Introduction In their contribution, Ugar and Malele 1 shed light on an often overlooked but crucial aspect of the ethical development of machine learning (ML) systems to support the diagnosis of mental health disorders. The authors restrain their focus on pointing to the danger of misdiagnosing mental health pathologies that do not qualify as such within sub-Saharan African communities and argue for the need to include population-specific values in these technologies’ design. However, an analysis of the nature of the harm (...) caused to said populations once their values remain unrecognised is not offered. Building on Ugar and Malele’s considerations, we add a further perspective to their analysis by showing the need to design considering intended values to avoid the occurrence of epistemic injustices. 2 First, we argue that failing to acknowledge the hermeneutical offerings of the populations interacting with these systems can qualify as _contributory injustice_.shrink)

Ugar and Malele 1 critique the use of ‘generic’ technologies like artificial intelligence (AI) and machine learning (ML) for mental health diagnoses, particularly in sub-Saharan African countries. They highlight how these AI medical tools often overlook traditional perspectives and local contexts. The article has the merit of working on ethical issues regarding the particularities and risks of using AI and ML for health diagnosis in the Global South, an urgent and neglected topic. According to the authors, the use of these (...) AI technologies leads to overgeneralisation in diagnosing mental disorders, which could be especially problematic in the mental health field because of value-laden judgements intrinsic to the definition of mental health disorders. This argument is theoretically grounded in the hybrid conceptualisation of mental disorders proposed by Wakefield. 2 This author’s perspective incorporates both factual and value components in defining mental disorders, framing them as context-dependent ‘harmful dysfunctions’... (shrink)

This commentary aims to respond to some criticism against our paper entitled ‘Designing AI for Mental Health Diagnosis: Challenges from sub-Saharan value-laden Judgments on Mental Health Disorders’.1 While we are sympathetic to the invaluable critiques of some authors, we show that some misunderstanding arises in reading our conceptualisation of the condition we use as a central example of disease in our paper. We argue, in our paper, that there are obvious context-specific value judgments when it comes to mental health disorders, (...) which obscures an application of generic machine learning (ML) in the diagnosis of these conditions. To contextualise our contention, our understanding of mental health disorder is grounded in Jerome Wakefield’s2 hybrid theorisation of disease as that which encompasses aspects of the normativist and naturalist accounts of health. Wakefield theorises that facts are necessary but insufficient to make a condition a disease. Wakefield claims that a disorder requires harm that involves values. Thus, values and facts must go concomitantly for a condition to be considered a disorder. Here, a disorder is conceived as something that is a ‘harmful dysfunction’. On the one hand, harmful is value laden—characterised by how a society conceives the notion due to their set norms. On the contrary, dysfunction is a solely scientific term due to a failure to function in accordance with the reference class of an organism. …. (shrink)

Ugar and Malele write that generic machine learning (ML) technologies for mental health diagnosis would be challenging to implement in sub-Saharan Africa due to cultural specificities in how those conditions are diagnosed. For example, they say that in South Africa, the appearance of ‘schizophrenia’ might be understood as a type of spiritual possession, rather than a mental disorder caused by a brain dysfunction. Hence, a generic ML system is likely to ‘misdiagnose’ persons whose symptomatology matches that of schizophrenia in the (...) South African context. The authors thus claim that ‘a generic or universal design cannot be effective given the heterogeneity of value judgements in defining what mental health disorders are in different contexts’.1 Should we take this to mean that ML systems ‘should not be designed with a generalised perception of mental disorders,’1 as the authors suggest? On the contrary, my view is that generic ML can be useful, with the caveat that issues of cultural sensitivity may engender translational challenges in various national and cultural contexts. Yet, if it can be demonstrated that a generic ML system can reliably and accurately pick up on shared symptomatologies across cultures despite what any individual or community might believe about their actual causes, this aspect is surely the function and value of diagnostic ML which we should take to matter. Cultural gaps in aetiological understandings of mental health conditions make it all the more important to advance generic ML as a calibrating diagnostic tool. In this commentary, therefore, I will endeavour to make a case for generic ML in mental healthcare. ### Aetiological (mis)understanding across cultures The fundamental complexity to which Ugar and Malele direct our attention is the fact that mental health conditions are attributed to variable causal explanations, rather than a …. (shrink)