The potential advances that AI will bring to medicine boggle the mind and portend massive changes, the likes of which we have not experienced since the advent of modern surgical techniques, the development of image-based diagnostic methods, and the discovery of effective medications. As a primary care doctor who has worked at San Francisco General Hospital for over 30 years, and as someone who has worked through the epidemics of AIDS, diabetes, opiate addiction, and COVID, I question what the introduction of AI into the clinician-patient interaction will mean for patients and clinicians.

Recently, a new patient who had just been discharged from the hospital was waiting to be seen at my clinic. I would have 20 minutes to review his medical records, get to know him, establish myself as his primary care physician, and make appropriate management decisions. I sighed with dismay as I reviewed his outside medical records: a young male with a rare congenital syndrome and a panoply of serious acute and chronic illnesses meant he was extremely vulnerable to rapid deteriorations and in need of very high-quality care. During his childhood, the patient had congenital heart disease that affected his valves and required two heart surgeries. He had a severe form of lymphedema (chronic swelling of his legs) that led to recurrent skin infections that would spread throughout his body. Third, he had brittle bones and a deformity of his elbow that made it impossible for him to extend his arm. He had just been hospitalized for a recurrent episode of endocarditis – a life-threatening bacterial infection of a heart valve. It had led to sepsis and had required that he take a total of 4 weeks of IV antibiotics, a course of which he was finishing up at home. I questioned whether I — or any of my busy colleagues for that matter — had the bandwidth to provide the level of care needed to manage the multiple complications that his congenital syndrome had generated.

In the face of these overwhelming data, I fantasized how his case would be a great test for an AI machine that could tirelessly and reliably handle all that this young man would throw its way. I have been awe-struck by the ability of AI to augment and even replace the communicative and relational functions of doctor-patient interactions. Studies in simulated environments show how AI outperforms clinicians in such medical dialogue-related tasks as question-asking, explanation, clinical synthesis and summarization, diagnostic accuracy, and the expression of empathy and compassion. The application of AI to actual health care has already begun, ranging from crafting automated responses to patients’ questions on health systems’ electronic portals to the automated provision of psychotherapy. These communicative computational feats force us to ask ourselves a set of existential questions: While a machine might outperform a clinician on each of these individual tasks, can a machine develop a meaningful relationship with that patient? How important is the human relationship—what we in health care call the therapeutic alliance—when it comes to patient engagement, effective doctoring, caring, healing, recovery, or accompaniment through suffering and death? Can a machine understand, appreciate, and respond to a patient’s story?

When I met my new patient, I noticed his very short stature, his hair pulled into a ponytail, and his Beatles T-shirt. His ballooning yet tight sweatpants hid his lymphedema well. In response to my questions—questions he had undoubtedly fielded from hundreds of physicians over the years—he openly described his early childhood of being ill while being carted around to developing countries by his father who worked for an oil company. He summarized his recent hospitalization, providing fresh details I could not garner from the skeletal chart. What struck me most was his upbeat, positive attitude, as well as his cooperativeness and communicativeness. When I asked him how he maintained such a positive attitude in the face of so many health challenges, he answered that growing up in the developing world meant that he had seen suffering much worse than his and that he felt deeply grateful for what life offered him.

During the physical exam, he willingly offered me his chest, revealing the midline scar by his heart, enabling me to listen to the whooshing sound of his right-sided heart murmur. Without a complaint, he then struggled mightily for the few minutes it took to remove his thigh-high compression stockings and show me his deformed, elephantine legs. Putting them back on caused him even more difficulty as he panted with effort.

I asked my new patient what he enjoyed doing since his illnesses prevented him from fully functioning. He loved music and attending concerts, being social and hanging out with friends, and walking in Golden Gate Park. We bonded over these little pleasures in life and shared stories of local places we both loved to go to. This young man was more than a series of objective data points. My goal needed to be bigger than optimizing a set of measures and numbers that reflected the state of his physiology. The magic of the therapeutic alliance is how a healthy clinician-patient relationship functions as a critical lubricant to successfully navigate the mechanics of modern medicine. As his primary care physician, I was responsible for his health but also with understanding his values and even with lifting his spirits when his chronic conditions would worsen. Now, after fifteen minutes of conversation, I was inspired to try to intervene on his behalf, to keep him as healthy as I could. At the end of the visit, as we shook hands, I believed he too felt that I would care for him in this way.

It is this human magic that we must hang on to at all costs as medicine—including its most intimate moments between patient and clinician—enters the age of AI.

Dean-David Schillinger is an internal medicine physician and author of Telltale Hearts: A Public Health Doctor, His Patients, and the Power of Story.