Machine learning for mental health diagnosis: power and extensions of epistemic injustice

Following the article of Ugar and Malele, Pozzi and De Proost provide a necessary addition to the discussion around machine learning (ML) in mental health diagnosis. However, their analysis of contributory injustice, a type of injustice in which the dominant social group does not recognise the epistemic offerings of the minority, provides only an introduction to the significant harm that can stem from the intersection of ML and culture. Not only does ML, untrained in certain cultural contexts, provide the broad dismissal of a group's values, but it also manifests in more direct, personal forms of harm. Testimonial injustice is a particularly troubling dimension of ML's integration into psychiatry, and this paper will examine how social power operates within these systems. First, it is necessary to understand ML as a social entity. Then, through an analysis of Fricker's account of social identity and an extension of this through Brey's framework of the social power of technology, a more complete appreciation of the epistemic injustice caused by ML algorithms in healthcare can be reached.