Generative artificial intelligence (genAI) is rapidly transforming education, but not the cognitive mechanisms through which humans learn. However, genAI’s effects on learning are often discussed in terms of technological capabilities rather than the cognitive processes through which learning occurs. In this paper, I build on learner-centered accounts for evaluating genAI in education. Through an educational psychology and cognitive science lens, I argue that technologies can influence learning only insofar as they support or interfere with learners’ active cognitive processing. This perspective is particularly important because genAI can increasingly perform cognitive operations on learners’ behalf, such as generating explanations, organizing information, and solving problems. Such capabilities may indeed increase efficiency, but they may also deprive learners of cognitive activities essential for learning. Therefore, the key question for genAI in education is not whether genAI improves learning, but whether the cognitive operations delegated to genAI are operations that learners need to perform themselves for the intended learning to occur. The same genAI function may therefore support learning under one learning objective and undermine it under another. Thus, performing better with genAI support should not be mistaken for learning that remains once the support is removed. Moreover, explicit learning objectives do not capture all cognitive, metacognitive, motivational, social, or developmental opportunities afforded by an activity. Educational psychology is uniquely positioned to determine which cognitive operations learners need to perform themselves and which may be delegated to genAI. It can also help identify what may be lost when such operations are outsourced.
Background Vedic Sanskrit recitation is an intensive oral tradition involving memorization, repeated retrieval, phonological precision, rhythmic production, and sustained attention, and we examined whether long-term training is associated with specific memory performance and intrinsic brain organization. Methods Thirty Vedic scholars and 30 non-Vedic men aged 21–28 years completed the PGI Memory Scale and resting-state functional MRI. Groups were compared on memory subtests, amplitude of low-frequency fluctuation (ALFF), local correlation (LCOR), functional-connectivity multivoxel pattern analysis (fc-MVPA) and ROI-to-ROI connectivity, adjusting for age and Raven score with false-discovery-rate (FDR) correction. Within scholars, training history was related to these measures. Results Scholars showed a large, selective advantage in immediate recall (adjusted difference = 2.51 points, 95% CI [1.68, 3.34]; adjusted d = 1.92; FDR-adjusted p = 9.4 × 10−7); no other subtest differed after correction. Scholars had greater ALFF in a cerebellar–posterior cluster (pFDR = .003), greater LCOR in cerebellar and postcentral–parietal clusters but lower LCOR in occipital and right anterior temporal clusters. fc-MVPA identified 10 clusters with distinct connectivity patterns, and ROI-to-ROI analysis a network with predominantly lower coupling in scholars (pFDR = .048). Within scholars, training variables were associated with ALFF, LCOR and fc-MVPA in thalamic, medial temporal, frontal and sensorimotor regions. No brain–behaviour or training–memory correlation survived correction. Conclusions Long-term Vedic oral education was associated with better immediate recall and distinct intrinsic organization of language- and memory-related brain systems, supporting further longitudinal study.
Background Vocabulary learning in real-world instructional settings requires coordinated engagement of language-selective and domain-general brain networks. However, it remains unclear how functional connectivity within and between these systems supports learning under varying levels of affective load and individual cognitive capacity in a naturalistic learning environment. Methods Thirty English learners (aged 18–28 years) and five instructors (aged 20–23 years) participated in this study. Functional near-infrared spectroscopy (fNIRS) was used to record cortical activity during a 30-minute English lesson. Physiological anxiety was indexed by heart rate variability, and prior proficiency was indicated by verbal fluency scores. Functional connectivity was estimated for the language network and the multiple-demand (MD) network to test how affective and cognitive factors shape neural recruitment. Results Anxiety produced distinct connectivity–performance profiles. Low-anxiety learners showed stronger within-language-network connectivity and greater vocabulary gains, whereas high-anxiety learners exhibited a reversed pattern, indicating disrupted engagement of language-selective regions. In contrast, strengthened MD–language connectivity predicted improved learning for high-anxiety learners, suggesting compensatory domain-general support. Proficiency further modulated these effects: high-proficiency learners demonstrated sharper shifts in connectivity–behavior relationships across anxiety levels, whereas low-proficiency learners showed weaker and less adaptively tuned patterns. Conclusion This study demonstrates that vocabulary learning is shaped by interactions among anxiety, prior proficiency, and brain network connectivity. Low anxiety supports efficient language network connectivity, whereas high anxiety shifts reliance toward language–MD network coupling as a compensatory mechanism. This shift depends on prior proficiency, highlighting dynamic neural reconfiguration during learning and the importance of considering affective and linguistic profiles in language education.
