{"slug": "two-ais-just-matched-or-beat-doctors-on-diagnosis-the-catch-none-of-the-patients", "title": "Two AIs just matched or beat doctors on diagnosis. The catch: none of the patients were real.", "summary": "Two AI systems, Mira and Amie, matched or outperformed doctors in diagnosing and treating simulated patients, with Mira achieving 87% accuracy versus doctors' 78% and Amie producing more guideline-adherent treatment plans. However, the tests used clean, text-only case notes without physical exams or real patient interactions, and experts note that Amie's advantage was partly due to following guidelines strictly while doctors are not bound to do so. The researchers emphasize the AI is an assistive tool, not a replacement, akin to an aircraft autopilot.", "body_md": "Two AI systems have matched, and in places beaten, doctors at diagnosing patients and planning their treatment. Then again, none of the patients were real.\n\nThe results, published in Nature this week, are some of the strongest evidence yet that specialist medical AI is closing in on human clinicians. They are also a textbook case of why a striking headline and real-world medicine are not the same thing.\n\n## What the studies found\n\nThe first system, Mira, was built by academic researchers in Germany.\n\nGiven access to a simulated medical record, it can choose from more than 85,000 actions: tests, prescriptions, even hospital admissions. Across more than 500 emergency-department cases, it reached a diagnostic accuracy of about 87 per cent, against 78 per cent for a panel of six doctors. It was strongest on conditions with clear test results, such as pancreatitis and appendicitis.\n\nThe second, Amie, is Google’s, and runs on its Gemini model.\n\nTested against 21 UK GPs across 100 multi-visit cases, it matched them on clinical reasoning and produced treatment plans that stuck more closely to official guidelines. On a benchmark for tricky medication decisions, it came out ahead.\n\n## Why the headline oversells it\n\nRead the fine print and the picture softens. Both systems were tested on simulated patients, fed clean, text-only case notes. There were no physical examinations, no scans, no reading a patient’s tone or body language, all things real doctors rely on.\n\nIndependent experts flagged more.\n\nAmie was rewarded for following guidelines, which doctors are not strictly bound to do, making the comparison lopsided. Mira ordered roughly twice as many tests as the doctors, and ordering more tests can flatter an accuracy score.\n\nAnd the models are already old: the versions tested are around two years out of date, which, the researchers note, arguably makes them weaker than what exists now, not stronger.\n\n## Autopilot, not replacement\n\nThe researchers are careful about what this means. Mira’s co-developer, Jakob Kather, compared the AI to an aircraft’s autopilot: it can take over routine work, but “ultimate responsibility will always remain with the physicians”.\n\nThat is the likely future, and it is already arriving.\n\nAI is being folded into [real health systems](https://thenextweb.com/news/nhs-england-microsoft-copilot-505000-staff-ai-rollout) to ease workforce shortages, used to cut [administrative load](https://thenextweb.com/news/frontier-health-16m-seed-atomico-nhs-admin-ai-juno), and pushed at patients as [consumer health advisers](https://thenextweb.com/news/microsoft-launches-copilot-health). The Nature studies do not show that doctors are obsolete. They show that, in a simulator, a machine can now reason like one, which is both genuinely impressive and a long way from a real hospital.\n\n## Get the TNW newsletter\n\nGet the most important tech news in your inbox each week.", "url": "https://wpnews.pro/news/two-ais-just-matched-or-beat-doctors-on-diagnosis-the-catch-none-of-the-patients", "canonical_source": "https://thenextweb.com/news/ai-mira-amie-matches-beats-doctors-nature-study", "published_at": "2026-06-19 12:02:19+00:00", "updated_at": "2026-06-19 13:12:23.054432+00:00", "lang": "en", "topics": ["artificial-intelligence", "large-language-models", "ai-research", "ai-safety", "ai-products"], "entities": ["Mira", "Amie", "Google", "Gemini", "Nature", "Jakob Kather", "TNW"], "alternates": {"html": "https://wpnews.pro/news/two-ais-just-matched-or-beat-doctors-on-diagnosis-the-catch-none-of-the-patients", "markdown": "https://wpnews.pro/news/two-ais-just-matched-or-beat-doctors-on-diagnosis-the-catch-none-of-the-patients.md", "text": "https://wpnews.pro/news/two-ais-just-matched-or-beat-doctors-on-diagnosis-the-catch-none-of-the-patients.txt", "jsonld": "https://wpnews.pro/news/two-ais-just-matched-or-beat-doctors-on-diagnosis-the-catch-none-of-the-patients.jsonld"}}