Healthcare policy has long treated the human practitioner as the indispensable safety net for clinical software. Regulatory frameworks across major Western markets operate on a single, defensive dogma: software must assist, never replace, licensed physicians. That consensus is now facing an open revolt from influential policy architects and venture-backed operators who argue that keeping a doctor in the loop is turning into an active clinical liability—and an artificial ceiling on healthtech unit economics.
The Performance Inversion
In a landmark viewpoint published in JAMA, lead author Ezekiel Emanuel—bioethicist at the University of Pennsylvania and a key architect of the Affordable Care Act—alongside Curai Health CEO Neal Khosla, argues that autonomous systems will inevitably surpass hybrid "doctor + AI" teams at clinical reasoning. This thesis directly collides with established medical cartels. The American Medical Association (AMA) and the American College of Physicians continue to lobby aggressively to keep AI in a strictly supportive straitjacket. When medical professor Robert Wachter dismisses fully automated clinical care as the "economy class" of medicine, the JAMA authors counter that his stance is completely detached from recent empirical benchmarks.
Since 2024, standalone foundation models have matched or outperformed practicing physicians across five core clinical workflows: patient history taking, differential diagnosis, test selection, guideline adherence, and chronic disease management. In simulated clinical encounters, Google's conversational diagnostic system AMIE beat primary care physicians across nearly every conversational and diagnostic metric. In a benchmark of 377 complex cases, ChatGPT o3 identified the correct diagnosis first in 60 percent of instances, compared to a dismal 15.9 percent success rate among 20 practicing internists. Furthermore, Microsoft's diagnostic orchestrator hit accurate diagnoses under tight budget constraints roughly four times as often as human clinicians, burning fewer systemic resources along the way.
When the AI is better, the human makes the result worse by overruling the system in the wrong places.
The critical friction point is that human oversight actively degrades diagnostic accuracy once model precision eclipses human capability. A meta-analysis of 106 clinical and cognitive experiments demonstrated that human intervention improves combined outcomes only when the clinician possesses superior baseline ability; when the model is superior, human overrides consistently inject unforced errors. In one diagnostic trial, GPT-4 working alone scored 92 percent on complex reasoning, whereas physicians armed with access to that exact model slumped to 76 percent. The dynamic mirrors competitive chess: human-computer centaur teams retained an edge after Deep Blue beat Garry Kasparov in 1997, but by 2017, standalone engines began systematically crushing hybrid human-machine pairs.
Regulatory Lock-In and Operational Limits
Forcing mandatory physician sign-offs into statutory law risks institutionalizing inferior, high-cost care. Emanuel and Khosla warn that autonomous models require an immediate overhaul of medical licensing, payer reimbursement schedules, and medical malpractice liability frameworks. Human skill degradation is already accelerating the gap: as documented in The Lancet, gastroenterologists rapidly lose baseline procedural proficiency and diagnostic vigilance when leaning continuously on computer-aided detection during colonoscopies.
Yet this debate is not merely philosophical—it is an anatomy of institutional lobbying and venture economics. Coauthor Neal Khosla's father, venture capitalist Vinod Khosla, holds major equity stakes in both OpenAI and Curai Health. The financial incentive to strip expensive human payroll out of telemedicine's margin structure is immense. Eliminating mandatory clinical sign-offs instantly lowers customer acquisition hurdles and transforms telemedicine from a low-margin services business into high-margin software.
For medtech founders and healthtech investors, this tectonic shift demands an immediate product bifurcation strategy. Companies must split product roadmaps into two distinct architectures: pure enterprise workflow assistants for defensive hospital systems, and autonomous diagnostic platforms built for markets where the doctor monopoly can be circumvented. While current benchmarks still rely predominantly on single-task synthetic simulations rather than messy, longitudinal EHR environments, the regulatory clash is here. Navigating the inevitable stalemate between tech capital and medical unions will define the next decade of digital health valuation.