According to a study, AI is unable to identify symptoms of sadness in social media posts by Black Americans

A recent study has uncovered a significant disparity in AI performance regarding the detection of depression symptoms on social media platforms. While the technology proves adept at identifying such signs among White users, it struggles when it comes to Black users. This disparity prompts serious considerations about the reliability of using diverse datasets to train AI algorithms for healthcare purposes.

Conducted on Meta Platforms (Facebook), the research underscores a notable inconsistency. The AI model utilized in the study demonstrated noticeably lower accuracy in pinpointing depression markers in posts from Black individuals compared to its effectiveness with posts from White users. This discrepancy, highlighted in a paper published in the Proceedings of the National Academy of Sciences (PNAS), suggests a failure to adequately consider racial factors in language-based mental health assessments.

Previous research has established linguistic patterns, like the use of first-person pronouns and self-critical expressions, as indicative of depressive symptoms. However, in this recent study involving 868 volunteers of diverse racial backgrounds, researchers found that these patterns did not universally apply. Co-author Sharath Chandra Guntuku from Penn Medicine expressed surprise at this departure from previous findings.

While cautioning against the use of social media data for diagnosing depression, Guntuku suggests its potential value in assessing risk levels. In a separate study, his team explored social media language to gauge the mental health impact of the COVID-19 pandemic.

Collaborator Brenda Curtis from the U.S. National Institute on Drug Abuse stresses the significance of social media language in predicting treatment outcomes for substance abuse disorders, highlighting its broader relevance beyond depressive symptoms alone.