Casting the Shadow for Understanding "Reasoning Ghosts"

AI systems increasingly shape everyday life through data capture and reasoning processes that often remain difficult to grasp, existing as ghosts within the shadows of data. This demo paper presents ShadowAI, an experiential AI system comprising the Shadow Capture Box and ShadowStories. Participants craft play-doh artefacts whose shadows are captured as silhouettes, and interpreted by GPT-4o which is embedded within ShadowStories through a structured prompt pipeline. This includes Shadow Parsing, Concept Proposal, Image Prompt, Story Outline, and Final Story, which are displayed as staged reasoning visualisations alongside DALL·E 3 illustrations and GPT-4o-generated narratives. Rather than relying on technical explanation alone, ShadowAI uses shadow play, creative making, and intentional ambiguity to support experiential encounters with how AI interprets. The resulting outputs may be surprising, erroneous, or absurd, prompting curiosity, shared reflection, and critical discussion about AI as an ambiguous and fallible simulation assembled from human-created data.