You do not currently have permission to view the full text for this document. Please sign in to an account with an active subscription or contact support.
Artificial intelligence has the potential to revolutionize the assessment of complex psychoanalytic constructs, particularly in narrative personality assessment. This study explores if artificial intelligence bots, powered by large language models, and utilizing expert-refined system prompts, can rate narratives accurately using four dimensions of the Social Cognition and Object Relations Scale–Global Rating Method (SCORS-G). Two experts familiar with the SCORS-G system developed highly detailed system prompts for four SCORS-G dimensions. Two bots were created for each dimension, one using ChatGPT-4o and one using Claude-3.5-Sonnet. Bots rated 196 narratives taken from SCORS-G training materials. These narratives are routinely used to train and assess human raters. Agreement between bot and human experts were assessed using intraclass correlation coefficients. Level of agreement was at or exceeded the level obtained by most human raters after initial training, with single-rater intraclass correlation coefficients being notably high. Of note, bots were particularly accurate at rating dimensions that focus on relational and emotional aspects of narratives (e.g., affective quality of representations; emotional investment in relationships). Overall, the study provides support for the utility of artificial intelligence bots in alleviating burdens associated with narrative assessment of personality. Findings indicate that they even have potential for assessing complicated features of object relations. Bots have some applications that can be employed immediately, particularly in research settings, while additional research is necessary before they are capable of being used in clinical settings.