Answers are grounded in this paper and cite their sources.
This study developed and pilot-tested an AI-based algorithm to detect potential child abuse in YouTube videos featuring children. An analysis of 634 videos revealed high overall sensitivity (0.801), with particularly strong performance in mukbang content (0.983). This outcome reflects the algorithm’s intentional design prioritising sensitivity, given the critical importance of avoiding missed cases in child abuse detection. However, its low specificity (0.407) and frequent false positives were due in part to its reliance on static visual cues and inability to interpret contextual features such as emotional expressions or caregiver – child interactions. This study suggests the feasibility of using AI as a first-level filtering tool for digital child protection. It also highlights the need for context-awareness and multimodal learning to improve accuracy and ethical applicability, while emphasising the importance of human-AI collaboration in social work practice for responsible interpretation and intervention.
No takes yet. Share an insight, caveat, or question.
Kang et al. (2025) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: