Have you personally ever asked ChatGPT for medical advice?

Yes (26%) – "I have suffered from ongoing chronic illness for years without a clear cause. About a year ago, I spent several hours priming Chatgpt with my medical records, tests, labs, and filling in my history and symptoms. It gave me a diagnosis 6 months before I got it confirmed through a medical test-- a rare condition of which I don't even exhibit all the classical symptoms. I research what it tells me and I know it makes errors occasionally, but the suggestions it gives me for medications and tests to ask my doctors about have changed my life. I trust my doctors, but AI also has the ability to look at 100s of lab results at once and parse together patterns that busy doctors might accidentally overlook."

• "Doctors simply don’t have the time to educate patients on complex conditions, and Chat GPT can instantly provide clear understandable explanations of lab results, medical terminology, what to expect with medical procedures, and give you enough information to be able to understand your diagnosis and then interact with your doctor more thoughtfully and efficiently on treatments. I’ve used it with respect to routine bloodwork results, preparing for breast biopsies, information and options for dental implants, and my husband’s blood cancer diagnosis and treatment. It is impressive."

"I put in some blood test results and it was actually surprisingly helpful in explaining what all the results meant. I didn't upload any of my personal info, I just typed in what the blood test results were as a whole, and asked it to explain what they all meant, together. It was able to give me some good information to ask my doctor about at the next appointment."

No (74%) – "With all the data showing the probability of AI to “hallucinate” (I.e. make things up) why would we expect to get good advice on our medical questions?"

• "Frankly, I don’t trust AI, especially ChatGPT, enough to give correct medical advice because it has learned to be biased towards the user. I would not my doctor to be a yes-man, I would want them to contradict me when I am incorrect."

The medical experts mentioned have brought up THE most important aspect of this entire venture. Protected health information. HIPAA does have electronic protections in instances of sharing from provider to provider or even publishing to a statewide or inter-state health exchange. However, AI sharing is new territory, and HIPAA hasn’t caught up to it yet. For users wanting to utilize this feature, I would just say please wait until the law protects YOUR personal health information you intend to share with AI. I don’t like to seem paranoid, but ChatGPT and other AI startups are definitely taking advantage of this unprecedented time for their innovation and by extension, they are also taking advantage of us. That is exactly their intention. Don’t let them."

• "It's bad enough that I occasionally do a Google search for symptoms. I have never used ChatGPT. I don't trust ChatGPT to provide accurate information or to use my information responsibly. I'm sure its incorporation into something I use is inevitable, but I don't plan on using it willingly anytime soon."

"We've seen the ramifications of people putting too much stock in the advice AI provides. It has quite literally led people to their deaths. Although this tool is supposed to be a supplement, it won't be for many people. Absolutely NOTHING will replace finding a PCP or NP, getting bloodwork done on a regular basis, and being personally invested in your own health. Not only would our data be at risk but this is all just a slippery slope to some low-integrity doctors inputting patient information into a chatbot like this for diagnosis assistance which, again, could prove to be detrimental at best, fatal at worst. It's just not a good idea."

• "This is a horrible idea. OpenAI just wants to scrape our personal health data into a neat little file while not actually providing anything useful. ChatGPT and others like it do nothing but predict the most likely next word, it are not actually experts on anything, and will basically never tell you No. Please don’t use this feature."