A Washington Post test had ChatGPT Health analyse ten years of Apple Watch data, including 29 million steps and 6 million heart-rate readings and asked the tool to grade the columnist’s heart health. The result varied dramatically between an F, a C, and a B when the same data was presented to the AI again and again, and his own doctor judged the assessments to be incorrect, inconsistent, and misleading. Expert Dr. Eric Topol called the analysis “baseless” and said the tool isn’t ready to give medical advice.
This is a useful reminder that generative AI in health management can be inspiring, but it isn’t reliable enough to function as a standalone decision support for patient assessments, especially since it may misinterpret complex medical data or produce seemingly definitive answers where none exist. We continually meet clients who have received incorrect “advice” or diagnoses from AI.
At Executive Health, we use AI support in several areas, primarily in MRI analysis, especially to improve MRI scans, but also to augment physicians’ ability to interpret images and detect subtle findings. But this is precisely that: suportit does not replace clinical judgment, experience, and the holistic discernment our specialists bring every day.
We believe in responsible implementation, where technology’s strengths are used together with human expertise to deliver truly safe and meaningful patient outcomes.
.png)
.png)
.png)
