Interpersonal neural synchronization underlies interactive concept learning in older adults

Abstract

Lifelong learning is essential for active aging, yet interactive learning remains underexplored in older adults. Using fNIRS-based hyperscanning, this study investigated instructor-learner neural interactions in older learners during a concept learning task. Study 1 compared passive and active learning approaches in younger and older learners, while Study 2 compared intergenerational (younger instructors) versus intragenerational (older instructors) learning among older learners, using the more effective approach from Study 1. Older adults performed better with passive learning (Study 1) and when taught by younger instructors (Study 2). Passive learning increased frontopolar interpersonal neural synchronization (INS), while intergenerational learning enhanced INS in both frontopolar and dorsolateral prefrontal regions. Intersubject representational similarity analysis (IS-RSA) showed that learners with similar performance exhibited greater neural similarity in the frontopolar cortex. These findings reveal the neural basis of interactive learning in aging and offer practical guidance for optimizing educational practices in older populations.

Introduction

Lifelong learning, the continuous pursuit of knowledge and skills throughout life¹, is emerging as a global phenomenon. For older adults, increased longevity, self-awareness, and social support create more opportunities to access learning^2–4. Senior universities are beginning to offer classroom-based learning programs involving higher cognitive demands, such as courses in financial literacy, digital skills, and social psychology^5–7. This type of learning, involving critical cognitive functions and social engagement, plays a pivotal role in promoting active aging^8–10. Interactive learning involves acquiring information by interacting with others (e.g., an instructor), as opposed to through direct personal experience or self-directed learning^11,12. However, while interactive learning in children, adolescents, and younger adults has been well studied^13,14, research on older adults is scarce, particularly regarding how to facilitate interactive learning in real-life contexts from the perspective of “how to learn” and “who to teach.”

The question of “how to learn” examines which learning approach is more effective in supporting interactive learning among older adults. Active and passive learning approaches are widely used in pedagogical practice, particularly in classroom-based interactive learning, which involves understanding overarching concepts and theories with the help of instructors^15,16. In passive learning, learners receive explanatory information from the instructor through listening or reading, with minimal reciprocal communication. In contrast, active learning features bidirectional exchanges between the instructor and the learner, such as question-and-answer (Q&A) sessions and feedback^17–20. According to the interactive-constructive-active-passive theory, two-way communication fosters deeper cognitive engagement, thereby enhancing learning outcomes^21,22. Empirical studies with younger adults have consistently shown that active learning improves learning performance in various domains^16,23–25. However, given the general cognitive decline experienced by older adults, the effectiveness of an active approach may not be equivalent to that observed in younger adults.

Aging is associated with cognitive decline across multiple domains related to interactive learning, including reductions in working memory capacity, processing speed, and inhibitory control^26–30. According to cognitive load theory, learning tasks that impose high cognitive demands tend to impair performance³¹. This effect is more pronounced in older adults, whose performance tends to deteriorate more steeply than that of younger adults under increasing cognitive load^32–35. Within interactive learning contexts, active learning involves reciprocal information exchange and requires learners to both comprehend and respond, thereby imposing greater cognitive demands than passive learning^21,36. While active learning with its emphasis on instructor–learner interactions has been shown to benefit younger adults, it may not necessarily facilitate interactive learning for older adults.

In addition to the learning approach itself, the question of “who to teach” concerns which instructors are better suited to implement it in order to maximize interactive learning among older adults. Recent studies have compared the outcomes of intragenerational learning, which refers to older instructors teaching older learners (old-teach-old), with those of intergenerational learning, which refers to younger instructors teaching older learners (young-teach-old)^37–41. Empirical findings have been mixed. Older adults demonstrated greater learning acquisition under intragenerational instruction, particularly for more demanding tasks³⁸. However, other studies have emphasized the benefits of intergenerational learning for older adults^42–44. Additional evidence from cognitively engaging learning tasks is needed to further clarify which instructor–learner dyad most effectively supports interactive learning in older adults.

One possible explanation for these mixed findings is the perceived match between instructor attributes (e.g., gender, age) and the learning content^45,46. For example, older adults tend to benefit more from older instructors when learning health-related topics, but perform better with younger instructors when acquiring technology-related skills³⁹. In the present study, learning new concepts relies on cognitive functions that are often stereotypically perceived as declining in older adults^47–49. From this perspective, younger instructors may be perceived as more competent and thus a better match, which reduces psychological distance and facilitates interpersonal coordination⁵⁰. The interactive learning can be viewed as a joint action^51,52, requiring instructors and learners to coordinate toward a shared goal. According to interpersonal synergy theory⁵³, such coordination emerges through reciprocal adjustment and role differentiation, with younger instructors guiding and older learners providing feedback that instructors adapt to. Based on this framework, we hypothesize that older learners will achieve better outcomes in interactive concept learning when taught by younger rather than older instructors, and that intergenerational dyads will exhibit learner-to-instructor directional influence in neural coupling.

