Health-care AI is here. We don't know if it helps patients (www.technologyreview.com)

AI is increasingly being integrated into healthcare settings, assisting doctors with tasks such as note-taking, interpreting medical results, and analyzing patient data. However, a recent paper published in Nature Medicine by Jenna Wiens from the University of Michigan and Anna Goldenberg of the University of Toronto raises critical questions about the effectiveness of these AI tools in actually improving patient health outcomes. While many hospitals have adopted AI technologies like ambient AI tools that summarize patient-doctor conversations, there is a significant lack of rigorous evaluation regarding how these tools impact clinical decision-making and, ultimately, patient care. The adoption of AI in healthcare is on the rise, with about 65% of U.S. hospitals reportedly using predictive tools. Yet, only two-thirds of these institutions have assessed their accuracy, and even fewer have explored potential biases. Wiens emphasizes the need for healthcare providers to investigate the real-world implications of AI tools, noting that while they may streamline processes and reduce clinician burnout, there’s no guarantee they enhance patient outcomes. The AI/ML community is urged to focus not only on the development of innovative tools but also on their measurable effects in clinical environments, ensuring that advancements genuinely benefit patient care.