The evolution of artificial intelligence is poised to continue sparking game-changing developments in the way in which we live our daily lives. For that reason, health equity proponents have had high hopes for AI’s potential for driving material enhancements in the delivery and outcomes of health care, particularly for historically marginalized populations. However, new research suggests that AI tools can actually perpetuate health disparities instead of reducing them.
The results of a study just published in Digital Medicine show that machine learning algorithms developed to diagnose a common infection that affects women can contain biases among certain racial and ethnic groups and, therefore, have inherent risks of furthering the very inequities that many hope AI could eliminate in the health care realm.
In analyzing the effectiveness of machine learning capabilities in diagnosing bacterial vaginosis (BV), a commonly occurring condition among women, the researchers reviewed data from 400 participants, including 100 women from each of the Asian, Hispanic, Black and White groups represented. The accuracy of the results varied widely across the study’s different populations with the most false-negative outcomes, for example, associated with the Asian women and the most false-positive results associated with Hispanic women. The algorithms performed lowest for the Asian participants, and highest for the White participants, underscoring that certain AI methods do not treat racial or ethnic groups in the same way.
The good news for health equity is that AI, including machine learning, has been shown in the larger historical body of research to serve as a potentially strong tool in medical diagnostics generally. But this latest research also makes clear that AI can exhibit biases towards certain racial and ethnic groups, which can exacerbate health disparities rather than eliminate them.
The research report can be found here. The local and global health equity communities should continue to monitor AI developments closely to help ensure that its benefits are leveraged, while its drawbacks are identified and avoided. With AI’s rapid growth set to move even faster in the months and years ahead, it will certainly be closely watched by health equity advocates going forward.
