A Harvard Medical School study published in Science has found that artificial intelligence systems outperform human doctors in emergency room triage diagnoses, raising urgent questions about oversight, liability, and the future role of human clinical judgment in high-stakes medical decisions.
The research tested AI against hundreds of physicians in scenarios where split-second decisions determine patient outcomes. In one critical test involving 76 patients arriving at a Boston emergency room, an AI system correctly identified or nearly identified diagnoses in 67% of cases, compared with 50-55% accuracy among human doctors working from the same electronic health records—vital signs, demographics, and nursing notes.
The performance gap widened dramatically in the conditions that matter most: rapid triage with minimal information. The AI's advantage was "particularly pronounced in triage circumstances requiring rapid decisions with minimal information," according to the study. When more detailed patient data became available, the AI's accuracy rose to 82%, compared with 70-79% for expert humans, though researchers noted this difference was not statistically significant.
Where the System Succeeds—and Where Accountability Fails
The AI also outperformed human doctors in developing longer-term treatment plans. When asked to examine five clinical case studies and propose comprehensive care strategies—including antibiotic regimens and end-of-life planning—the AI scored 89% compared with just 34% for doctors using conventional resources like search engines.
One striking example illustrates the potential: a patient presented with a blood clot to the lungs and worsening symptoms. Human doctors attributed the deterioration to failing anti-coagulants. The AI, however, identified a crucial detail humans missed—the patient's lupus history, which could explain lung inflammation. The AI's diagnosis proved correct.
Yet despite these capabilities, a fundamental governance gap remains unaddressed. "There is not a formal framework right now for accountability," acknowledged Dr. Adam Rodman, a lead author and physician at Boston's Beth Israel Deaconess Medical Centre, where the study took place. This absence of accountability structures is particularly concerning given the stakes: decisions made in emergency triage directly determine which patients receive immediate intervention and which do not.
Adoption Accelerating Without Safeguards
The technology is already entering clinical practice at scale. Nearly one in five US physicians are using AI to assist with diagnosis, according to research published last month. In the UK, 16% of doctors use the technology daily and another 15% weekly, with "clinical decision-making" among the most common applications, according to a Royal College of Physicians survey. Yet UK doctors identified AI error and liability risks as their biggest concerns.
Dr. Arjun Manrai, another lead author heading an AI lab at Harvard Medical School, sought to temper expectations about the study's implications. "I don't think our findings mean that AI replaces doctors," he said. "I think it does mean that we're witnessing a really profound change in technology that will reshape medicine."
Dr. Rodman proposed a "triadic care model" for the next decade: "the doctor, the patient, and an artificial intelligence system." But he emphasized a critical human need: patients ultimately "want humans to guide them through life or death decisions [and] to guide them through challenging treatment decisions."
Critical Gaps in the Evidence
Independent experts flagged significant limitations that complicate any rush toward deployment. The study tested AI only on text-based patient data—electronic records and written notes. The AI never evaluated a patient's visible distress, physical appearance, or non-verbal cues that experienced clinicians routinely use to inform triage decisions. "That meant the AI was performing more like a clinician producing a second opinion based on paperwork," the study noted.
Dr. Wei Xing, an assistant professor at the University of Sheffield's school of mathematical and physical sciences, raised a concern that extends beyond technical performance: some findings suggested doctors may unconsciously defer to the AI's answer rather than thinking independently. "This tendency could grow more significant as AI becomes more routinely used in clinical settings," he warned.
Xing also highlighted missing data critical to equitable deployment: the study does not show which patients the AI performed worse with, or whether it struggled disproportionately with elderly patients or non-English speakers—populations already facing healthcare disparities.
"It does not demonstrate that AI is safe for routine clinical use, nor that the public should turn to freely available AI tools as a substitute for medical advice," Xing cautioned.
Prof. Ewen Harrison, co-director of the University of Edinburgh's centre for medical informatics, offered a more measured assessment: the study showed these systems are "starting to look like useful second-opinion tools for clinicians, particularly when it is important to consider a wider range of possible diagnoses and avoid missing something important."
Why This Matters:
This study arrives at a moment when billions in investment capital are flowing into AI healthcare companies, yet regulatory and accountability frameworks lag far behind deployment. The technology's demonstrated capacity to improve diagnostic accuracy in emergency settings could save lives—but only if accompanied by robust governance structures that protect patients and preserve human clinical judgment. The absence of formal accountability mechanisms, combined with evidence that doctors may unconsciously defer to AI recommendations, suggests the technology is advancing faster than the democratic and institutional safeguards necessary to manage its risks equitably. Questions about performance disparities across patient populations, the role of human oversight, and liability when AI systems fail remain largely unanswered, even as adoption accelerates in clinical practice.