Abstract Artificial intelligence (AI) has recently surpassed human performance in several domains, and there is great hope that in healthcare, AI may allow for better prevention, detection, diagnosis, and treatment of disease. While many fear that AI will disrupt jobs and the physician–patient relationship, we believe that AI can eliminate many repetitive tasks to clear the way for human-to-human bonding and the application of emotional intelligence and judgment. We review several recent studies of AI applications in healthcare that provide a view of a future where healthcare delivery is a more unified, human experience.
Introduction Artificial intelligence (AI) powers the digital age. While this reality has become more tangible in recent years through consumer technology, such as Amazon’s Alexa or Apple’s Siri, the applications of AI software are already widespread, ranging from credit card fraud detection at VISA to payload scheduling operations at NASA to insider trading surveillance on the NASDAQ. Broadly defined as the imitation of human cognition by a machine, recent interest in AI has been driven by advances in machine learning, in which computer algorithms learn from data without human direction.1 Most sophisticated processes that involve some form of prediction generated from a large data set use this type of AI, including image recognition, web-search, speech-to-text language processing, and e-commerce product recommendations.2 AI is increasingly incorporated into devices that consumers keep with them at all times, such as smartphones, and powers consumer technologies on the horizon, such as self-driving cars. And there is anticipation that these advances will continue to accelerate: a recent survey of leading AI researchers predicted that, within the next 10 years, AI will outperform humans in transcribing speech, translating languages, and driving a truck.3 Despite a flurry of recent discussion about the role and meaning of AI in medicine, in 2017 nearly 100% of U.S. healthcare will be delivered with 0% AI involvement. In healthcare, there is great hope that AI may enable better disease surveillance, facilitate early detection, allow for improved diagnosis, uncover novel treatments, and create an era of truly personalized medicine. There is also profound fear on the part of some that it will overtake jobs and disrupt the physician–patient relationship, e.g., AI researchers predict that AI-powered technologies will outperform humans at surgery by 2053.3 The wealth of data now available in the form of clinical and pathological images, continuous biometric data, and internet of things (IoT) devices are ideally suited to power the deep learning computer algorithms that lead to AI-generated analysis and predictions. Consequently, there has been a substantial increase in AI research in medicine in recent years. We believe, based on several recent early-stage studies, that AI can obviate repetitive tasks to clear the way for human-to-human bonding and the application of emotional intelligence and judgment in healthcare. Physician time is increasingly limited as the number of items to discuss per clinical visit has vastly outpaced the time allotted per visit,4 as well as due to the increased time burden of documentation and inefficient technology.5 Given the time limitations of a physician’s, as the time demands for rote tasks increase, the time for physicians to apply truly human skills decreases. By embracing AI, we believe that humans in healthcare can increase time spent on uniquely human skills: building relationships, exercising empathy, and using human judgment to guide and advise.
Black box warning AI has already exceeded human performance in visual tasks,6 large-scale image recognition,7 and strategy games8 due to rapid advances in the field of deep learning.9 Previously, machine-driven predictions relied on algorithms designed to extract specific features provided by a human expert. For example, the designer of a melanoma detection program might input rules that detect asymmetry and border irregularity. First-gen machines were limited in their accuracy by relying only on rules that could be programmed, and unless new rules were specifically added, the machines were unable to adapt. Now, the advent of deep learning algorithms allows for machines to receive data and self-develop complex functions to provide predictions. Using the melanoma example, now the designer of our melanoma detection program simply feeds the computer labeled images of confirmed melanomas and non-melanomas, and the computer creates its own internal rules to differentiate malignant from benign. And as the machine collects more data, it can continue to improve its predictions. Current AI therefore creates an uncomfortable situation for physicians and patients: we cannot tell which features the machine uses to generate its predictions. Without a thorough understanding of how AI is working, it may be difficult to assuage the fear of “runway machines” that has been stoked by movies like The Matrix, The Terminator, and 2001: A Space Odyssey. There has accordingly been significant discussion on the ethical implementation of AI.10 Regardless, it will quickly become clear that AI can equal or outperform humans at simple, repetitive tasks. And the simpler or more helpful the task—for example an AI system that can largely automate electronic medical record documentation—the easier it will be to allow these technologies into the clinic. Physicians by and large don’t enjoy repetitive, rote tasks—they enjoy the application of reason and judgment to complex problems in order to help patients. Rather than take over, we believe that these systems may take on much of the unpleasant work of healthcare.