A Harvard study has revealed that artificial intelligence systems surpassed human doctors in emergency medicine triage diagnoses, identifying conditions more accurately in initial hospital assessments, while billions are simultaneously being invested in AI healthcare companies. The findings, published in the journal Science on the same day, indicate that large language models have “eclipsed most benchmarks of clinical reasoning,” according to the study’s authors.
In an experiment involving 76 patients at a Boston hospital emergency room, an AI system and two human doctors were provided identical electronic health records, including vital signs, demographic data, and a nurse’s summary. The AI identified the exact or a very close diagnosis in 67% of cases, outperforming human doctors who achieved 50%-55% accuracy. This advantage for the AI, specifically OpenAI’s o1 reasoning model, was particularly evident in rapid decision-making scenarios with minimal information. When more detailed information was available, the AI’s accuracy rose to 82%, compared to 70-79% for expert human clinicians, though this difference was not statistically significant.
The study further demonstrated the AI’s superior capability in formulating longer-term treatment plans, such as antibiotic regimens or end-of-life processes. When tasked with examining five clinical case studies alongside 46 doctors, the computer developed significantly better plans, scoring 89% compared to 34% for humans utilizing conventional resources like search engines.
Capital's New Frontier
The deployment of AI in healthcare represents a new frontier for capital accumulation, with “billions being invested in AI healthcare companies.” This influx of capital underscores the drive to integrate advanced technology into medical practice, promising a “profound change in technology that will reshape medicine,” as stated by Arjun Manrai, a lead author from Harvard Medical School. This reshaping is not merely technological but economic, aimed at optimizing processes and potentially reducing labor costs in a sector ripe for surplus extraction.
Already, nearly one in five US physicians are using AI to assist diagnosis, according to research published last month. In the UK, 16% of doctors employ the technology daily, with an additional 15% using it weekly, primarily for “clinical decision-making,” as reported by a recent Royal College of Physicians survey. This rapid adoption indicates the swift penetration of capital-intensive technologies into the medical profession.
The Cost to Labor
While capital celebrates these advancements, the implications for medical labor are significant. The study’s findings suggest a potential deskilling of the medical profession, as AI systems demonstrate superior diagnostic and planning capabilities. Dr. Wei Xing, an assistant professor at the University of Sheffield, highlighted that doctors might “unconsciously defer to the AI’s answer rather than thinking independently,” a tendency that “could grow more significant as AI becomes more routinely used in clinical settings.” This deference erodes the autonomy and critical reasoning skills historically central to medical practice.
Concerns among UK doctors primarily revolve around AI error and liability risks, issues for which “there is not a formal framework right now for accountability,” according to Dr. Adam Rodman, another lead author. This absence of a clear accountability framework effectively shields corporations developing and deploying these AI systems from full responsibility, shifting potential burdens onto individual practitioners or patients, and thus protecting accumulated wealth.
The study’s limitations also reveal potential class and social biases in the technology’s application. The AI was only tested on patient data communicable via text, excluding visual cues or a patient’s level of distress. Dr. Xing noted the lack of information regarding which patients the AI performed worse at diagnosing, specifically mentioning elderly patients or non-English speakers. This omission raises questions about the technology’s efficacy and safety for vulnerable populations, who often face systemic disadvantages within the healthcare system.
Despite the promise of a “triadic care model” involving doctor, patient, and AI, as proposed by Dr. Rodman, the underlying drive for capital efficiency suggests a future where human labor may be increasingly marginalized or restructured to serve the technological apparatus, rather than the other way around. The “most impactful technologies in decades” are those that serve the interests of capital, and the current trajectory of AI in healthcare is no exception.