Interrupting the Oracle : Rephrasing and Questioning as Designed Friction

Sammanfattning

Conversational AI lets users move quickly from one answer to the next. This thesis examines what happens when that movement is interrupted through epistemic pauses. In the pause condition, users had to rephrase each AI explanation and state what remained unclear, unverified, or unresolved before continuing. These entries were logged by the system, but the AI did not use them when generating its next response. A qualitative within-subjects design probe with six participants compared the pause interface with a seamless baseline across two conceptual topics: opportunity cost and technical debt. The analysis drew on interaction logs, written reflections, and semi-structured interviews, analysed through reflexive thematic analysis and interpreted with concepts from Rabardel, Ihde, and McDermott et al. Rephrasing showed whether participants repeated, struggled with, or reformulated the preceding AI response. The question step helped participants name what they wanted next: a source, example, clarification, or scope distinction. Across repeated turns, some participants treated the pause more procedurally, while others continued to use it to organise their own understanding. The enforced pause made participants' current grasp and uncertainty visible. The thesis contributes the epistemic pause as intermediate-level design knowledge for designing epistemic friction in conversational AI.