Cognitive and neural mechanisms of improving informal reasoning in human-GenAI interactive learning contexts: An fNIRS study

While generative artificial intelligence (GenAI) has advanced personalized interactive learning, the cognitive and neural mechanisms underlying learners' informal reasoning improvement remain underexplored. Thus, we conducted sliding-window correlation and k-means clustering to capture learners' dynamic functional connectivity (dFC) states, time-varying correlations among brain regions, and examined their correlations with the informal reasoning improvement. 78 participants completed a learning task under either a traditional search engine-supported interactive learning mode (TSE group) or a human-GenAI interactive learning mode (GenAI group). Functional near-infrared spectroscopy (fNIRS) was employed to measure cortical hemodynamic responses from the medial and dorsolateral prefrontal cortex and the right temporo-parietal regions. The results showed that the two groups demonstrated no significant difference in informal reasoning improvement. Moreover, both groups presented a series of dynamic dFC states throughout the learning process, and the properties of these dFC states were similar across groups. Nevertheless, the neural correlates underlying informal reasoning improvement differed across groups. In the GenAI group, State 1, associated with goal-directed sense-making processes, showed a significant positive correlation with informal reasoning improvement. In contrast, in the TSE group, State 3, associated with the retrieval and extraction of task-relevant information, was significantly positively correlated with informal reasoning improvement. These findings deepen our understanding of brain dynamics in learning and uncover shared and distinct neural mechanisms that characterize the GenAI-supported and traditional search engine-supported interactive learning modes.