Energy Considerations for Scaling Artificial Intelligence Adoption in Medicine: First Do No Harm

AbstractAI is transforming health care delivery, but its environmental impact remains underexamined. In the age of rapidly scaling AI applications across medicine, we risk falling into a reactive rather than proactive cycle as it relates to carbon emissions in health care. This perspective outlines the energy and resource demands associated with AI development and deployment, including training and inference phases of implementation, along with larger life-cycle considerations such as data-center cooling and precious-mineral procurement. Cost-emission analysis is discussed as a critical method to translate AI impact into standardized metrics that can facilitate head-to-head comparison while enhancing interpretability for policy makers and hospital-based decision-makers. Throughout the article, we also emphasize several actionable modifications for modeling, data handling, and computing that can reduce emissions and maximize energy efficiency while preserving clinical utility. We argue that the most critical steps moving forward include defining a measurement framework and requiring transparent reporting practices for developers and health care organizations alike. Together, these deliberate steps toward computational efficiency and sustainability can collectively harness the potential of AI to improve care quality while ensuring environmental responsibility. With aligned policy and practice, the promise of AI can be realized without compromising ecological longevity.