people bring to it. The episode keeps coming back to the gap between generation and critique. Peter points out that one of the model's most useful abilities is to review a clinician's thinking and offer a second look. He uses a differential diagnosis as the example, where a doctor can ask the model to look at the labs, the presentation, and the list of possibilities. The point is not to replace the doctor. The point is to catch what might have been missed. The conversation also gets specific about where the model cannot be trusted. Peter explains that if you ask it to read a chapter of a book or do complex arithmetic without tools, it can fail in ways that feel oddly human. It may hallucinate when it tries to do something it cannot reliably do. That limitation is part of the argument, not just a caveat. Brandon and Peter then return to the idea that critique and revision may be the real superpower of generative AI. The model can help sharpen a thought, test an assumption, or push a clinician to revisit an early conclusion. The tone stays warm and practical, with enough technical language to be useful and enough explanation to keep it accessible. The episode keeps bringing the discussion back to healthcare use, not AI theater. It asks what this means for doctors, for institutions, and for the broader experience of care. Near the end, the conversation turns playful with a comparison between a human written question and a ChatGPT written question. That moment works as a small test of the listener's assumptions. It also reinforces the show's larger point: AI can help, but human judgment still matters. The closing lands on appreciation rather than certainty. The show invites listeners to keep thinking about how AI and ideas can come together to transform healthcare in practice, not just in theory.

As heard by us

A grounded look at AI as a clinical second opinion.

This episode treats generative AI as a working clinical reasoning aid, not a novelty, and Peter Lee makes that case clearly. The strongest stretch is the discussion of differential diagnosis, where a model can review labs, the initial presentation, and a working diagnosis to…

Read the full review in PlayNext →

Why you'd press play

Want a clearer way to think about AI in clinical decisions?

Read the full recommendation in PlayNext →

Listen to the show on

Apple Podcasts Spotify

Play clips from this episode More from The Real Chemistry Podcast