AI has the potential to improve health equity - but it must be trained with diverse datasets: here’s why:

What is AI?

AI stands for Artificial Intelligence. It is a program used to process and analyze large quantities of data. AI can handle a substantially larger amount of data than a human being ever could. Patterns and clues in medical data can be detected by AI that would have otherwise been missed by human eyes. The AI sector is valued to reach $150 billion by 2026. AI is a powerful tool that has already revolutionized the way healthcare operates; it can also assist us in improving inequities.

How AI can improve health equity:

AI can identify the source of a medical problem that is the product of inequity. For instance, AI could analyze a dataset and discover a link between a disease and a certain racial or socioeconomic group. From there, AI can assist in identifying further action and allocating resources that can be used to provide more effective care for underrepresented groups. Links to diseases that affect one race more than another or one social group more than another can be detected by AI and treated sooner, improving health equity. Recently, there was a discovery that AI could predict the race of a patient when there were no other racial indicators, something a human being wouldn’t be able to do. This finding has great implications for catering to the different biological needs of every group and ensuring that they receive proper, personalized care.

What are the implications of AI on health equity?

Better care for all groups, not just the majority, is improving because of AI. Healthcare providers can better serve minority communities (LGBTQ+, Black, etc.) that suffer disproportionately from disease and lower quality of life by providing complete and personalized care. Additionally, AI allows healthcare to become more affordable and accessible because of the efficiency of the system. This can allow access to low-income communities that may not have been able to gain access to healthcare.\

Why AI must be trained with diverse datasets:

While all of these positive outcomes are great, we must be wary of the way we use AI. Before accepting the outcome AI provides, datasets must be trained on a broad set of data to avoid biases that exclude certain races. If AI is trained on a broad, rather than a narrow set of data, the technology can be applied to better serve all races and social groups, rather than just one. Often datasets can become non-inclusive because they are based on the demographics of patients within a healthcare region. If one region is made up of primarily white patients, AI datasets might not have enough information to properly treat minority communities, and this is a problem.

Without data training, racial and social disparities can worsen because of biases embedded in AI trained with narrow data, and the benefits AI can provide in health equity are lost.

Overall, AI can be a valuable tool to improve health equity, but we must use it properly.