A JAMA opinion piece by prominent bioethicist Ezekiel Emanuel and AI telemedicine CEO Neal Khosla argues that autonomous AI has already matched or surpassed doctors on core medical reasoning tasks and predicts it will outperform any doctor-AI team by 2030.
The authors warn regulators against mandating human sign-off on AI decisions, saying such rules could lock in outdated care standards once AI becomes clearly superior.
However, the evidence remains largely simulation-based, and physical procedures and AI failure modes still pose real risks.
What happened
A JAMA opinion piece by bioethicist Ezekiel Emanuel and Curai Health CEO Neal Khosla argues that autonomous AI already matches or beats physicians on core medical reasoning tasks—patient history, diagnosis, test selection, guideline-based treatment, and chronic disease management—and predicts AI will outperform any doctor-AI team by 2030 for some workflows.
Why it matters
The piece directly challenges the American Medical Association and American College of Physicians, which insist AI should support doctors, never replace them. If regulators lock in requirements for human final approval (as some propose), they may enshrine care standards that will soon lag behind AI-only performance—forcing a choice between compliance and optimal care.
What to watch
The authors acknowledge major gaps: evidence relies mostly on simulations of single tasks, not real patient care; physical procedures (surgery, childbirth, colonoscopies) remain with humans; and AI systems fail differently than doctors do (hallucinations, outages, cyberattacks). Liability, payment rules, and medical training will need rethinking if autonomous AI moves forward.
Ask the AI about this article →
The opinion piece rests on two pillars: empirical evidence and a forecast about the future. On evidence, the authors cite studies from 2024 onward showing AI matching or exceeding physicians on five core reasoning tasks (patient history, diagnosis, test selection, guideline-based treatment, chronic disease management). They dismiss opposing studies as outdated or methodologically flawed. The second argument—and the piece's core claim—is a trajectory argument: AI models improve rapidly while physician skills may atrophy through disuse (the authors cite a Lancet study on colonoscopies as evidence). Once AI is clearly superior, a human reviewer becomes a liability rather than a safeguard; a meta-analysis of 106 experiments backs this claim, showing that when AI outperforms the human, the human's intervention worsens outcomes. The authors use chess as a historical precedent: after Deep Blue defeated Kasparov in 1997, human-machine teams dominated until pure AI beat the teams by 2017.
The piece's real stakes lie in regulation. If policymakers lock in rules requiring physician sign-off on AI decisions, those rules would cement care standards that may soon be demonstrably inferior. The authors argue liability, payment, and medical training frameworks all need rethinking now. However, the authors themselves acknowledge critical limitations: almost all evidence comes from simulations of single tasks, not real patient care; the handoff of information between human and model is a weak point; physical procedures remain beyond current robotic capability; and autonomous systems fail in ways humans do not—through hallucinations, internet outages, and cyberattacks. These risks must be weighed against the higher accuracy claims.
AI-summarized, only the topics you pick — one digest a day via Email, Slack, or Discord.
Free · takes 30 seconds · unsubscribe anytime
Ask AI anything about this article. Q&As are published on this page for other readers too.
