Reflexion: An AI prototype for guided emotional self-exploration

Reflexion is an experimental AI system created to move beyond the surface level of emotional recognition toward deeper self-reflection. Rather than stopping at labeling how someone feels, the prototype guides users through layered conversations designed to explore emotions more thoroughly and connect them to personal values and goals.

How it works#

The system combines three core mechanisms. Emotion detection identifies what a user is experiencing; reflective prompting guides progressively deeper engagement with those feelings; and metaphorical storytelling helps users articulate their emotional experiences in richer ways. The design follows a structured pathway—starting with basic emotional awareness and advancing toward thoughtful action planning informed by what matters most to the user.

Grounding in established theory#

Reflexion draws on three established frameworks: expressive writing (the therapeutic practice of writing about emotions), cognitive restructuring (reframing thoughts and beliefs), and self-determination theory (which emphasizes autonomy, competence, and relatedness). This combination aims to create a coherent experience where reflection genuinely supports personal growth rather than offering generic sentiment analysis.

Current status and limitations#

The researchers emphasize this is an early prototype without empirical evidence of effectiveness. Informal feedback from reviewers suggested the layered interaction model felt understandable and potentially valuable, though the final stage—action planning informed by agency and empowerment—remains partly aspirational. The work documents one theoretical approach to affective computing rather than a validated intervention.

The research was submitted to arXiv in April 2025 under computer science's human-computer interaction category, with a revision posted in July 2026.