Design for Cognitive Friction in Human-AI Interaction: A Student‑Centred AI Scaffold for Cognitive Resilience and Intellectual Agility

Unlike answer-giving chatbots, Erwin adopts a supportive-critic stance: it provides stage-aware, dialogic scaffolding that widens idea spaces and supports criteria-based consolidation through nudges and commitments. The analysis drew on 1,999 student and Erwin turns organised into 171 sessions nested within 78 student logs, alongside post-module qualitative feedback.

The study combined reflexive thematic analysis with a full-corpus, non-inferential descriptive analysis of selected interaction-process indicators, including initiative balance, futures probes, completed repair sequences, and system boundary or refusal episodes. A purposive sample of six final artefacts provided illustrative triangulation of interaction patterns and student project trajectories. The analysis identified four practices: disciplined divergence, in which Deliberate Cognitive Friction (DCF) prompts appeared alongside expanded frames without displacing agency; criteria-aware convergence, in which students translated broadened views into clearer problem statements, constraints, and next steps; cognitive resilience visible through repair and persistence sequences; and a supportive-critic experience shaped by answer resistance and meta-reflection.