There is a version of AI in medicine that Sejal Desai finds genuinely compelling: a tool that catches what a physician might miss, steps in earlier, and helps a patient understand their own condition in plain language. In obesity medicine, where treatment is highly individual and patients often arrive carrying years of frustration, that kind of support matters. The technology, Desai argues, has real potential to make care more personalized, more proactive, and more accessible to people who have historically been underserved by the system.
But there is another version that gives her pause. It is one where the efficiency gains come at a cost that does not show up in outcomes data, where the act of sitting with a patient and building trust gets quietly squeezed out by the same tools that were supposed to free up time for exactly that.
"AI can support clinical decision-making, but it cannot replace empathy, intuition, lived experience, or the trust built between a physician and patient."That line is not a platitude for Desai. It is the tension she believes the broader conversation about AI in medicine keeps glossing over.
The concerns she raises are practical as much as philosophical. Bias embedded in AI models, the spread of medical misinformation, and the risk of clinicians becoming overly dependent on algorithmic recommendations are all live problems, not hypothetical ones. And then there is the equity question. If these tools are rolled out without careful attention to who has access and who does not, Desai warns, they risk making existing disparities worse rather than better.
She is equally attentive to what AI could do well, specifically in translating complex medical information into language patients can actually use. Empowering patients to understand their conditions and advocate for themselves is not a secondary benefit in her view. It is central to what good care looks like. The promise of AI helping to close that communication gap is one she takes seriously.
What she keeps returning to, though, is the condition under which any of this works. The physician-patient relationship has to remain the anchor. AI, in her framing, is a powerful tool, but the word "tool" is doing real work there. Tools serve the person using them. They do not replace the judgment, the presence, or the relationship that makes medicine medicine. The question the field has not fully answered yet is whether the institutions building and deploying these systems share that priority, or whether the machine in the room eventually starts setting the terms.
