Why Psychiatric AI Needs Psychiatrists: One Clinician's Experience at IXO

By Dr. Robert Banks — 2025-11-05

Dr. Robert Banks chose psychiatry because he believed language was medicine's most underused diagnostic tool. Fourteen years in, he still believes it — and he's increasingly concerned about what happens when AI tries to replicate psychiatric assessment without clinical grounding.

Why psychiatry is different

"Psychiatry is different from most medical specialties in a way that matters enormously for AI," he explains. "There's almost no objective test. No blood result, no imaging finding that says 'this person has depression' or 'this person is at risk of suicide.' The diagnosis emerges from a conversation — from noticing what someone says, how they say it, what they don't say, how they present. That's extraordinarily hard to train an AI to do well."

Banks joined IXO after reviewing a mental health AI tool being piloted by a large employer. His assessment was stark: the tool was producing safety flags based on keyword detection rather than clinical context, creating false positives that were damaging people's employment situations while missing genuine risk. "It had been built by engineers, reviewed by data scientists, and no psychiatrist had looked at it closely. That's a category error."

Three areas of focus

His IXO work covers three areas. Safety assessment evaluation — reviewing AI responses to disclosures of suicidal ideation or self-harm for appropriateness and clinical safety. Diagnostic reasoning evaluation — testing whether AI assessments of depression, anxiety, and psychosis presentations meet clinical standards. And therapeutic response evaluation — assessing whether AI mental health support tools respond in ways that are helpful and not harmful.

The invisible harm

"The second-order effects in mental health AI are significant. A tool that normalizes avoidance behavior, that reinforces unhelpful cognitions, that fails to recognize when someone needs urgent care — these cause harm that's invisible in accuracy metrics. You need psychiatrists to catch them."

Read Why Psychiatric AI Needs Psychiatrists: One Clinician's Experience at IXO on the IXO blog