My take: Any research saying that ChatGPT provides more empathetic responses to vulnerable patients is irrelevant as long as the company can not guarantee that a response from ChatGPT won’t encourage self-harm, be racist or in others ways provide advice that hurts a patient.
As it turns out, no such guarantees can be given.
Also, in the study everyone is quoting, the study evaluators were also co-authors. Hmm...
Do read Jenniffer Lycette's reflection on this study, bringing everyone back to reality. Lycette is a rural community hematologist/oncologist and writer in Oregon. Her first novel is “The Algorithm Will See You Now".
https://www.statnews.com/2023/05/04/medical-chatbots-empathy-research-human-doctors/