Dream content analysis using Artificial Intelligence

An artificial intelligence algorithm trained on over 35,000 dreams from 635 users identified 47 reliably recurring dream themes, building on prior dream content ontologies. Female dreams showed higher incidence for most themes compared to male dreams, though effect sizes were small. The algorithm also identified which themes best predicted whether a dream was associated with positive or negative mood. The AI-based system is a promising tool for analyzing dream content patterns.

Study at a glance

Abstract

We developed a dream content analysis system (DCAS) based on an artificial intelligence (AI) algorithm that was trained using a relatively large corpus of over 35,000 dreams. This sample of dreams were supplied by 424 female and 211 male users over 4 years who had posted them at the dream posting website and app Dreamboard.com. Building upon previous dream content ontologies developed by Hall, Van de Castle, Domhoff and Bulkeley, forty-seven reliably identified dream themes emerged from repeated application of algorithm and agent training procedures. DCAS reproduced most of the key dream content themes from these previous ontologies but also returned some unexpected findings. Mixed-model estimation detected significant male-female content differences for 34 dream themes, with female dreams evidencing higher incidence percentages for most themes, but effect sizes were small. Mixed-model logistic regression identified those themes that best predicted self-reported positive or negative mood associated with dreams. We conclude that the AI-based DCAS algorithm developed here is a promising tool for detailed analyses of dream content patterns.