Robots, radiologists, and results

In 2017 a robot called Xiaoyi—Chinese for little doctor—attempted China’s medical licensing examination.1 In its first practice examination, the robot scored 100 out of 600 points; after studying 400 000 articles and millions of patient records, it scored 456, substantially higher than the passing score of 360. Notably, although Xiaoyi excelled at questions based on memorisation, it did not do as well dealing with patient cases.2 The company that developed the robot is clear that it is not meant to replace doctors,3 but other people have been less restrained about the effect of artificial intelligence systems on human physicians, particularly in areas requiring speed, accuracy, and radiographs.4 As technological advances capture ever higher resolution images, and more such images are produced, the advantages of machine over human analysis are increasingly lauded.

Consequently, radiology—a specialty that uses imaging technology for diagnostics and interventions—has long been predicted to become extinct as a result of artificial intelligence. Geoffrey Hinton, a computer scientist sometimes called the father of deep learning, stated bluntly in 2016 that: “We should stop training radiologists now.”5 After Hinton’s …