Four equity considerations for the use of artificial intelligence in public health

New technologies can either improve or worsen health inequities.1 Innovative technologies involving artificial intelligence are no exception, particularly where they are adopted and implemented in health systems. Indeed, determining whether and how artificial intelligence might contribute to reducing or exacerbating health inequities has been identified as a priority research area by several stakeholders and by numerous ethics and policy guidance documents.2–4 Understanding the connection between health inequities and artificial intelligence should be a priority when deploying these technologies in public health. Since public health activities typically target populations instead of individuals and require collective action instead of individual intervention,5 introducing artificial intelligence technologies to support these activities may influence (either positively or negatively, intentionally or unintentionally) health inequities more than in other areas. As such, identifying the distinctive equity considerations and dimensions that might emerge in the public health context is critical. However, doing so is not a straightforward task. First, we cannot simply look to past technological innovations to determine which health equity considerations or implications might arise with the use of artificial intelligence in public health because technological innovations and their diffusion in health systems each produce or interact with health inequities in novel ways.1 We may not be able to assume that the trends or pathways that create or prevent inequities will be the same when implementing artificial intelligence technologies as they are with other technological innovations. This limitation may be particularly challenging with artificial intelligence technologies given their use of big data and machine learning. Second, artificial intelligence represents a vast and sometimes contested area of study and application. Here we define artificial intelligence as a branch of computer science that explores the ability of computers to imitate aspects of intelligent human behaviour, such as problem-solving, reasoning and recognition.2 Technologies that are supported by artificial intelligence are therefore numerous, and include natural language processing, object recognition and reinforcement learning, among others. The ways in which these technologies might be deployed in public health are equally numerous, including digital disease surveillance, machine learning to predict incidences of noncommunicable diseases, and others. Finally, given that health inequities are often defined as differences in health that are unjust, even what should be counted as health inequities and what it means to achieve health equity may differ according to the nature of the new technology, how it is or has been integrated into health systems and our judgements about its interaction with the public’s health.6 As a result, before research or health system interventions in this area are developed or implemented, we should first seek to conceptually map the unique ways in which inequities might manifest when artificial intelligence is implemented or used in public health. Indeed, important work examining the unique equity dimensions associated with specific artificial intelligence technologies in this area has begun.7 Yet, we posit that there are general equity considerations and dimensions that can be identified and used as starting points for the reflection of equitable artificial intelligence in public health, and that it would be of benefit for the field to have these identified and enumerated. We will briefly describe four key equity considerations and dimensions and conclude by discussing how they can be used as starting points to further understand and enhance the equitable deployment of artificial intelligence in public health.