How Task Design Transforms AI Interactions in the Classroom

"As generative AI tools like ChatGPT become more embedded in education, a critical question emerges: how do we design learning activities that promote critical engagement rather than passive dependence? While recent research has documented concerning patterns of student over-reliance on AI, many universities' AI guidelines remain vague, lacking concrete strategies to support deep, reflective engagement. The key lies in examining how specific features of task design shape the way students interact with AI tools."

"Luther et al.'s (2024) individual writing study, which we discussed in our previous post, exemplified exactly these concerning patterns, with many students copying AI-generated text with minimal revision, treating ChatGPT as a ghostwriter rather than collaborator. But might it be possible to change our task design to alter these dynamics? In a recent study, Zhang, Sun, and An (2025) analysed over 1,000 minutes of recorded debate sessions and interview transcripts totaling 370,551 words to examine a fascinating GenAI use case."

"They observed 22 students working together iteratively over four weeks in teams of five in fast-paced, time-constrained classroom debates. Teams of five could leverage support from GenAI, but only one student had access to the sixth team member, ChatGPT. This use case scenario required real-time collaboration, rapid decision-making, and strategic use of AI-generated responses under time pressure-10 minutes preparation, five minutes initial arguments, 25 minutes cross-examination-which created conditions for"

Individual writing tasks often produce copy-paste behavior and heavy reliance on AI, with students treating ChatGPT as a ghostwriter rather than a collaborator. Fast-paced, team-based debates with restricted AI access encouraged critical evaluation, rapid decision-making, and real-time synthesis of evidence. Public accountability and stringent time constraints shifted students from passive consumers of AI output to active synthesizers who strategically used AI-generated responses. Analysis covered over 1,000 minutes of recorded debate sessions and 370,551 words of transcripts involving 22 students over four weeks, organized in five-person teams where only one member accessed ChatGPT under specific time limits.

Read at Psychology Today