If you want to start, or continue using, a chatbot for your health questions, take these expert-recommended steps as you come up with prompts:
1. Test the model with misinformation or inaccuracies first.
[Dr. Girish N. Nadkarni], an AI health researcher and director of Mt. Sinai’s Hasso Plattner Institute for Digital Health, says it’s important to ask the chatbot about medical misinformation or known falsehoods prior to querying it about specific health questions.
…
2. Consider the cues or information you may be giving the chatbot.
When Nadkarni and his colleagues tested ChatGPT Health earlier this year, they discovered that how users frame their symptoms may influence the model’s accuracy.
If, for example, the prompt included statements about friends or family downplaying the symptoms in question, ChatGPT Health’s recommendation shifted in that direction as well.
3. Take into account whether you’re a novice or expert.
Share this article
…
[Dr. Robert Wachter, professor and chair of the Department of Medicine at the University of California, San Francisco], says that recent research demonstrates a clear risk for patients when they don’t know the right information to use in a prompt, and when they misinterpret the chatbot’s response.
…
4. Ask for references and cross-check the answer.
