AI Struggles To Detect Depression Signs In Social Media Posts By Black Americans

AI’s inability to detect signs of depression in social media posts by Black Americans was revealed in a study published in the Proceedings of the National Academy of Sciences (PNAS).

This disparity raises concerns about the implications of using AI in healthcare, especially when these models lack data from diverse racial and ethnic groups.

The Study

The study, conducted by researchers from Penn’s Perelman School of Medicine and its School of Engineering and Applied Science, employed an “off the shelf” AI tool to analyze language in posts from 868 volunteers. 

These participants, comprising equal numbers of Black and white adults with similar age and gender demographics, also completed a standard questionnaire used in healthcare to screen for depression.

The study involved analyzing social media posts using an AI tool to detect linguistic markers of depression.

Prior research indicated that frequent use of first-person pronouns and certain negative emotions were signs of depression. 

AI’s Inability To Detect Depression For Black Americans

However, the findings revealed that while depression severity in white individuals led to increased use of first-person pronouns (“I-usage”), such a correlation was not found in Black individuals. 

However, this new study revealed that these markers apply predominantly to white individuals, not Black individuals. 

Additionally, negative emotions relating to feelings of outsider-belongingness, self-criticism, worthlessness/self-deprecation, and anxious-outsider language were linked to greater depression severity among White, but not Black, individuals.

“We were surprised that these language associations found in numerous prior studies didn’t apply across the board,” said study co-author Sharath Chandra Guntuku.

This revelation indicates a significant oversight in the AI’s training process, particularly in the context of mental health assessment.