AI providing medical information to the public has permanently arrived, for better or worse.
The inflection point with this was reached a few years ago and this toothpaste will never be put back in the tube. This year I see around half of my naturopathic patients bringing AI diagnoses and recommendations to at least one of their appointments. As long as they are bringing in this information for my professional review I welcome it! As there have been times where a patient's ChatGPT or Claude health analysis has provided us with relevant, accurate information. However it is important to be aware of the major and potentially serious drawbacks of patients relying too heavily on AI for their healthcare.
Input Errors
Aside from objective data like lab test results, vital signs and wearable device readings the health information we input into an AI about ourselves for analysis can be highly subjective and often inaccurate. This can lead to an amplification of our own misinformation causing various downstream errors in the AI's assessment of our health.
Output Sycophancy & Hallucination
It is well established that mainstream consumer AIs default too often to telling us what we want to hear...i.e. are our sycophants! This is not a good thing when it comes to health and medicine. In addition AI "hallucinations" remain common, where the output analysis or recommendations appear very plausible but are highly inaccurate...which is especially concering when it is our health and lives at stake.
Incorrect Prioritization
For patients with complex multi-system chronic illness, the clinical prioritization of testing and treatment is crucial to their recovery...as for many people not all underlying issues can possibly addressed at once. The successfull ordering of clinical interventions is highly nuanced, often counterintuitive and requires human experience.
Not All AIs Are Equal When It Comes to Health & Medicine
The main consumer AIs do not have healthcare-specific focus or programming and therefore apply their extremely generalized learning models to individual health and medical inquiries. As a naturopathic and functional doctor I invest in a subscription to the FunctionalMind AI, which has been created by experts in my profession with a complete evidence-based focus on what we do.
Addiction Potential
With it's always-on availability and nearly limitless access to health information, using consumer AIs for individual health evaluation and recommendations has extremely high addiction potential. Several of my patients in the last few months have described losing sleep and experiencing burnout from excessive AI research on their own health and wellness. The term "rabbit hole" that a few years ago described long searches into conspiracy theories is now often being used by patients to describe their lengthy and futile AI research into their own health. And ironically these searches have often been related to symptoms and concerns that were easily evaluated and successfully treated at my clinic.
AI Can't See, Hear & Touch You
ChatGPT and Claude can't (at least not yet!) hear your voice, see the surface of your skin or touch your body...can't give you a physical exam, which is an extremely important part of medicine and healthcare. I recently had a young, otherwise healthy male patient who came in with a rash on his face, neck and chest that he was sure from his online research was fungal. Upon very close inspection I saw some signs that pointed in the direction of a much more serious infection and referred him to our local emergency department. Once there he was quickly diagnosed with a severe and rapidly spreading staph infection that required a week of daily IV antibiotics to clear.
An Unprofessional Relationship
Our individual healthcare relationship with our AIs are not healthy or professional! As I've shared above they are flawed in several major and potentially serious ways. This is not a relationship that should be relied upon when our lives are potentially at stake. AI can be a powerful tool in medicine today (especially for doctors like me using true medical-grade, evidence based AI) when used correctly. But for the average patient and AI user it remains buyer beware.
Dr. Arjuna
Dr. Arjuna Veeravagu
Contact Me