There are unique difficulties in solving the learning puzzle in later life. First, most existing studies on lifelong learning in older adults have adopted a phenomenological perspective^54–56, leaving the underlying neural mechanisms largely unexplored. Second, learning and teaching are two sides of the same coin, forming a dynamic and intertwined process that makes examining brain mechanisms challenging. Functional near-infrared spectroscopy (fNIRS) hyperscanning provides a useful approach, as it simultaneously records brain activity from two interacting individuals in real time. A growing body of research has revealed that interpersonal neural synchronization (INS), the coupling of neural processes across brains, supports dyadic interaction through shared attention, mutual prediction, and co-regulation^57,58. In interactive learning contexts, INS predicts successful performance in concept learning²⁵, numerical reasoning⁵⁹, and theory comprehension⁶⁰. Given that the prefrontal cortex (PFC) is critically involved in these processes^61–63, the present study focused on this region. In addition to INS, we employed intersubject representational similarity analysis (IS-RSA)⁶⁴, an extension of traditional representational similarity analysis (RSA)⁶⁵, to examine whether learners with similar behavioral outcomes also display analogous neural response patterns^61,62, thereby linking inter-individual differences in performance with neural activity.

The present study addresses two key questions of “how to learn” and “who to teach,” with a particular focus on the underlying neural mechanisms of interactive learning in aging. To this end, we designed two fNIRS hyperscanning studies using concept learning tasks. Study 1 examined how different learning approaches affect learning outcomes in younger and older adults. Study 2 focused on older adults and compared intergenerational and intragenerational learning. We hypothesized that (a) older adults would benefit more from passive than active learning due to reduced cognitive resources, (b) younger instructors would be more effective than older instructors in supporting older learners, and (c) enhanced instructor–learner neural synchrony, particularly learner-to-instructor influence, would underlie these behavioral advantages. Together, these studies aim to identify both behavioral and neural mechanisms that optimize interactive learning in later life.

Results

Study 1: How to learn through passive versus active learning

Study 1 employed a 2 (learner age: younger vs. older) × 2 (learning approach: passive vs. active) mixed design to examine how instructional format affects learning outcomes across age groups.

Passive learning improves outcomes for older learners

A repeated-measure analysis of variance (ANOVA) on learning outcomes revealed a significant main effect of learner age (F_(1,56) = 107.30, p < 0.001, η_p² = 0.66), with younger adults performing better than older adults. The interaction between learner age and learning approach was also significant (F_(1,56) = 44.97, p < 0.001, η_p² = 0.45). Post hoc tests showed that passive learning was more effective than active learning for older learners (t₍₃₀₎ = 7.95, p < 0.001, Cohen’s d = 1.43), whereas the reverse was observed for younger learners (t₍₂₆₎ = −4.02, p < 0.001, Cohen’s d = −0.77). There was no significant main effect of learning approach (F_(1,56) = 0.75, p = 0.39, η_p² = 0.013; Fig. 1a).

Fig. 1. Effects of learning approach and learner age on learning outcomes and instructor–learner INS.

a Learning outcomes across all conditions, showing a significant interaction effect between learning approach and learner age. b Instructor–learner INS in the FPA (CH12 and CH19) as modulated by learning approach and learner age. Density plots show null distributions of INS from 1000 pseudo dyads. The green line marks the real dyad mean INS, which exceeded the 95% CI (gray dashed lines), indicating a significant effect. c F-maps from a mixed-design ANOVA with learning approach and learner age as independent variables, revealing frontopolar channels with significant effects. OL old learner, YL young learner, FPA frontopolar area. *p < 0.05, **p < 0.01, ***p < 0.001.

Instructor–learner INS supports effective passive learning in older adults

A repeated-measure ANOVA revealed a significant main effect of learner age on task-related INS at 0. 228–0.484 Hz (CH2, 14, 17, 21, 22; ps < 0.05) and 0.057–0.064 Hz (CH5, 9, 10, 19; ps < 0.05). A significant interaction between learner age and learning approach emerged in the frontopolar area at 0.228–0.484 Hz (CH12: F_(1,56) = 4.80, p = 0.03, η_p² = 0.08; CH19: F_(1,56) = 4.54, p = 0.04, η_p² = 0.08; Fig. 1b, c). Post hoc tests for CH12 indicated that active learning elicited higher INS than passive learning in younger learners (t₍₂₆₎ = −2.00, p = 0.05, Cohen’s d = −0.39), whereas no significant difference was observed in older learners (t₍₃₀₎ = 1.09, p = 0.28, Cohen’s d = 0.20). In CH19, passive learning resulted in higher INS than active learning in older learners (t₍₃₀₎ = 2.00, p = 0.055, Cohen’s d = 0.36), but no significant difference was found in younger learners (t₍₂₆₎ = −1.21, p = 0.24, Cohen’s d = −0.23). Permutation analysis confirmed that real dyads showed greater INS than pseudo dyads in the frontopolar area (ps < 0.05, Fig. 1b).

