Calibrate assistant support forms to preserve users' cognitive participation rather than making every interaction tutor-like¶

Strategy · All strategies
Evidence · no claims cited

Description¶

The article recommends that the educational value of future AI assistants lies in calibrating support rather than universal tutoring: assistants should choose "when to answer, explain, give feedback or step back, preserving the cognitive agency through which people question, test, revise and build judgment while solving real problems." The support-form findings ground this: feedback and explaining aligned most consistently with constructive engagement, while instructing and questioning showed weaker or negative associations, especially under explicit learning-oriented framing. The article also proposes that sustaining users' cognitive participation can become an explicit objective of model behaviour, evaluated with the process measures developed in the study.

Design Implications¶

Context¶

Requirements¶

• Support must be positioned within the exchange so that part of the epistemic work remains available to the user, as feedback, explanation and hints do

Constraints¶

• Support was observed rather than assigned, so the form-level associations are not causal estimates
• Form–task fit varies: instructing suited coding whereas modelling suited writing

Target Learners¶

• Adult users of general-purpose LLM assistants in everyday coding and writing work

Target Learning Goals¶

• Preserving opportunities for users to reason, test ideas and construct understanding during everyday problem solving

Affordances¶

• Turn Level Informal Learning Signatures Framework

Related Strategies¶

• AI-assisted learning systems should scaffold cognitive development rather than replace it, and require active student contribution to preserve ownership
• Design AI coaching systems to preserve user agency, recognize vulnerability, and stop sycophantic encouragement during maladaptive thought processes
• Design AI learning assistants to make the implications of AI use visible rather than restricting assistance
• Pursue research yielding conditions and parameters for when to use information giving versus information withholding interactions
• Design AI configurations that preserve generative work: feedback-only audiences, generate-then-compare, and calibrated teaching interactions

Examples¶

-

Key Sources¶

• Chen, Z., Li, H., Xiao, Z., Qu, H., & Xie, X. (2026). Informal learning emerges in everyday human–LLM interaction. arXiv. https://arxiv.org/abs/2607.17643. doi:10.48550/arxiv.2607.17643