Abstract
The most pressing ethical challenge which Artificial Intelligence poses in child welfare lies in the capacity of AI-assisted decision support systems to covertly restructure professional practice—in a manner which hollows out relationship-based care, distributes responsibility across many shoulders, and constrains the exercise of professional discretion in ways that remain largely invisible to all parties involved. The analysis takes as its starting point a sociotechnical premise which is sometimes missing from normative discussions of AI and social work: Most systems currently deployed in child welfare operate in the so-called human-in-the-loop model, as decision support tools whose outputs guide or constrain professional judgement without replacing it. It is precisely this hybrid character which produces the ethically most substantial and pressing problems: automation bias, responsibility diffusion and epistemic closure. Drawing on the capability approach in its elaboration for socially disadvantaged children and youth, this contribution argues that explainability should be understood as a structural condition for the preservation of professional-ethical integrity and for the realisation of children’s rights.