MIT Scientists Build AI Tool to Predict Suicide Risk From 16,000 Crisis Texts
Updated
Updated · MIT News · Sep 24
MIT Scientists Build AI Tool to Predict Suicide Risk From 16,000 Crisis Texts
3 articles · Updated · MIT News · Sep 24
Summary
Researchers at MIT said their new language-processing model accurately classified suicide risk in unseen crisis-text conversations, aiming to help counselors spot people facing imminent danger.
About 16,000 de-identified Crisis Text Line exchanges trained the system, which sorts conversations into non-suicidal, suicidal ideation, and imminent-risk cases involving a plan or intent to die within 48 hours.
A custom lexicon links roughly 60 words or phrases to each of 49 risk factors, letting the model show which signals drove an assessment instead of producing only a black-box score.
Mentions of lethal means, substance use, active suicidal ideation and self-injury emerged as stronger markers of imminent risk than depressed mood, while anxiety, PTSD and emotional pain were intermediate predictors.
The team said the lightweight model can run on a personal computer and may ease privacy and cost concerns, but still needs further validation and human oversight before any clinical deployment.