AI Speech Analysis for Child Mental Health

Artificial intelligence is increasingly being applied to clinical prediction, including the prediction of which children might develop mental health problems in the future. One research programme is developing methods to identify, from samples of speech, how warm or negative mothers feel towards their children, and to use that signal to predict future mental health outcomes in childhood. Before such approaches are rolled out in clinical settings, it is important to understand how they might be experienced by the people they affect, and what hopes and fears they provoke.

A clinical tool that can predict a child's future mental health risk from a parent's speech raises profound questions about consent, surveillance, about what it means to be identified as at-risk before anything has happened, and about what responsibilities that identification creates.

The Research

The project uses AI to analyse speech samples for markers of expressed emotion, specifically the warmth or negativity with which a parent speaks about or to their child. These markers are then used as predictors of child mental health risk, with the aim of enabling earlier identification and intervention. This approach sits at the intersection of clinical psychology, natural language processing, and mental health prediction, a domain with significant technical potential and equally significant ethical complexity.

To support responsible development of the technology, the research team commissioned a set of speculative fiction stories, written by a creative writer in collaboration with young people, clinicians, and researchers, set 10–20 years in the future and designed to explore how the use of such technology might affect different people in different circumstances. The stories were not intended as predictions or endorsements, but as starting points for structured public conversation about the technology's implications before it reaches clinical practice.

My Involvement

I was invited to a consultation session to provide feedback on the speculative fiction stories. The session asked participants to reflect on their initial reactions - what they liked and disliked, what the stories provoked, and what they felt was important to consider when researchers and clinicians implement approaches of this kind. The involvement was designed to ensure that the voices and concerns of young people were part of the conversation at an early stage, before the tool was deployed, rather than after.