Background. Recent advances in algorithmic content generation hold promise and controversy in many areas, including work, education, art, research, and psychological assessment and intervention. Drawing on Higgins’s Self-Discrepancy Theory, we explore how AI-generated images can be used to assist in expressing people’s thoughts about their self as they see it, desire it to be, and feel others expect them to be. Method. Participants used the software Midjourney to generate images representing actual/own, ideal/own, and ought/other self-states by providing a list of verbal prompts as input, iteratively modifying the results, and selecting a single image to illustrate each state. The images were used within the same session as visual anchors for semi-structured interviews to derive verbal accounts of self-states and discrepancies, complemented by psychometric assessment. Results. Participants found it feasible to depict their self-states and the resulting images helpful in describing and contrasting their actual/own self-concept with their ideal/own and ought/other self-guides. The protocol fits within a single session and, beyond conventional verbal transcripts, yields a record of verbal prompts and images for subsequent analysis. We describe the protocol and provide an analysis of verbal accounts alongside their pictorial anchors. Conclusions. Text-to-image generation as an assistive interview tool promises to be engaging for clients and useful in addressing the long-standing difficulty of qualifying and quantifying discrepancies between self-states by allowing the rapid creation of highly descriptive images without reliance on artistic ability. We interpret the interaction process within Vygotsky’s Sociocultural Theory, where AI is used as a tool or interactive artefact in a mediated action aimed at a more complex and accurate self-expression through the facilitation and visualisation of ideas.