Pearson correlations further revealed that frontopolar INS during passive learning positively correlated with learning outcomes for older learners (r = 0.50, p = 0.01; Fig. 2a), whereas no such correlation was found in the active learning (r = 0.33, p = 0.09). For younger learners, INS was not significantly associated with performance in either condition (ps > 0.05; Fig. 2b).

Fig. 2. Correlation between INS and learning outcomes.

a Among older learners, INS significantly predicted learning outcomes in the passive learning condition. b Among younger learners, no significant correlation between INS and learning outcomes was observed in either condition. OL old learner, YL young learner. *p < 0.05, **p < 0.01, ***p < 0.001.

Neural-behavioral correspondence in passive learning

IS-RSA was used to examine whether similarity in learning performance corresponded with similarity in neural responses in the frontopolar area. Significant correlations were observed for older learners under the passive condition (CH11: r = 0.23, p < 0.001; FDR-corrected; Fig. 3a–c). Other significant channels were presented in Supplementary Figs. 1–3 and Supplementary Data 2. No significant correlations emerged for older learners in the active condition or for younger learners in either condition (ps > 0.05).

Fig. 3. Neural-behavioral alignment in the frontopolar area during passive learning among older adults.

a Behavioral and neural dissimilarity matrix derived from IS-RSA for older learners and younger learners under passive and active learning conditions. A significant neural-behavioral correlation was observed only in the OL passive condition (r = 0.23, p < 0.001). Color scales indicate pairwise dissimilarity (Euclidean distance). b Brain localization of CH11 in the frontopolar area. c Null distributions of IS-RSA r-values generated from 1000 permutation tests for each condition. The red line represents the observed r value in the OL passive condition, which lies outside the 95% confidence interval (dashed lines), confirming statistical significance. OL old learner, YL young learner.

Cognitive load reflected in behavioral and neural measures

Task completion time and task-related oxyhemoglobin (HbO) increases were examined as indices of cognitive load. A paired-sample t-test revealed a significant effect of learning approach. Older adults required more time to complete active learning compared to passive learning (t(30) = −2.11, p = 0.04, Cohen’s d = −0.38). Region-based analysis further showed that passive learning elicited focal activation in the task-relevant frontopolar region (CH11: t(30) = 2.13, p = 0.04, Cohen’s d = 0.38; CH13 & 15 see Supplementary Data 3). By contrast, the active condition did not exhibit this localized activation pattern, suggesting a more distributed neural response.

Study 2: Who to teach through intergenerational versus intragenerational learning

Study 2 focused on older learners, comparing the effectiveness of intergenerational and intragenerational instructor–learner dyads using the passive instructional format identified as more effective in Study 1.

Intergenerational instruction enhances older adults’ learning

An independent samples t-test revealed a significant main effect of instructor–learner dyad (t₍₆₁₎ = 3.64, p < 0.001, Cohen’s d = 0.93; Fig. 4a). Older learners performed significantly better when paired with younger instructors compared to when learning with older instructors.

Fig. 4. Effects of instructor-leaner dyad on learning outcomes and INS.

a Learning outcomes were significantly higher in the intergenerational learning group compared to the intragenerational learning group. b INS in the dlPFC (CH8) and FPA (CH19) was significantly greater in the intergenerational learning group. The density plot shows the distribution of INS values from 1000 pseudo dyads generated through permutation tests. The green line indicates the mean INS of real dyads, confirming dyad-specific neural synchronization. c Correlation between learning outcomes and INS. A significant positive association was observed in the intergenerational learning group (r = 0.56, p < 0.001), but not in the intragenerational learning group. d T-maps from independent-sample t-tests comparing INS between intergenerational learning and intragenerational learning groups. Inter- intergenerational, Intra- intragenerational, dlPFC dorsolateral prefrontal cortex, FPA frontopolar area. *p < 0.05, **p < 0.01, ***p < 0.001.

Instructor-leaner INS supports effective intergenerational learning for older adults

An independent samples t-test revealed a significant main effect of instructor–learner dyad on INS in the dorsolateral prefrontal cortex (dlPFC; CH8; t₍₆₁₎ = 3.80, p < 0.001, Cohen’s d = 0.97) and frontopolar area (CH19; t₍₆₁₎ = 3.24, p = 0.002, Cohen’s d = 0.83; Fig. 4b, d) within 0.228–0.484 Hz frequency band. Specifically, the intergenerational learning group exhibited higher INS than the intragenerational learning group. A permutation analysis further confirmed that the enhanced INS observed in intergenerational learning was specific to real dyads (ps < 0.05; Fig. 4b).

