Zixin Chen, Haotian Li, Ziang Xiao, Huamin Qu, et al. (2026) — arXiv preprint. arXiv:2607.17643 [cs.HC]. 📄 Full text (arXiv)
As LLMs take over task execution, a central worry is that everyday AI use becomes cognitive offloading that erodes people's own capability development. This study analyses 128,569 naturalistic human-LLM conversations, translating learning-science constructs into turn-level behavioural signatures to test whether informal learning actually emerges in routine use.
The authors find that users do engage in learning-supporting behaviours — cognitive engagement, self-explanation, and elaboration — within ordinary llm interactions, tempering the pure offloading concern. The analysis contributes to debates on ai-literacy and the risks of over-reliance, and reframes student-ai-interaction as a site where incidental learning can occur. It also informs self-regulated-learning and metacognition research by quantifying how conversational patterns either preserve or displace opportunities to think.
Related Pages
- llm — related AIED work
- ai-literacy — related AIED work
- over-reliance — related AIED work
- student-experience — related AIED work
- self-regulated-learning — related AIED work
- metacognition — related AIED work