Informal learning emerges in everyday human–LLM interaction

As LLMs become integral to daily tasks, a primary concern is that they may lead to cognitive offloading—where users delegate thinking to the machine, potentially eroding their own problem-solving capabilities. This study investigates whether everyday human-LLM interaction also serves as a site for informal learning. The researchers analyzed over 128,000 naturalistic conversations from public datasets (WildChat, LMSYS Chat, and ShareChat), focusing on coding and writing tasks. They translated learning-science constructs into turn-level behavioral signatures, categorizing user engagement into passive, active, and constructive levels, and assistant responses into scaffolded versus direct-answer support.

The study reveals that everyday LLM use is a genuine, albeit selective, space for informal learning. While cognitive engagement (general effort) appeared in nearly 32% of user turns, deeper constructive engagement—where users elaborate, test, or revise ideas—occurred in about 5% of turns. This constructive engagement was not random; it was ecologically organized, appearing more frequently in coding tasks (which often involve explicit error-checking) and in sustained, multi-turn conversations. Furthermore, the researchers found that assistant scaffolding (e.g., providing hints or explanations rather than just answers) consistently marked richer constructive participation, suggesting that the way an AI responds can preserve or enhance a user's cognitive agency.

These findings shift the focus of AI evaluation from simple answer-delivery efficiency toward the preservation of cognitive opportunities. By demonstrating that informal learning occurs naturally in everyday AI interactions, the study suggests that the educational value of future AI assistants lies in their ability to calibrate support—knowing when to provide an answer and when to step back to allow the user to reason, test, and build judgment. This perspective reframes AI as a potential partner in cognitive development rather than just a tool for task completion.