Moreover, a significant positive correlation was found between learning performance and the frontopolar INS in the intergenerational learning group (r = 0.56, p = 0.01), whereas no such correlation was observed in the intragenerational learning group (ps > 0.05; Fig. 4c).

Neural-behavioral correspondence in intergenerational learning

IS-RSA was used to examine the relationship between learning performance and neural signals in the frontopolar area. Significant correlations were found in the intergenerational learning group (CH11: r = 0.23, p < 0.001; FDR-corrected; Fig. 5a–c). Other significant channels were presented in Supplementary Figs. 4–8 and Supplementary Data 4. No significant correlations were observed for older learners in the intragenerational learning group (ps > 0.05).

Fig. 5. Neural-behavioral alignment in the frontopolar area during intergenerational learning.

a Behavioral and neural dissimilarity matrix for intergenerational learning and intragenerational learning groups. A significant neural-behavioral correlation was observed only in the intergenerational learning group (r = 0.23, p < 0.001). The color scale represents the dissimilarity values (Euclidean distance). b Brain localization of CH11 in the FPA. c Null distributions of IS-RSA r-values from 1000 permutation tests. The red line indicates the observed r value for the intergenerational learning group, which falls outside the 95% confidence interval (dashed lines), confirming statistical significance. Inter- intergenerational, Intra- intragenerational, FPA frontopolar area.

Learner-driven effects in intergenerational dyads

We conducted Granger causality analysis to examine the directionality of INS in intergenerational learning. Results showed that learner-to-instructor influence was significantly stronger than the reverse direction (t₍₃₀₎ = −4.00, p < 0.001, Cohen’s d = −0.72; Fig. 6). By contrast, intragenerational dyads exhibited no significant directional difference (t₍₃₁₎ = 0.42, p = 0.68, Cohen’s d = 0.07).

Fig. 6. Directional coupling between the learner and the instructor during learning.

Granger causality analysis revealed a significant directional effect in the intergenerational learning group: information flow from the learner to the instructor (L → I) was stronger than the reverse (I → L), indicating learner-driven influence. No directional difference was found in the intragenerational learning group. The right panel depicts this connectivity pattern in the learner’s and instructor’s dlPFC (CH8). Inter- intergenerational, Intra- intragenerational, dlPFC dorsolateral prefrontal cortex. *p < 0.05, **p < 0.01, ***p < 0.001.

Discussion

Interactive learning plays an important role in promoting active aging. The present study examined how older adults learn in naturalistic instructional settings. We focused on two key questions of “how to learn” (passive versus active) and “who to teach” (intergenerational versus intragenerational). Results showed that older adults performed better when learning passively (Study 1) and with younger instructors (Study 2). Using fNIRS-based hyperscanning, we observed that passive learning elicited greater instructor–learner INS in the frontopolar cortex, while intergenerational learning enhanced INS in both the frontopolar cortex and dlPFC. Stronger INS in these regions was further associated with better learning outcomes. Under these optimal conditions, IS-RSA revealed greater neural pattern similarity in the frontopolar region among older adults with similar learning performance. Collectively, these findings highlight the neural mechanisms underlying interactive learning in aging and offer valuable insights for instructional design.

When studying “how to learn,” Study 1 found that older adults benefited more from the direct, explanatory instruction typical of passive learning, whereas younger adults performed better with active engagement. Older adults also required more time to complete active learning than passive learning, even though task complexity and concept difficulty were matched. These results support cognitive load theory. Active learning requires reciprocal interactions and greater real-time cognitive processing, while passive learning emphasizes one-way information transmission³⁶. Older adults experience declines in working memory, processing speed, and attentional control^27,28, which limit their ability to manage the demands of active learning. They also show reduced social-cognitive capacities, such as theory of mind and perspective-taking^63,66. These deficits may hinder spontaneous interaction and flexible role-switching during active engagement. Collectively, these findings suggest that passive learning is more effective for older adults, as it allows them to allocate limited resources more efficiently to task processing.

At the neural level, Study 1 revealed increased instructor–learner INS in the frontopolar area during passive learning, and this INS was positively correlated with learning outcomes. Passive learning was also characterized by localized activation in the frontopolar area, whereas active learning elicited more widespread neural recruitment. These findings align with cognitive load theory. The frontopolar region plays a critical role in interactive learning because it supports information integration and attention control^67–69. During passive learning, older adults can focus their limited cognitive resources on the instructor’s input. This focus leads to robust activation in the task-related frontopolar area and fosters stronger instructor–learner neural coupling. Together, these findings suggest that frontopolar neural synchronization represents a potential neural mechanism through which a passive approach facilitates effective learning in older adults. Nevertheless, interactive learning is inherently social and unfolds through real-time coordination between instructors and learners. The next important question, therefore, concerns who is better suited to implement the more effective approach.

