Kids-SIT: AI tool analyzes children's social behavior via video with 74% accuracy

Kids-SIT (Simulated Interaction Task for Children) is a novel web-based platform designed to computationally quantify children's social interaction behavior by engaging them in a standardized video conversation scenario. In a pre-registered study with 21 healthy children (ages not specified), the tool successfully elicited naturalistic behaviors: children maintained eye contact more while listening than speaking, and smiling/gazing occurred more during person-directed segments than picture-description phases. Researchers manually annotated three socially relevant non-verbal behaviors—gaze deviation, smiling, and nodding—and compared them against three automated extraction methods. Semantic analysis of verbal responses using LIWC confirmed the scenario felt natural.

In an exploratory clinical classification using automated behavioral features, Kids-SIT differentiated healthy children from 11 children with social anxiety disorder (SAD) with an AUC of 0.74, significantly above chance. The tool's accessibility (web-based, no specialized hardware) makes it scalable for schools and clinics. While sample sizes are small, the results demonstrate that automated video analysis can capture clinically meaningful social interaction patterns in children. The researchers suggest Kids-SIT could become a low-cost screening tool for childhood social anxiety and developmental disorders, paving the way for larger validation studies.