#309 - What AI Sees That Doctors Can’t | Dr. Ziad Obermeyer & Mike Haney

<p>Medical AI is already getting good at doing things doctors do. Dr. Ziad Obermeyer thinks the bigger opportunity is using AI to discover things medicine doesn’t yet know.</p><p><br></p><p>His work shows both sides of that future: algorithms can amplify bad assumptions when trained on the wrong targets, but they can also uncover signals in medical data that humans miss. Obermeyer argues that as more health data is collected outside the hospital, AI could turn it into a continuous picture of health rather than a series of isolated snapshots.</p><p><br></p><p><strong>Free course: Improve your metabolic health</strong></p><p>Get our free email course on how glucose, nutrition, exercise, sleep, and measurement can help you build habits that support better energy and long-term health: <a href="https://www.levels.com/partner/multimedia?utm_source=podcast&utm_medium=link&utm_content=nextlevel&partner=Obermeyer" target="_blank" rel="ugc noopener noreferrer">⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠https://levels.link/wnl⁠</a></p><p><br></p><p><strong>What We Cover:</strong></p><ul><li>Why AI should learn from patients and outcomes, not just doctors</li><li>How an ECG model found hidden risk of sudden cardiac death</li><li>Why medical data access is still a major bottleneck</li><li>How more measurement could actually mean fewer unnecessary tests</li><li>Why healthcare may be entering a “mainframe to PC” transition</li></ul><p><br></p><p><strong>🎙️ About the Guest:</strong></p><p>Dr. Ziad Obermeyer is an emergency medicine physician and an Associate Professor at the UC Berkeley School of Public Health. He also co-founded Dandelion Health and the non-profit Nightingale Open Science. </p><p><br></p><p><strong>📍What Dr. Ziad Obermeyer &amp; Mike Haney discussed:</strong></p><ul><li>02:43 Why better data is fundamental to medical AI</li><li>05:54 The problem: There’s no variable called “get sick”</li><li>09:27 Algorithms optimize exactly what you tell them to</li><li>23:03 Why teaching AI to copy doctors limits what it can discover</li><li>26:28 How AI could create a new science of medicine</li><li>28:40 Could AI predict sudden cardiac death before it happens?</li><li>34:14 Can an algorithm teach us what it sees?</li><li>36:50 Medicine’s biggest AI bottleneck: access to data</li><li>49:00 Healthcare’s “mainframe to PC” transition</li><li>55:08 Why more measurement could actually mean fewer tests</li><li>58:46 Why medical AI is aiming too low</li><li>1:02:13 What the next generation of wearables needs to measure</li></ul><p><br></p><p><strong>🔗 Helpful Links:</strong></p><ul><li><strong>Dissecting Racial Bias in an Algorithm Used to Manage the Health of Populations</strong> (<em>Science</em>, 2019)<a href="https://www.science.org/doi/10.1126/science.aax2342" target="_blank" rel="ugc noopener noreferrer">https://www.science.org/doi/10.1126/science.aax2342</a></li><li><strong>An Algorithmic Approach to Reducing Unexplained Pain Disparities in Underserved Populations</strong> (<em>Nature Medicine</em>, 2021)<a href="https://www.nature.com/articles/s41591-020-01192-7" target="_blank" rel="ugc noopener noreferrer">https://www.nature.com/articles/s41591-020-01192-7</a></li><li><strong>An ECG Biomarker for Sudden Cardiac Death Discovered with Deep Learning</strong> (<em>Nature</em>, 2026)<a href="https://www.nature.com/articles/s41586-026-10674-6" target="_blank" rel="ugc noopener noreferrer">https://www.nature.com/articles/s41586-026-10674-6</a></li><li><strong>Predicting the Future — Big Data, Machine Learning, and Clinical Medicine</strong> (<em>NEJM</em>, 2016)<a href="https://www.nejm.org/doi/full/10.1056/NEJMp1606181" target="_blank" rel="ugc noopener noreferrer">https://www.nejm.org/doi/full/10.1056/NEJMp1606181</a></li><li><strong>Nightingale Open Science</strong><a href="https://www.nightingalescience.org/" target="_blank" rel="ugc noopener noreferrer">https://www.nightingalescience.org/</a></li><li><strong>Dandelion Health</strong><a href="https://dandelionhealth