Regarding “who to teach,” Study 2 demonstrated that older adults learned more effectively when instructed by younger rather than same-age partners. In line with our hypothesis, this intergenerational advantage likely reflects a perceived match between instructor attributes and the learning task. Older adults may view younger instructors as more competent for concept learning^{47–49,70–72}, a perception that reduces psychological distance and promotes smoother coordination. These results are consistent with the view of interactive learning as a joint action^51,52, in which shared representations and prediction make intergenerational dyads more effective in coordinating toward a common goal. According to interpersonal synergy theory⁵³, such coordination relies on reciprocal compensation. Younger instructors dynamically adjust to learners’ feedback, while older learners concentrate on internal processing. This complementary role distribution increases interaction efficiency and ultimately enhances learning outcomes in intergenerational settings.

At the neural level, older adults showed greater instructor–learner INS in the dlPFC and frontopolar cortex when taught by younger instructors. The dlPFC supports mentalizing and motivational regulation^73–75, while the frontopolar cortex contributes to integrative reasoning and sustained attention^67–69. Synchrony in these regions reflects stronger cognitive and social alignment, and critically, higher frontopolar INS was correlated with better learning outcomes. From a joint action perspective^51,52, such neural synchrony reflects the alignment of shared goals and predictions that support effective coordination. These findings indicate that neural alignment serves as a key mechanism through which intergenerational matching enhances learning. Granger causality analysis further revealed stronger learner-to-instructor directional coupling in intergenerational dyads, consistent with prior findings on asymmetry in teaching-learning interactions^76–78. This learner-led influence suggests that younger instructors dynamically adjusted their teaching in response to learners’ signals. Such reciprocal compensation, as articulated by interpersonal synergy theory, supports real-time coordination and sustained engagement, thereby reinforcing the learning benefits of intergenerational pairing.

IS-RSA revealed significant correlations between neural response similarity in the frontopolar area and learning performance among older learners, specifically under passive (Study 1) and intergenerational learning (Study 2). The frontopolar cortex plays a well-established role in attention, reasoning, and information integration—processes central to interactive learning^67–69,79. These correlations, therefore, suggest that older learners in optimal learning contexts adopted more consistent cognitive strategies. In passive learning tasks, the structured format and lower cognitive demands may have promoted uniform attention and deeper processing, leading to similar neural activation patterns. Likewise, in intergenerational learning, perceiving younger instructors as more effective may have encouraged sustained attentional and integrative engagement. This, in turn, aligned neural responses across learners. Together, these findings highlight consistent cognitive engagement in the frontopolar cortex as a core neural mechanism supporting effective interactive learning in older adults.

However, this study has several limitations that may constrain the generalizability of its findings. First, given the Confucian tradition of lecture-based learning in China, the observed pattern may be culture-specific. Future research should employ cross-cultural designs to test generalizability. Second, although learning and testing were paper-based, the fNIRS apparatus and laboratory environment may still induce tension for older adults. Adaptation sessions could be included in future studies to minimize such bias. Third, our one-to-one laboratory setting does not fully reflect real-world lifelong learning contexts, where group instruction and peer interactions are common. Future research should examine multi-learner settings in senior universities or community education programs^80,81. Fourth, although instructional variability was minimized through standardized training, residual differences in teaching style could not be eliminated. Future studies may address this issue by using standardized pre-recorded instruction or by modeling instructor variability statistically. Finally, the fNIRS setup captured only frontal cortical activity, leaving other cortical and subcortical regions unexplored^82–86. Future studies should integrate multimodal neuroimaging to provide a more comprehensive neural account.

In conclusion, this study provides converging behavioral and neural evidence that passive and intergenerational learning can significantly improve interactive learning outcomes in older adults. These benefits are associated with increased instructor–learner INS in key prefrontal regions, particularly the frontopolar and dorsolateral prefrontal cortices. These regions are known to support integrative reasoning and socio-cognitive engagement. By addressing the questions of “how to learn” and “who to teach,” our findings highlight the importance of instructional format and instructor attributes in optimizing learning for aging populations. Beyond theoretical insights, these results have practical implications for the design of aging-friendly education. Digital platforms and group-based learning interventions could be improved by incorporating structured, explanatory content and intergenerational instructional formats. Such adaptations could foster neural alignment and reciprocal coordination, ultimately promoting more effective and inclusive lifelong learning.

