Teach students when not to trust AI
Trust calibration means learning when to rely on a system and when to question it. ICE-T treats that as teachable. The classroom move is not vague confidence-building; it is letting learners see, modify, and explain how a machine-learning system can work and fail.
What the source reports
Haritz, Krone, and Liebig present ICE-T, a teaching plan for calibrated trust in AI systems. Calibrated trust means knowing when to rely on a system and when to question it. The plan has three classroom moves. First, show ideas in more than one form. Second, give students step-by-step control through a Use-Modify-Create path. Third, ask them to explain errors with a process model. The authors link these moves to research on algorithm aversion, AI literacy, and mental models. Their claim is plain: students need more than a black-box demo. They need practice seeing how an AI system frames a problem, changing parts, and explaining mistakes.
Original source
- Title
- Addressing Trust in AI Systems through Education: A Didactic Perspective (ICE-T)
- Author
- Haritz, Krone & Liebig
- Publication
- arXiv
- Date
- Wednesday, September 2, 2026