AI fails to detect signs of depression in black Americans post

Analyzing social media using artificial intelligence may reveal signs of depression in white Americans, according to a study highlighting the risks of training AI models for health care-related tasks without data from different racial and ethnic groups. Can, but not in black counterparts.

The researchers reported that the AI ​​model used for the study was three times less predictive of depression when applied to Black people using the Facebook Meta platform compared to white people.

“Race appears to have been particularly neglected in work on language-based assessment of mental illness,” the authors of the US study wrote in a report. Published in PNAS, Proceedings of the National Academy of Science.

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

For the new study, researchers used “off the shelf” AI tools to analyze language in the posts of 868 volunteers, including an equal number of black and white adults who shared other characteristics such as age and gender. Was.

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

Co-author Sharath Chandra Guntuku of the Center for Insights said that the use of “I-talk” or self-focused meditation, and self-deprecation, self-criticism, and feeling like an outsider contribute to depression, especially for white individuals. Was related to. Results in Pain Medicine.

“We were surprised that these language associations found in many prior studies did not hold across the board,” Guntuku said.

Guntuku acknowledged that social media data cannot be used to diagnose a patient suffering from depression, but it can be used for risk assessment of an individual or group.

An earlier study conducted by his team analyzed language in social media posts to evaluate the mental health of communities during the COVID-19 pandemic.

Brenda Curtis, of the US National Institute on Drug Abuse at the National Institutes of Health, said that in patients with substance use disorders, language suggestive of depression on social media provides information about the likelihood of treatment dropout and relapse. , who has also worked on this. the study.