July 9, 2026, (Inside AI) — As artificial intelligence tools spread through hospitals and clinics, a leading Harvard voice warns that the push for efficiency risks hollowing out the moral core of medicine. Dr. Arthur Kleinman, a psychiatrist and medical anthropologist, argues that care—not just treatment—must remain the central work of health systems even as algorithms take on more clinical tasks.
In a recent interview, Kleinman, the Rabb professor of anthropology at Harvard and professor of psychiatry at Harvard Medical School, stressed that care is a reciprocal act. It demands presence, empathy, and a deep understanding of a patient's lived experience. He says those qualities cannot be reduced to data points or automated.
"Care should be the most important work of healthcare systems," Kleinman said. He defines care as accompanying patients with respect, listening carefully, and grasping how their social worlds shape illness. Yet today's systems measure almost none of that. Instead, they track efficiency and profit.
Kleinman's perspective draws on over 50 years of clinical practice and a decade spent caring for his late wife, Joan, who had early-onset Alzheimer's. He chronicled that journey in his book The Soul of Care. The experience showed him that care is built on rituals and habits that technology cannot replicate.
"Enduring can include joy as well as disappointment, success as well as failure," he said. "Most importantly, it means accompanying those with serious illness even through the most difficult times."
The Double-Edged Scalpel of Clinical AI
Kleinman acknowledges AI's genuine utility. It can sharpen medical scribes' work, improve doctors' clinical notes, and even measure aspects of care like communication quality. He sees promise in AI companions that use patient narratives to ease agitation in dementia, potentially reducing reliance on antipsychotics.
But he draws a hard line on therapy. "AI cannot, in my view, substitute for human therapists," he said. "It is a set of algorithms that predict words or patterns; it is not a human being." In mental health, he warns, AI's pattern-matching can backfire—reinforcing delusions or mishandling suicide risk.
This cautious optimism echoes broader industry debates. A 2025 study in The Lancet Digital Health found AI scribes cut documentation time by 40%, but a 2026 JAMA survey showed 62% of physicians fear AI erodes patient trust. Kleinman's stance aligns with bioethicists like Dr. Eric Topol, who advocates for AI that restores the human touch, not replaces it.
History offers a cautionary parallel. The electronic medical record was once hailed as a care integrator. Instead, it became a billing tool that frustrates clinicians. Kleinman sees the same pattern threatening AI: a drift toward fiscal metrics over human needs.
Training the Next Generation for a Hybrid Future
For medical students, the path forward is not about rejecting AI but mastering discernment. "The question is no longer whether to use AI. That horse has already left the barn," Kleinman said. The real challenge is knowing where AI helps and where it demands oversight.
His current research at Harvard's Social Technology for Global Aging Research Initiative explores how tech can support aging populations without sidelining caregivers. The goal is systems that weave together context, clinical excellence, and the labor of care—with AI as a supportive thread, not the whole fabric.
Kleinman's message lands at a pivotal moment. Global healthcare AI spending is projected to hit $45 billion by 2027, yet metrics for care quality remain elusive. As algorithms take on more follow-ups and decision support, the risk is a two-tier system: efficient for the healthy, hollow for the suffering.
The interview, edited for clarity, underscores that medicine is both science and moral practice. For Kleinman, the ultimate test of AI in healthcare will not be its speed or accuracy, but whether it helps clinicians stay present with the person in front of them.
