Can AI be used to promote equality in healthcare access?

Access to healthcare remains a pressing concern globally, especially among minority communities. The inherent societal bias has been a significant concern across the healthcare spectrum, especially in the wake of growing economic and social disparity.

With the increasing adoption of AI technologies in different areas, there has been rising hope that its adoption in healthcare will help improve on existing medical technology, personalised medicine and provide reprieve and equity of access to the underserved and socially disadvantaged communities.

However, the adoption of AI in healthcare must overcome critical challenges to ensure that AI technologies do not amplify the existing biases, which would render such a promoter of healthcare access inequality.

Based on these concerns, the adoption of AI provides hope as it promises to personalise medicines, improve existing technologies, and provide access to customised healthcare to underserved communities by utilising big data. Nonetheless, all these potential benefits are put at risk by the possibility of AI magnifying the existing societal disparities. According to a professor at MIT Institute of Medical Engineering, biases in healthcare AI arise from algorithms and datasets that propagate the inherent societal bias. The three main sources of bias are statistical bias, variance in sample size, and noise that emanates from large data size or changes in the model used.

Overcoming these challenges is crucial if the goals of AI are to be achieved. Stakeholders and thought leaders in the field propose several ways. The key to these solutions is to ensure that the AI tools are trustworthy, which calls for the need for the said tools to be inclusive, fair, and support personalisation. According to researchers at Jameel Clinic, an initiative that supports AI research in healthcare, AI tools should be diverse in the operationalisation with the ability to serve any community, population, or subpopulation. This calls for algorithms to be trained and validated in multiple communities, cities, and countries while also considering the need to honour patient privacy. It is also imperative to train the AI to consider individual uniqueness, which is important in addressing the pre-existing biases in the healthcare system. Besides, conducting exploratory error analysis could identify common error threads, thus helping identify bias.