Speaker: David Gunderman
Medicine has long relied on numbers, formulas, and models to classify patients, predict risk, and guide decisions. Machine learning and AI have made these tools more powerful, but also more opaque: physicians and patients may rely on outputs without fully understanding how they were produced or what assumptions they contain. Is this opacity a new problem, or part of a much longer history of medicine’s trust in mathematics? Drawing on examples from the four humors to modern AI, this talk explores how mathematical models shape real clinical decisions and patient lives. Questions that may seem purely technical, such as measurement, modeling, thresholds, and prediction, also turn out to be questions about ethics, communication, bias, and responsibility.
David Gunderman is an incoming resident physician at Ascension St. Vincent Hospital in Indianapolis and a future dermatologist whose work brings together medicine, mathematics, and public health. A summa cum laude graduate of Wabash College with majors in Mathematics and German, David received his PhD in Applied Mathematics from the University of Colorado Boulder and his MD from Indiana University School of Medicine. He was a Fulbright Fellow in Germany in 2015 and the Lillian Gilbreth Postdoctoral Fellow at Purdue Engineering and Stanford Medicine in 2021. During medical school, Gunderman served as President of the Indiana University School of Medicine Medical Student Council, was named one of the Premier 10 graduate students across IU Indianapolis, and received the national U.S. Public Health Service's Excellence in Public Health Award. He serves on the board of Mental Health America–Wabash Valley Region, a nonprofit serving 11 counties and more than 500,000 residents in north-central Indiana. He has taught college- and graduate-level statistics, mathematics, and computer science for nearly a decade and has published more than 25 peer-reviewed research articles.
Program: From Antiquity to AI: Mathematics, Medicine, and the Black Box Problem
Speaker: David Gunderman, MD, PhD, resident physician, educator, researcher and much more
Introduced By: Bill Halsema
Attendance: NESC: 115; Zoom: 30
Guests: Jenny Hootman, Stou Spinola
Scribe: Doug Ellrich
Editor: Carl Warner
Talk’s Zoom recording found at: https://www.scientechclubvideos.org/zoom/09212026.mp4
David Gunderman, MD, PhD, is an Ascension St. Vincent resident physician in Indianapolis planning to specialize in dermatology. A Wabash graduate and Fulbright Fellow, he is fluent in German, earned a doctorate in applied mathematics at the University of Colorado Boulder, completed postdoctoral work at Purdue Engineering and Stanford Medicine, and earned his medical degree at Indiana University. He has taught mathematics, statistics, and computer science and published more than 25 peer-reviewed articles.
Gunderman explored medicine’s “black box” problem: a number may shape care even when neither physician nor patient can fully explain how it arose. A model may be opaque because its formula is proprietary, its statistics unfamiliar, or its AI calculations too complex to trace. Medicine is full of uncertainty, and diagnosis, prognosis, and treatment decisions depend on probabilities. AI estimates probabilities from data, but their value depends on the quality and relevance of those data and the clinician’s judgment.
The problem predates AI. The four-humor theory long justified bloodletting, and likely killed George Washington. Ignaz Semmelweis (1818 – 1865) showed that handwashing saved lives before germ theory explained why. The body mass index (BMI) illustrates how a useful population guideline can become a rigid threshold for an individual. Evidence-based guidelines help, but physicians must examine how studies were designed, whom they included, and whether their findings apply to the patient in front of them.
The Black Box of AI systems can learn misleading signals: a skin-lesion classifier treated physicians’ ink marks near moles as evidence of cancer, while a COVID chest X-ray classifier appeared to rely on equipment clues instead of the lungs. Testing on new cases can expose such errors. Gunderman urged clinicians to ask who built a model, what it predicts, whose data trained it, where it works, what biases it may have, and who is responsible for its use. AI may assist with documentation and focused diagnosis, but its output requires skepticism, testing and human review. Dr. William Osler’s maxim captures the duty to attend to the patient first: “See, and then reason and compare and control. But see first.”
David Gunderman
