As electronic medical records begin incorporating AI, Manisha J. Loss, associate chief medical information officer for Johns Hopkins Medicine, envisions a health care future with fewer administrative hassles, better patient experiences and more time for patient-centered care.

Already, pilot programs are underway at Johns Hopkins to use AI to draft notes based on clinical conversations, generate responses to patient queries in MyChart and categorize MyChart messages for triage based on clinical versus administrative concerns from patients.

But while AI technology can write notes, letters and MyChart responses, it doesn’t discern if the information is accurate or if its message would devastate a recipient. “We need to make sure we have transparency in understanding how these models generate their responses,” Loss says. “And even when we have that transparency, this is clinical care — we still need a clinician to lay eyes on it to make sure it’s appropriate.”

Yet in its own way, AI can be remarkably empathetic, able to explain difficult health news without the all-too-human burdens of emotion, exhaustion or distraction. “The patient gets a [written] response that’s thorough and kind and grammatically correct,” says Loss. It’s then up to clinicians to review those computer-generated words and use them as a starting point for their conversation. 

Computer scientist Mark Dredze and colleagues made headlines last spring with a study, published in JAMA Internal Medicine, that looked at whether ChatGPT can respond accurately to the types of questions patients send to their doctors. Comparing written responses from human physicians and those from ChatGPT, a panel of licensed health care professionals preferred ChatGPT’s answers 79% of the time. The panel perceived ChatGPT’s responses to be more empathetic.

While physicians are under punishing time pressure, notes Dredze, “Sometimes all that is needed is to stop and say, ‘I’m sorry you’re going through this.’ It’s such a simple thing, and computers are easily able to remember to do that — especially in writing. Computers have all the time in the world to make empathetic statements.”