The case for near-term artificial intelligence risks to be considered a public health problem.

Abstract

AI is being rapidly integrated across nearly all sectors, including within healthcare and health systems. The prevailing dominant narrative frames AI's impact on health as overwhelmingly positive and destined to improve further. However, this benefit-focused discourse overlooks a growing set of serious risks to human health arising from AI systems already in use or clearly on a path to deployment-termed as near-term AI risks to health. The most prominent near-term AI risks to health are outlined, including algorithmic bias, erroneous clinical diagnoses and recommendations, AI-enabled health disinformation, harms to mental health, AI-driven mass unemployment, lethal autonomous weapons systems, AI-enabled chemical and biological weapons and AI as both a vulnerability and an enabler of cyberattacks on health systems. Each risk is already causing harm, or could plausibly do so soon, to the health of substantial proportions of populations, thereby collectively satisfying the criteria for a public health problem. Recognising near-term AI risks to health as a public health problem rebalances the benefit-centric narrative. The article addresses counterarguments and briefly outlines potential risk management responses: mandatory AI health impact assessments for high-risk systems, public health-oriented surveillance of AI-related harms and systematic embedding of public health expertise within AI governance structures.