MEDITATION EXPERIENCE WITH A SOCIAL ROBOT USING LARGE LANGUAGE AND VISION LANGUAGE MODELS Exploring Non-Duality Awareness and Supporting Self-Reflective Deconstruction through Multimodal Dialogue with a Robot

Sammanfattning

Nowadays robots are becoming more and more part of our society. We always expect fromrobots to solve problems, give advice and provide answers. Robots are often designed to becomputationally intelligent, but it is still a challenge the design of the emotional intelligence.The project explores how a robot with a multimodal dialogue system can be designed to supportself-reflective deconstruction within a human-facilitated workshop setting. Rather than treatingdialogue as a tool for efficiency or task completion, the project approaches conversationexploring the ambiguity of the language and of the human thoughts. The system was designedas a meeting point between meditation practice, psychotherapy, art and AI. A large languagemodel was constrained to support this self-reflective inquiry. A vision-language modelaccompanies the dialogue, with natural images. These components were brought togetherthrough an embodied conversational interface using the FurHat social robot, creating aworkshop experience in Stockholm, attended by several participants from Sweden. Theresearch unfolded in three stages. Participants first interacted with a web-based dialogue systemin a pre-interview designed to capture how they describe their sense of self and experience.Their responses were analysed to measure self-reflective deconstruction: how their wordchoices, hesitations, and references to direct experience revealed a gradual questioning,loosening, or stepping back from their ordinary sense of self. Finally, participants joined ahuman-facilitated workshop where they encountered the embodied multimodal system andengaged in guided reflective dialogue. Instead of measuring clinical outcomes, the projectfocused on the qualities of the experience itself: curiosity, openness, emotional engagement andlast but not least the self-reflective deconstruction. The robot became less of a tool and more ofa conversational mirror. The project contributes to research in multimodal dialogue and human-AI experience design by exploring how AI systems support open-ended self-inquiry rather thanproblem solving. Non-duality can be added to a robot’s dialogue by teaching it to acceptuncertainty, allow silence, and ask questions instead of always giving explanations. In this way,the robot creates a sense of conversational presence. It can respond to the participant’s statewithout using rigid scripts and without treating the human as a problem that needs a measurablesolution. AI then becomes a tool for relational, flexible, and non-reductionist interactionbetween humans and robots.