In this episode, host Mariana Fernández Mora explores, in conversation with Janine Armin, Andy Dockett, and Sabine Niederer, how AI can act as a collaborator in collective imagination and storytelling. Returning to early experiments within the research group, the discussion examines how machine-generated text once introduced unruly narrative structures, fluid pronouns, and unfamiliar relationships between subjects, objects, and environments. Together, they reflect on why unpredictability felt so engaging and how early models enabled forms of co-creation that resisted closure, efficiency, and fixed hierarchies. The conversation considers the narrative assumptions embedded in contemporary AI systems, and how alternative, messier exchanges might reopen space for other voices, perspectives, and ways of telling stories. Soundscapes in this episode were created by artist and researcher Angelo Custódio during the Material Playground Everything Evaporates.