Boston — Scientific journals have become a game for Artificial Intelligence (AI) in which it can quickly tell the difference between studies. However, the question remains: which studies matter and how can clinicians classify them? At a recent AI conference, Atman Health Chief Medical Officer and Associate Physician at Brigham and Women’s, Rahul Deo, summarized this issue. He presented a hierarchy of medical AI research, ranked from least impactful to the riskiest and most impactful. The AI models with the greatest impact are those that figure out how to replace the most complex physician tasks with automation, followed by research that takes steps towards that goal.
Replacing the primary work performed by doctors with machine learning could reduce healthcare costs or expand access to treatment in areas where doctors are in short supply. However, this would require strict regulation by the Food and Drug Administration. Deo believes that everything else in medicine will become very possible if AI receives reliable input data. Rapid, iterative learning of optimal treatment approaches will require vast amounts of data. However, such data is almost nonexistent. Empathy, reasoning, and decision-making are also very difficult for AI to replicate without prompting.
Clinical decision support algorithms are already gaining traction in clinical practice, with computer-assisted mammography now used in patient care. Models that point out people at higher risk often use wearables and other devices to gather digital clues about human risk. However, the direct-to-consumer nature of these technologies poses major workflow challenges that the healthcare system does not yet understand. The infrastructure to integrate this kind of data into the traditional healthcare system is also lacking.
AI has given physicians more tools to eliminate low-risk tasks, such as responding to patient portal messages. However, these features do not advance the state of AI or impact the healthcare system. The healthcare system has always defined acceptable amounts of risk, but new features raise questions about the line between high-stake and low-risk activities. The goal is to allow AI to replace the most complex physician tasks with automation.