Citation:
Abstract:
Understanding how LLM provide emotional support differently from humans is crucial given the prevalence of AI emotional support. We focused on a highly researched emotion regulation strategy, cognitive reappraisal, the ability to reinterpret emotional situations to impact their intensity (N = 3,740). We compared trained humans and GPT-4 in their ability to perform reappraisal of vignettes (Study 1) and in real-time interactions rated by the person experiencing the reappraisal (Study 2), showing that GPT-4 outperformed humans. Incentivizing humans to produce better reappraisals led to increased time spent on the task but did not close the gap (Study 3). Labeling the reappraisals as produced by AI reduced participants’ evaluation of the effectiveness of the reappraisal, though GPT-4 was still considered more effective (Study 4). Language analysis revealed differences in specific language use that were associated with quality. These results advance our understanding of emotional support provided by humans and AI.