Methods

Study 1: How to learn through passive versus active learning

Participants

The sample size was determined via a prior power analysis using G^*Power⁸⁷, with α = 0.05, power = 0.95, and an expected effect size of d = 0.40 based on previous studies employing a similar intergenerational learning paradigm³⁸. The analysis suggested a minimum of 52 dyads would be required to detect a reliable interaction effect. A total of 62 dyads were recruited, exceeding the minimum requirement. This included 31 young learners and 31 older learners. The inclusion criteria for the older learners were: (1) aged 55 years or over; (2) without cognitive impairment, as indicated by a Mini-Mental State Examination⁸⁸ score of at least 26; (3) normal or corrected-to-normal vision with no speech or hearing difficulties. For both young and older learners, additional inclusion criteria included having no prior knowledge of social psychology and having no history of psychiatric or neurological illness. Two young learners did not complete the posttest, and two others were excluded due to poor fNIRS signal quality, leaving 27 young (age: M = 20.78, SD = 1.40; 21 females) and 31 older (age: M = 68.62, SD = 5.45; 18 females) learners for the final analyses.

Two additional young adults (age: M = 23; one male, one female) were recruited and trained as instructors. Each learner was matched with an instructor of the same sex, resulting in a total of 58 dyads. The study was approved by the Ethics Committee of the Faculty of Psychology at Southwest University (No. H23215) and pre-registered on AsPredicted (#166392).

Materials

A concept learning task was employed. The concepts used in the present study underwent a process of selection, construction, and assignment. First, 13 social psychological concepts were selected from an online course at the Seniors University of China (https://lndx.edu.cn/). Eight of these were chosen after 32 non-psychology students rated their familiarity with and perceived difficulty of each concept, including availability heuristic, mental set, fundamental attribution error, door-in-the-face effect, representativeness heuristic, projection effect, foot-in-the-door technique, and hindsight bias. Second, each concept was structured with one term, one definition, one illustrative example, and supplementary explanations from psychology textbooks^89,90 (see Supplementary Table 1). Finally, the eight concepts were randomly divided into two sets of four and matched for familiarity (t₍₃₁₎ = 1.24, p = 0.23) and difficulty (t₍₃₁₎ = −0.24, p = 0.81).

The learning approach (passive versus active) was manipulated as a within-subject factor. Each participant learned two sets of concepts, with one set presented in the passive format and the other in the active format. The learning approach assigned to each set of concepts was randomized, and the order of the two learning blocks was counterbalanced across participants. In passive learning, concepts were conveyed directly from the instructor to the learner (e.g., “Availability heuristic means the tendency to estimate the likelihood that an event will occur by how easily instances of it come to mind”). In contrast, active learning involved a Q&A format that encouraged learner participation (e.g., “What do you think the availability heuristic is? Please share your understanding”). Each concept was presented only once in the format assigned to its corresponding learning block. To ensure consistency across dyads, detailed learning scripts were developed to standardize the instructional process (see Supplementary Table 2).

The two instructors learned the eight concepts and two scripts corresponding to different learning approaches. They then practiced the instructional process together, adhering to predefined criteria: duration, speech rate, and script consistency⁹¹. The experiments were conducted only after these criteria were successfully met.

Procedures

To control for the effects of interpersonal distance, the learner and the instructor were seated face-to-face at a 90° angle and 80 cm apart (Fig. 7a). The task consisted of two blocks, each comprising a 180-s resting phase to establish a baseline, followed by a 600 to 1000-s learning phase (Fig. 7b). A 1-min interval was allotted between the two blocks. Tests were administered before and after the task to evaluate learners’ understanding of the social psychological concepts. During the resting phase, both participants were instructed to relax, remain still, and close their eyes. During the learning phase, dyads completed the concept learning task using both passive and active learning approaches in a counterbalanced order (Fig. 7c). Written informed consent was obtained from all individuals shown in the images for publication. Facial features have been blurred to ensure anonymity. The experimental procedure was recorded using a digital video camera (Sony, GZ-HM550) for process checking.

Fig. 7. Experimental design, instructional procedure, and optode configuration.

a Experimental setup of Study 1. The left panel shows the older learner group, and the right panel shows the younger learner group. In both groups, the learner (left) and instructor (right) were seated face-to-face at a 90° angle, approximately 80 cm apart. b Experiment procedure of Study 1, including pretest, two blocks of resting (180 s) and learning (600–1000 s) phases, a break between blocks, and posttest. c Instructional procedures for passive and active learning approaches. These were identical across both studies, with Study 2 applying only the instructional format found to be more effective for older adults in Study 1. d FNIRS optode configuration and regions of interest, consistent across both studies. PTr pars triangularis Broca’s area, dlPFC dorsolateral prefrontal cortex, FPA frontopolar area.

