What I'm Working On

Most recently: The Human Values Project has become a major focus. With over 1,000 clinicians enrolled from three continents, we are systematically studying how AI models make value-laden clinical decisions and how those decisions compare with those of human clinicians and patients. This work led to the RAISE symposium consensus statement proposing a "Values in the Model" (VIM) transparency framework for clinical AI, published in NEJM AI. Alongside this, we have introduced MedLog, a universal logging protocol for clinical AI—think syslog for medicine—now being piloted at four sites across three continents. I continue to work with the Coordinating Center of the Undiagnosed Disease Network where we are ramping up efficiency through the use of AI, including providing self-service solutions for patients. Also, the cases of several of the patients that we were not previously able to diagnose have yielded to new genomic insights, also powered by machine learning (e.g. splicing variants in compound heterozygous cases). I am also quite busy plowing through submissions to NEJM AI and discussing the more interesting ones with the rest of the editors. Finally, I am revisiting a project I undertook in the 1990's, how to diagnose kids from their growth patterns—but now with LLMs and far more pediatric data.