Deepfakes, bias and the loss of human contact: what keeps healthcare awake at night about AI

By Zendoric · September 8, 2026.

The source material is modest and it is worth saying so: this is a highlights episode of KFF's podcast 'The Business of Health with Chip Kahn', published by the American Enterprise Institute (AEI), in which the host —Charles N. Kahn III, an AEI Visiting Senior Fellow— closes every installment by asking his guests the same question: what keeps you up at night? This episode gathers several of those answers focused on artificial intelligence and health. As AEI itself notes, the themes that emerge are deepfakes and disinformation, bias, and the risk that technology erodes the human connection that underpins the act of caring.

The podcast is part of a broader series that AEI links alongside this episode: recent installments have brought in Michelle Mello, professor of law and medicine at Stanford and co-director of its lab on the ethical evaluation of healthcare AI, to discuss the guardrails AI in health should have; Caroline Pearson, executive director of the Peterson Health Technology Institute, on who really benefits when AI moves into healthcare's administrative back office; and Seema Verma, former administrator of the Centers for Medicare and Medicaid Services (CMS) and now an executive at Oracle Health, on the reshaping of electronic health records. The episode at hand works as a summary of that thread, not as a new piece of reporting.

That these concerns come from people who know the U.S. healthcare system from the inside —academics, sector managers and former Medicare officials, judging by the guests the series itself lists— matters more than the content. They are not alarmed activists on the outside: they are the people who will have to implement AI tomorrow morning in a real hospital, with real patients and real legal liability. That their list of fears includes something as untechnical as 'losing human contact' is revealing: the problem is not just whether the algorithm gets the diagnosis right, it is whether the system still treats the person as a person.

This fits with what we have been flagging at Zendoric about AI deployment in healthcare: the short term is one of real friction. Algorithmic bias in models trained on unequal historical data can perpetuate existing disparities in access to care; deepfakes and AI-generated disinformation threaten to contaminate the doctor-patient relationship of trust before that relationship even begins; and administrative automation, even as it frees up time for doctors and nurses, can end up serving the bottom line of whoever implements it more than the patient —the question of who really benefits is precisely the one raised in the series' installment with Pearson.

Our reading is that these internal voices are doing necessary watchdog work while the technology matures. The underlying promise —more accurate, earlier diagnoses, personalized medicine, relief from the administrative burden that is burning out doctors and nurses today— still stands, and in the long term it is what could bring us closer to a healthcare system that catches diseases before they turn serious and frees up human resources for the kind of care only a person can provide. But that promise is only fulfilled if the sector does not lose sight, along the way, of the very things that keep those running it up at night: bias, disinformation and the risk of dehumanizing precisely what the technology was supposed to relieve.