fNIRS data acquisition

A Shimadzu multichannel high-speed continuous wave system (FORE3000, Shimadzu Corp., Kyoto, Japan) was used to record the hemodynamic signals from each dyad at a sampling rate of 7.69 Hz using three wavelengths: 780 nm, 805 nm, and 830 nm. Oxyhemoglobin (HbO), deoxyhemoglobin (HbR), and total hemoglobin (HbT) concentrations were determined using the modified Beer-Lambert law. The study focused specifically on analyzing HbO concentration due to its high signal-to-noise ratio^92,93. A 3 × 5 optode array was positioned over the PFC of both the instructor and the learner in each dyad, a region known to be associated with interactive learning (Fig. 7d). Each participant used eight emitters and seven detectors, resulting in 22 measurement channels (CH). The bottom row of the optode was centered at Fpz according to the international 10/20 system⁹⁴. The probe set was positioned so that its bottom touched the tops of the participants’ eyebrows, with the middle row aligned with the nose. The emitter-detector pairs were spaced 3 cm apart to allow cerebral blood volume measurements at a depth of 2–3 cm from the skin. The measurement channels were mapped onto cortical regions using the virtual registration method⁹⁵ (Montreal Neurological Institute coordinates see Supplementary Table 3), and were then assigned to the corresponding Brodmann area (BA). Channels within the same area were grouped to create regions of interest (Supplementary Table 4).

Behavioral data analyses

The learners’ knowledge of the concepts was assessed before and after the task using tests consisting of eight subjective and eight objective questions (see Supplementary Table 5). The subjective questions, which were asked by the experimenter, required learners to explain the meaning or their understanding of the concepts they had learned. The learners answered the objective questions in a multiple-choice format.

Two independent raters, who were blinded to the experimental conditions, scored the learners’ responses. For each learning approach, the subjective questions were assigned a maximum score of 31 points, and the objective questions 4 points, resulting in a total possible score of 35. Learning outcomes were calculated as the difference between pretest and posttest scores. A repeated-measure ANOVA was conducted on the learning outcomes, with learning approach (passive vs. active) as the within-subjects variable and learner age (older vs. younger learner) as the between-subjects variable.

fNIRS data analyses

Learner brain activation was first analyzed. Raw data from the resting (180 s) and learning (600–1000 s) phases were preprocessed using the Homer2 toolbox in MATLAB⁹⁶. A principal component analysis spatial filter was applied to remove physiological noise (e.g., blood pressure, respiration, blood flow) using a variance threshold that removes up to 80% of components^97–99. Motion artifacts were corrected using the correlation-based signal improvement method¹⁰⁰. β values were then estimated for each phase using a general linear model (GLM). A one-sample t-test comparing learning and resting phases revealed significantly greater activation during learning in the dlPFC (CH13: t₍₁₇₉₎ = 3.16, p = 0.01, Cohen’s d = 0.24; CH18: t₍₁₇₉₎ = 2.94, p = 0.02, Cohen’s d = 0.22) and frontopolar area (CH11: t₍₁₇₉₎ = 3.60, p = 0.005, Cohen’s d = 0.27; CH15: t₍₁₇₉₎ = 4.42, p < 0.001, Cohen’s d = 0.33; Fig. 8b).

Fig. 8. FNIRS data analysis workflow.

a INS analysis pipeline. HbO concentration time series from each CH of both the learner and the instructor were extracted and preprocessed. WTC was then applied to compute instructor–learner INS, generating a coherence matrix for each dyad. Warmer colors represent higher coherence values. b Activated CH during the learning tasks. CH with significant increases in coherence during learning, compared to rest, are shown. The color scale represents FDR-corrected p values. c Frequency range of interest. Significant frequency bins were identified using statistical testing, with color indicating FDR-corrected p values. d Permutation analysis. Real dyads (left) were compared to pseudo dyads (right) generated by randomly re-pairing time series from different individuals (e.g., sub 1 was paired with sub 5). e IS-RSA. Neural (left) and behavioral (right) dissimilarity matrices were constructed by computing pairwise Euclidean distances between learners. Spearman’s correlation coefficient (r) was used to assess the correspondence between the two matrices. Higher values represent greater dissimilarity.

INS analysis was conducted to examine the underlying neural mechanisms. The raw data underwent the same preprocessing pipeline as the learner brain activation analysis. HbO time series were extracted from both the learner and the instructor. Wavelet transform coherence (WTC) analysis was applied to compute instructor–learner INS for each dyad, generating coherence value matrices (Fig. 8a)¹⁰¹. Paired-sample t-tests were used to compare INS between the resting and learning phases, with False Discovery Rate (FDR) correction applied (p < 0.05) to identify the frequency range of interest (FOI). A significant increase in INS was observed at 0.091–0.144 Hz, 0.228–0.484 Hz, and 0.057–0.064 Hz (Fig. 8c), consistent with the frequency band reported in a previous fNIRS-based hyperscanning study¹⁰².

