Augmented intelligence (AI) has numerous benefits to the field of medicine. From improving study design and enhancing data collection, to improving diagnostic accuracy and reducing medical errors, to earlier disease detection and even remote monitoring, AI will be ubiquitous in medicine. However, in aspects influencing patient decisions to seek or forgo medical care, AI will be a deterrent. AI will de-individualize care, eroding trust in medical professionals, and provide online alternatives that are less beneficial, or even harmful, than receiving care directly from a trained professional. The root cause is threefold: AI cannot feel, AI cannot experience, and AI cannot evaluate comprehensively.
First, augmented intelligence reduces the human-to-human aspect of shared decision-making. While AI may identify and recommend clinically appropriate aspects of care, these will be biased to the population, not the individual patient. This is inherent to the data itself, as randomized controlled trials aggregate data from individual interventions. Percentages of positive or negative effects are generated based on the conglomeration of individual outcomes, and results pertaining to the population of individuals are concluded. Medical professionals then apply this science when practicing the art of medicine. AI cannot perform the art of medicine; AI can only suggest therapies based on their effect on a population that is medically similar to the individual patient. But human decisions are not solely based on what is likely to happen, but also on what could happen. AI can’t weigh factors such as the patient’s desires, fears, or lifestyle to a meaningful degree. Decisions in medicine, which can impact both quantity of life and quality of life are not exercises in data aggregation and algorithmic analysis. This is because quality can only be defined by the individual living it, and only a human can attempt understanding and empathy. Applied to a dataset or population, frequency may be the most important measure of outcome. But each person in that population has a different definition of quality.
Furthermore, before even weighing individual patient factors, medical providers and AI can come to different recommendations. AI algorithmically synthesizes data and draws conclusions from it. Humans makes conclusions based on information (not necessarily data), experience, and intuition. Take for example the scenario of a patient who has suffered a myocardial infarction and now has a reduced left ventricular ejection fraction. Based on randomized data, the augmented intelligence algorithm could suggest that 1) a wearable cardiac defibrillator (WCD) should be prescribed to reduce death from any cause, or 2) a WCD should not be prescribed because it will have no effect on incidence of sudden cardiac death, or 3) a WCD should be prescribed if the patient agrees to actually wear it, to lower the chance of dying due to a ventricular arrhythmia. The way the algorithm interprets data can vary just as much as physician interpretation. However, and especially in cases like this, human experience matters. I have cared for patients who survived out-of-hospital cardiac arrest solely due to a WCD defibrillation. My practice is influenced by these experiences; AI decision-making is not.
Lastly, AI will enable patients to bypass professionals in diagnoses, and maybe someday, treatment. As easy as it is to self-diagnose online now, (often incorrectly, Google headache symptoms and you may be ‘diagnosed’ with brain cancer), AI will only worsen this. Plugging symptoms or objective data into a publicly accessible online AI generates diagnoses made without appropriate oversight and may lead to delayed or incorrect medical care if patients act on this information. In medical school I vividly remember an anecdote from a visiting alumnus. She detailed the challenges she faced in caring for a community in rural Tanzania, noting specifically that the three largest detractors were lack of nutritious food, being the only physician for 100 miles, and visiting aid groups. These visiting aid groups would often travel through the area and distribute donations. One group specifically distributed packets of vitamins, since local diets were deficient in some vital nutrients. A young child came to her practice critically ill with malaria. Unfortunately, the parents thought that the aid they received from the visiting volunteers, a packet of vitamins, was all that could be done, so did not seek further care. Besides asking some basic questions, the aid group did not medically evaluate the patient. The child received care, just not the best care, and died because of it. While this example is an extreme, the implications to modern medicine are analogous. We as medical professionals must ensure that the ability to find quick answers does not detract from seeking appropriate care.
The term “Augmented Intelligence,” used by the American Medical Association to differentiate from artificial intelligence, is meant to imply that this technology will support rather than replace medical personnel. However, medical professions don’t decide if our expertise is replaced; patients do. We must ensure that patients continue to receive individualized evaluation and treatment, otherwise, AI will deter medical care.