Beware the algorithm

Predictive analytics are increasingly embedded in children’s social care, often without adequate scrutiny of their assumptions, accuracy or social consequences. While racialised harms from criminal justice algorithms are well documented in the US, concerns in social work remain underexplored. Models trained on data from unequal systems risk reproducing or intensifying existing social gradients. This scoping review maps empirical research on predictive analytics in children’s services. From 5,393 records, 37 studies meet the inclusion criteria. Using O’Neil’s framework of opacity, scale and damage, we examine how algorithms are developed, operationalised and reported. Interest is growing in using predictive tools for resource allocation, service design, risk screening and decision support. However, most models lack transparency, provide limited information on development or error rates, and rarely examine algorithmic bias or the harms of false positives. Many studies anticipate rapid scale-up using linked data sets, raising significant concerns about governance, consent and accountability

Analytics · Empirical research · Predictive analytics · Scrutiny · Child Abuse and Trauma · Ethics and Social Impacts of AI · Social Work Education and Practice · Social Work