Artificial Intelligence Is Becoming Natural

You don’t have to sit in a self-driving Tesla to feel the impact of artificial intelligence (AI) on your daily life. From voice-powered personal assistants like Alexa or Siri to help you track and organize information to tailored online shopping, AI is no longer in the realm of science fiction. Machine-learning platforms for clinical purposes are also making the headlines. Early last year, Stanford-based scientists harnessed aGoogle algorithm to classify skin cancers as accurately as board-certified dermatologists (Esteva et al., 2017). This algorithm distinguishes harmless from potentially fatal moles at an early stage, which is critical, given that melanoma is one of the deadliest cancers and its global incidence is on the rise. 2018 itself has already seen significant AI advances. These bring many unprecedented opportunities—and daunting challenges. ‘‘The eyes are a window to the heart’’—we’ve all heard it before. While popular sayings are not meant to be taken literally, recent research suggests theremay be some truth to this one. In collaboration with the Stanford School of Medicine, Google and its sister company, Verily Life Sciences, recently reported a deep-learning model that can recognize elevated cardiovascular disease risk from photographs of the retinal fundus (Poplin et al., 2018). Around the same time, a team of scientists from the University of California, San Diego, and Guangzhou University described an AI platform for the screening and diagnosis of common causes of severe vision loss at a stage where the diseases are still treatable. Further, the authors demonstrated the general applicability of their machine-learning system by showing its potential for diagnosing pediatric pneumonia using chest X-rays (Kermany et al., 2018). Last month, a paper published in Nature Digital Medicine reported that computer vision can also be leveraged to interpret echocardiograms and does so at accuracies that exceed those of trained experts (Madani et al., 2018). While these developments nicely illustrate the potential for AI in imaged-based medical diagnosis, they are not completely unanticipated. It is well accepted that machines can be fed large amounts of data and be trained to recognize patterns much better than humans. What is surprising is the speed with which such potential is now being unleashed.