Health AI for Good Rather Than Evil? The Need for a New Regulatory Framework for AI-Based Medical Devices

Artificial intelligence (AI), especially its subset machine learning, has tremendous potential to improve health care. However, health AI also raises new regulatory challenges. In this Article, I argue that there is a need for a new regulatory framework for AI-based medical devices in the U.S. that ensures that such devices are reasonably safe and effective when placed on the market and will remain so throughout their life cycle. I advocate for U.S. Food and Drug Administration (FDA) and congressional actions. I focus on how the FDA could - with additional statutory authority - regulate AI-based medical devices. I show that the FDA incompletely regulates health AI-based products, which may jeopardize patient safety and undermine public trust. For example, the medical device definition is too narrow, and several risky health AI-based products are not subject to FDA regulation. Moreover, I show that most AI-based medical devices available on the U.S. market are 510(k)-cleared. However, the 510(k) pathway raises significant safety and effectiveness concerns. I thus propose a future regulatory framework for premarket review of medical devices, including AI-based ones. Further, I discuss two problems that are related to specific AI-based medical devices, namely opaque (“black-box”) algorithms and adaptive algorithms that can continuously learn, and I make suggestions on how to address them. Finally, I encourage the FDA to broaden its view and consider AI-based medical devices as systems, not just devices, and focus more on the environment in which they are deployed