AI is poised to solve capacity problems by cutting wait times, easing overburdened systems, and helping more people get served, faster. Therefore, AI is a promising solution to the mental health industry and others facing these strains.

But scaling AI raises a question: does the AI model work equally well for everyone? Drawing on a nationwide Danish AI study on health outcomes, this session demonstrates how an AI model can appear highly accurate while systematically underperforming for particular groups. We show how hidden bias emerges, how it can be detected, and what happens when richer contextual data is incorporated.

Main takeaways from the session will be an insight to a playbook for spotting hidden bias in real-world data and its impacts on your AI model.

We also look at what this means in clinical practice: where AI should be used in psychiatry, where it shouldn't, and how to draw that line responsibly.