Generative AI (GenAI) promises to reshape education, yet, beyond gains in efficiency and scalability, crucial questions remain regarding its deeper cognitive, epistemic, socio-emotional and ethical implications for learners. This special section collects emerging empirical evidence from diverse educational settings to examine how interactions with advanced GenAI systems—ranging from intelligent tutors and conversational agents to AI-generated feedback providers—shape learners' cognitive engagement, epistemic agency, metacognitive regulation and emotional relationships with technology. While studies reveal evidence of positive effects in personalised guidance, collaborative inquiry and enhanced self-reflection, findings also highlight risks such as diminished epistemic vigilance, superficial learning and emotional dependence on AI interlocutors. These insights advocate for an educational re-imagining that actively experiments with novel AI technologies while continuously gathering and critically evaluating evidence—envisioning education not merely as a spectrum ranging from fully human- or AI-led activities but also as genuinely collaborative endeavours that balance human intelligence and AI capabilities in meaningful and ethically grounded partnerships. Looking ahead, in this editorial, we propose an interdisciplinary research roadmap that urges deeper examination of the cognitive and neural consequences associated with sustained AI exposure, together with the development of comprehensive frameworks to cultivate AI literacy and evaluative judgement. It also encourages rigorous assessment of the long-term ethical implications of increasingly blurred human-AI boundaries. Ultimately, advancing education in GenAI-infused contexts requires proactive engagement with these emerging complexities, so that teachers and learners not only survive in an AI-infused future but also creatively, critically and ethically co-author it.
Large language model (LLM)-based conversational agents (CAs), with their advanced generative capabilities and human-like conversational interfaces, can serve as reading partners for children during dialogic reading and have shown promise in enhancing children's comprehension and conversational skills. However, there is limited research on the efficacy of LLM-based bilingual CAs in children's language acquisition in English as a Foreign Language (EFL) contexts. This randomized controlled trial study investigated the effectiveness of LLM-powered CAs compared with traditional parent–child shared reading in promoting engagement and improving learning outcomes among children with EFL. An interactive e-book featuring a LLM-powered CA was developed to engage children in dialogic reading through questioning and scaffolding. Sixty-seven children, aged 5 to 8, were randomly assigned to either an experimental (AI-led) group or a control (parent-led) group. The study found that children in the experimental group outperformed the control group in reading comprehension, with comparable benefits in vocabulary acquisition and story retelling, both immediately and in delayed tests. In the meantime, this study unpacks children's different engagement patterns when reading with the CA versus reading with their parents. Children reading with the CA demonstrated higher behavioural engagement and visual attention, while those in the parent-led group showed greater affective engagement and narrative-relevant vocalizations. The findings highlighted insights into the potential of LLM-powered CAs in children's language acquisition and suggested key design implications for developing better CAs for children from multilingual backgrounds.
Understanding learners' preferences in educational settings is crucial for optimizing learning outcomes and experience. As artificial intelligence (AI) becomes increasingly integrated into educational contexts, it is crucial to understand learners' preferences between AI and human tutors to support their learning. While AI demonstrates growing potential in education, the phenomenon of algorithm aversion, which is a tendency to favour human decision making over algorithmic solutions, requires further investigation. To explore this issue, an experiment involving 114 university students was conducted to measure learners' preferences for different feedback sources before and after exposure to one of four conditions: no feedback, human tutor feedback, ChatGPT feedback through a free-dialogue user interface, and AI-powered writing analytics tool feedback through a structured interface. Our results revealed a strong initial preference for human tutors. However, the post-task analysis showed an important nuance. While the general preference for human tutors persisted, learners' preference towards the free-dialogue interface (ChatGPT 4.0) of ChatGPT increased, whereas the structured AI interface (AI-powered writing analytics tool) reinforced the preference for human tutors. These findings offer theoretical and practical contributions by extending algorithm aversion theory to educational contexts and demonstrating that appropriate interaction design can mitigate this aversion. The success of free-dialogue interfaces suggests that overcoming algorithm aversion may depend more on creating natural, flexible interaction experiences than purely technical optimization. However, we must also consider that increased preference for AI tools, particularly those with more engaging interfaces, may potentially lead to over-reliance and metacognitive laziness among learners, highlighting the importance of balancing technological support with the development of independent learning skills.
What is already known about this topic?
What this paper adds?
A promising approach to support students' math learning effectively, automatically and at scale within existing learning environments is conversational artificial intelligence (ConvAI). Although previous studies have suggested ConvAI's potential to guide, facilitate and enhance learning, its effects on students' conceptual change and academic motivation—the latter a crucial moderator of conceptual change—in math education remain understudied. Our study expands understanding of how ConvAI can be used to support Algebra learning from a conceptual change perspective. Using a between-subjects, pre- and posttest design, we conducted an experimental study with 151 participants, with the experimental group accessing ConvAI developed with induction, concretization and exemplification teaching strategies. Results showed that participants in the ConvAI group exhibited higher mastery goal orientation and self-efficacy compared with the control group post-intervention. The frequency of visiting recommended learning resources by ConvAI significantly predicted participants' motivation changes, with increased visits correlating with higher motivation. Additionally, although there was no significant main effect on misconceptions between ConvAI and no-AI participants, significant interaction effects on misconceptions emerged between treatment conditions and student motivation. Our findings, revealed through open-sourced implementations, provide support and implications for educational practitioners and researchers to design and develop pedagogically meaningful ConvAI for math learning.
This study explores whether AI-generated adaptive feedback or static feedback is favourable for student interest and performance outcomes in learning statistics in a digital learning environment. Previous studies have favoured adaptive feedback over static feedback for skill acquisition, however, without investigating the outcome of students' subject-specific interest. This study randomly assigned 90 educational sciences students to four conditions in a 2 × 2 Solomon four-group design, with one factor feedback type (adaptive vs. static) and, controlling for pretest sensitisation, another factor pretest participation (yes vs. no). Using a large language model, the adaptive feedback provided feedback messages tailored to students' responses for several tasks on reporting statistical results according to APA style, while static feedback offered a standardised expert solution. There was no evidence of pretest sensitisation and no significant effect of the feedback type on task performance. However, a significant medium-sized effect of feedback type on interest was found, with lower interest observed in the adaptive condition than in the static condition. In highly structured learning tasks, AI-generated adaptive feedback, compared with static feedback, may be non-essential for learners' performance enhancement and less favourable for learners' interest, potentially due to its impact on learners' perceived autonomy and competence.
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：