An experimental study exploring human–AI complementarity in early social-emotional learning

Many children are not receiving crucial social-emotional learning (SEL) support due to systemic constraints, such as high educator workload and training burdens. Pedagogical conversational agents (PCAs)—generative AI-powered virtual characters trained to converse with students like a teacher—offer promising solutions that could extend to SEL. However, the current affective-computing capacities of PCAs may be limiting, as SEL facilitation requires deeper emotional and relational attunement than cognition-heavy subjects (e.g., literacy, numeracy). The emotional and relational skills of human educators likely confer benefits beyond those of PCAs in SEL instruction. Despite this potential complementarity, no studies have experimentally assessed the relative performance of PCAs and educators in SEL contexts. It remains unclear if and how PCAs can help educators close the SEL gap. In this study, we used a controlled comparison to reveal complementary strengths, comparing the performance of a PCA to that of human educators when independently facilitating story-based SEL activities with a static AI child. We used a static AI child to help ensure that any performance differences could be attributed to facilitator type. The PCA (n = 18 simulations) and educators (n = 18) each delivered three SEL activities (N = 108 observations). Expert raters, blind to facilitator type, coded dialogue excerpts for evidence-based SEL support techniques and indicators of pedagogical quality. A mixed ANOVA and t-tests revealed significant differences. The PCA showed strengths in basic relational and instructional domains, maintaining respectful tone and routinely providing procedural scaffolding. Educators showed strengths in deeper SEL instruction, guiding reflection and promoting social-emotional knowledge. We discuss implications for SEL through the lens of human–AI complementarity.