AI Model Lacks Accurateness In Detecting Depression In Black People On Social Media

Sharath Chandra Ggunu from the Center for Insights to Outcomes at Penn Medicine. (photo: x @PennEngOnline)

JAKARTA - A study highlights the risks of using artificial intelligence models for health-related tasks without data from diverse race and ethnic groups. The study found that artificial intelligence models can detect depression signals in white Americans but not in black counterparts. This is reported in a study highlighting the risk of artificial intelligence model training for health-related tasks without data from diverse race and ethnic groups.

The artificial intelligence model used for the study is more than three times less predictive for depression when applied to black people who use Meta Platforms social media, Facebook.

"Ras appears to have been ignored primarily in studies on language-based mental illness assessments," the authors of the US study wrote in a report published at PNAS, National Academy of Sciences Prosiding.

Previous research on social media posts has shown that people who frequently use first-person pronouns, like me or me, and certain categories of words, such as demeaning terms, have a higher risk of depression.

For this new study, the researchers used an artificial intelligence tool "ready to use" to analyze the language in shipments from 868 volunteers, including the same number of black and white adults who have other characteristics such as age and gender.

All participants also filled out a validated questionnaire used by health care providers for depression screening.

The use of "I speak" or self-focused attention, and self-lowening, self-criticism and feeling like an extraordinary person associated with depression exclusively for white individuals, said study co-author Sharath Chandra Gantiu of the Center for Insights to Outcomes at Penn Medicine.

"We are shocked that the language associations found in many previous studies do not apply to everyone," said Gunau.

Social media data cannot be used to diagnose a patient with depression, admits Gunau, but it can be used for individual or group risk assessment.

A previous study by his team analyzed the language in social media posts to evaluate people's mental health during the COVID-19 pandemic.

"In patients with substance abuse disorders, language in social media showing depression has proven to provide insight into the possibility of dropping out of treatment and recurring," said Brenda Curtis of the National Institute on Drug Abuse at the US National Institutes of Health, which also works on the study.