Artificial intelligence in health care and social inequality: will AI reduce or widen health gaps?

Abstract

Artificial intelligence (AI) could transform health care, particularly in lowresource settings (1). The technology enables new capabilities, ranging from data acquisition to decision support, that amplify ongoing investments in digital technology to improve access to information, diagnostics, treatment, and decision-making (2). As health gaps persist between and within countries, a key question emerges: will deployable AI solutions widen or narrow existing disparities? (3). In many countries, poverty, education, geography, and race constitute risk factors that contribute both to health status and to access to health services and other determinants of health (4). Two competing hypotheses exist (5). The optimistic view holds that AI can reduce these gaps by improving the accuracy of diagnostics and therapeutics and by expanding access to services that would otherwise be out of reach demonstrates that these capabilities can profitably influence health status and survival in under-resourced settings where access to trained human capital, diagnostic devices, and therapeutics is limited (6). The gloomy view maintains that significant mechanisms exacerbate the very gaps that AI might alleviate: 1) the data on which AI training relies often reflects a chronic lack of representation, 2) the post-deployment conditions under which models operate may drift further from those on which models were trained, 3) the infrastructure and knowledge necessary to deploy new capabilities may be absent in the most vulnerable settings, and 4) oversights in governance and use could further endanger already vulnerable groups as the market for AI grows (7, 8).