CHI 26 Workshop · Understanding and Engaging Critical Resistance to AI in Education
When Not to Help: A Human-Teaching Lens on AI Tutoring
We invert the usual question about GenAI in education: rather than ask how to deploy new capabilities, we ask what role GenAI should play given what we know about effective human teaching. This position draws on an observational study of ten teacher-student pairs using speech, visual annotations, and remote screen control to teach Figma to novices. Teachers consistently showed restraint—avoiding high-precision modalities when it would undermine learning—which we interpret as pedagogical wisdom, not inefficiency. We argue this reveals a precision-agency tradeoff and digital territoriality as design constraints for GenAI tutors. We contribute empirical evidence of how teachers balance help with learner agency, methodological guidance for evaluation under authentic motivation, and implications for metrics and design process that treat restraint and friction as valid design goals.
This paper had many concerns that were not included in the study.