AI fails to detect depression signs in posts by Black Americans

 Analyzing social media using artificial intelligence may pick up signals of depression in white Americans but not in Black counterparts, according to a study that highlights the risk of training AI models for healthcare-related tasks without data from diverse racial and ethnic groups.

The AI model used for the study was more than three times less predictive for depression when applied to Black people who use Meta Platforms' Facebook than for white people, the researchers reported.

"Race seems to have been especially neglected in work on language-based assessment of mental illness," the authors of the US study wrote in a report published in PNAS, the Proceedings of the National Academy of Sciences.

Previous research on social media posts had indicated that people who frequently use first-person pronouns, such as I, me or mine, and certain categories of words, such as self-deprecating terms, are at higher risk for depression.

For the new study, researchers used an "off the shelf" AI tool to analyze language in posts from 868 volunteers, including equal numbers of Black and white adults who shared other characteristics such as age and gender.

All participants also completed a validated questionnaire used by healthcare providers to screen for depression.

The use of "I-talk" or self-focused attention, and self-deprecation, self-criticism and feeling like an outsider were related to depression exclusively for white individuals, said study co-author Sharath Chandra Guntuku of the Center for Insights to Outcomes at Penn Medicine.

"We were surprised that these language associations found in numerous prior studies didn't apply across the board," Guntuku said.

Social media data cannot be used to diagnose a patient with depression, Guntuku acknowledged, but it could be used for risk assessment of an individual or group.

An earlier study by his team analyzed language in social media posts to evaluate communities’ mental Health during the Covid19 pandemic.

In patients with substance abuse disorders, language on social media indicating depression has been shown to provide insight into the likelihood of treatment dropout and relapse, said Brenda Curtis of the US National Institute on Drug Abuse at the National Institutes of Health, who also worked on the study.