For each identified CH and FOI, INS values were averaged across time and frequency for both the resting and learning phases. Task-related INS was calculated by subtracting resting-phase coupling from learning-phase coupling. A Fisher z-transformation was applied to normalize the distribution for subsequent analyses. The resulting values were entered into a mixed-design ANOVA with learning approach (within-subjects: passive vs. active) and learner age (between-subjects: older vs. younger learner) as independent variables. F-maps were visualized using BrainNet Viewer software¹⁰³. Parallel analyses were performed for each FOI. To determine whether group differences in INS were specific to real dyads, pseudo dyads were generated by randomly re-pairing HbO time series from the 58 real dyads (Fig. 8d)¹⁰⁴. INS was computed for each pseudo-dyad and repeated 1000 times. Statistical significance was assessed by comparing real dyad INS values against the null distribution from pseudo dyads (ps < 0.05). To assess the functional significance of INS, Pearson’s correlation coefficients were computed between learning outcomes and INS values in the significant channels identified in the ANOVA.

To further examine the relationship between learning performance and brain activity, we conducted IS-RSA (Fig. 8e)^64,99. A behavioral dissimilarity matrix was constructed by calculating the Euclidean distance between participants’ learning scores. A neural dissimilarity matrix was then generated by applying a fast Fourier transformation to the preprocessed HbO signals to obtain the power spectrum. The amplitude of signal fluctuations, defined as the square root of the power spectrum, was averaged across the 0.01–0.5 Hz frequency range. Euclidean distances were computed between participants’ neural signals, with larger values in both matrices reflecting greater dissimilarity. Spearman’s correlation between the behavioral and neural dissimilarity matrices was then computed to assess their correspondence. A permutation test was used to evaluate the significance of the observed correlation by comparing it against a null distribution generated through 1000 iterations of random shuffling of the neural dissimilarity matrix⁶⁵.

Study 2: Who to teach through intergenerational versus intragenerational learning

Participants

A power analysis was conducted using G^*Power, specifying an alpha level of 0.05, a power of 0.95, and an expected medium effect size (d = 0.40), based on prior work employing a similar intergenerational learning paradigm³⁸. The analysis indicated that a minimum of 52 dyads was required to detect a reliable effect. A total of 63 older learners (age: M = 70.97, SD = 5.47; 36 females) were recruited from local communities, exceeding the minimum requirement. Inclusion criteria were consistent with those used in the “Participants” section of Study 1.

Additionally, two older adults (age: M = 65; one male, one female) and two young adults (age: M = 23; one male, one female) were recruited and trained as instructors. Thirty-one older learners were randomly assigned to a younger instructor (intergenerational learning group) of the same sex, and the remaining 32 to an older instructor (intragenerational learning group) of the same sex. The study was approved by the Ethics Committee of the Faculty of Psychology, Southwest University (No. H23215).

Materials

The same concept learning task used in Study 1 was employed in this study, as detailed in the “Materials” section of Study 1.

Procedures

Each dyad in Study 2 completed a single block, which exclusively employed the passive learning approach identified as more effective for older learners in Study 1. This block included a resting phase (180 s) followed by a learning phase (600–1000 s), during which all dyads completed the concept learning task.

fNIRS data acquisition

Hemodynamic signals from the PFC of both the instructor and the learner were recorded using the same procedure described in the “fNIRS data acquisition” section of Study 1.

Behavioral data analyses

Learners’ conceptual knowledge was assessed and scored as in Study 1. An independent-sample t-test was conducted to compare learning outcomes between intergenerational learning and intragenerational learning groups.

fNIRS data analyses

fNIRS data were processed following the same procedures used in Study 1. Additionally, Granger causality analysis was performed using the HERMES MATLAB toolbox to determine the directionality of instructor–learner neural coupling within each dyad¹⁰⁵. Granger-causality values were calculated in both directions (i.e., from the instructor to the learner, and from the learner to the instructor), and compared using a paired-sample t-test. The same validation procedures were applied to assess the specificity of INS and IS-RSA results.

Author contributions

Z.W.L.: conceptualization, investigation, data curation, formal analysis, methodology, software, validation, visualization, writing—original draft, writing—reviewing and editing. Z.J.Z.: investigation, data curation. Y.C.L.: investigation, data curation. H.Z.X.: supervision, writing—review and editing. J.Y.: conceptualization, supervision, project administration, funding acquisition, writing—review and editing.

Data availability

The datasets analyzed during the current study are available in the OSF https://osf.io/34kqy/.

Code availability

The underlying code for this study is available on OSF and can be accessed via this link https://osf.io/34kqy/.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41539-025-00368-5.