Can AI Diagnose Mental Disorders

MIT researchers recorded conversations with depressed patients and found: an algorithm detects depression from speech pauses, response speed, and word choice more accurately than standardized questionnaires — and several weeks before the patient themselves reports worsening. This was 2019. Since then the technology has moved far ahead.

AI diagnostics in psychiatry is one of the most promising and simultaneously most debated areas. Its possibilities make it promising. Questions of accuracy, ethics, and what "diagnosis" means make it contested.

What Modern Algorithms Can Do

The field is developing rapidly, and the application spectrum is already impressive.

Speech analysis. Natural language processing systems identify patterns associated with depression, bipolar disorder, and even early stages of dementia: speech speed, pause length, vocabulary diversity, prevalence of negative words, use of the pronoun "I" (elevated — a depression marker). Accuracy in several studies: 72–85%.

Facial expression analysis. Computer vision analyzes video recordings and detects microexpressions characteristic of depression: reduced amplitude of emotional expressions, slowed movements, a particular quality of eye contact.

Passive smartphone data. Movement patterns (GPS), social media activity, phone usage time, sleep quality (via accelerometer) — all correlate with mental state. Monitoring systems using this data can track changes in state in real time.

Biomarkers. Combining neuroimaging data (fMRI, EEG) with machine learning identifies patterns of brain activity associated with specific disorders — and potentially predicts treatment response before it begins.

Real Successes: From Screening to Monitoring

AI application is most evidence-based in two areas.

Population screening. Identifying people with elevated risk of disorders in large groups — among students, military personnel, company employees. Not a diagnosis — but a signal: "this person should speak with a specialist." This can substantially reduce time from symptom onset to seeking help.

Dynamic monitoring. For patients already in therapy, AI systems can track state changes between sessions — and alert the therapist to signs of worsening or suicidal risk. This is already used in several clinical programs.

Why This Cannot Replace Specialist Diagnosis

Despite the impressive capabilities, there are fundamental limitations.

Context. The algorithm sees a pattern but does not understand context. Slowed speech may indicate depression. Or fatigue, a cup of coffee, anxiety before an important conversation, cultural features. A clinician considers history, environment, a person's life. The algorithm does not.

False positives. When screening a large population, even a high-accuracy algorithm (90%) generates thousands of false positives. A person is told "you show signs of depression" — and this changes their self-perception. This is not neutral information.

Data bias. Algorithms are trained on data that reflects existing biases in medicine. If depression in men was historically diagnosed less often — the algorithm will reproduce this pattern.

The Correct Role of AI in Psychiatric Diagnosis

AI in diagnosing mental disorders is a tool for the specialist, not a replacement. Its optimal role: help the clinician notice what they might have missed. Provide monitoring between visits. Lower barriers to initial contact.

But the decision about diagnosis, about treatment, about how to work with a specific person in the specific context of their life — that remains with the live specialist. And likely should remain so.