#68: Beyond One-Sided Arguments: Evaluating Perspective-Taking in Middle School Essays with GPT-4o and Human Scoring

In this exploratory study, we compared GPT-4o zero-shot scoring with human ratings of middle school students’ essays collected across multiple studies. Results showed that GPT-4o achieved acceptable reliability with human raters, highlighting the potential of GenAI for assessing perspective-taking. Further, we compared linguistic features of essays with a single perspective versus those with multiple perspectives. Future research should examine the quality of perspective-taking, moving beyond recognition of alternative perspectives to evaluating attempts at rebutting counterarguments.