U.K.-based startup Babylon Health announced today that its artificial intelligence has shown an ability to diagnose heath issues as well as a human doctor, and can occasionally surpass mere mortals in its skills. No word on its bedside manner, though.
To make this *ahem* diagnosis, Babylon had the AI take the final exam that all trainee General Practitioners in the U.K. have to take, as designed by the Royal College of General Practitioners (RCGP). When GPs pass the test they can start practicing. A key part of the exam tests a doctor’s ability to diagnose, so Babylon took a representative sample-set of questions testing diagnostic skills from exam prep materials and the RCGP curriculum—then fed them into the AI.
After taking the test, Babylon says its AI scored 75%, which was 5% higher than the average marks for human doctors over the last five years (70%). As the AI accumulates more knowledge, Babylon expects its scores will go up. To further test the AI’s capabilities, the company pitted it against seven highly experienced primary-care doctors using 100 independently devised symptom sets. Babylon’s AI scored 80% for accuracy, the company says, while the seven doctors achieved an accuracy range of 64-94%. The AI accuracy went up to 98% when assessing the conditions seen most frequently in primary care medicine, compared to human clinicians whose accuracy ranged from 52-99%.
While it sounds like yet another slightly unsettling job-stealing robot story, it could potentially help solve a global doctor shortage and provide better access to health advice. As reported by TechCrunch, Babylon raised $60 million in funding last year to develop its AI capabilities.
Babylon’s research paper, entitled Evaluation of AI Powered Symptom Checker, can be downloaded from the company’s website and will be available over coming days via ArXiv.com.