Beyond the "Wow" factor: Using Generative AI for Increasing Generative Sense-Making

Beyond the "Wow" factor: Using Generative AI for Increasing Generative Sense-Making | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Beyond the "Wow" factor: Using Generative AI for Increasing Generative Sense-Making Guido Makransky, Ban M. Shiwalia, Tue Herlau, Steven Blurton This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-5622133/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Generative Artificial Intelligence (GenAI) has emerged as a transformative tool in education, offering scalable, individualized learning experiences. However, there is a notable lack of theoretically informed and methodologically rigorous research on how GenAI can effectively augment learning. This study addresses this gap by investigating the potential of a theoretically informed GenAI chatbot, ChatTutor, to facilitate generative sense-making, leveraging principles from generative learning theory. The study had two primary goals: first, to build on theory to propose how GenAI could be used to support generative sense-making; and second, to empirically test its impact on conceptual knowledge, self-efficacy, and trust immediately after the intervention, as well as on conceptual knowledge, enjoyment, and behavioral intentions in a follow-up test four weeks later. Conducted in an authentic university course, the pre-registered experiment with 175 students compared ChatTutor to a generic GenAI system (ChatGPT) and a teaching-as-usual condition. Results show that ChatTutor significantly enhanced trust, enjoyment, and behavioral intentions but not self-efficacy. While ChatTutor improved conceptual knowledge over ChatGPT immediately after the intervention, it did not outperform teaching-as-usual. Four weeks later, ChatTutor significantly outperformed teaching-as-usual but not ChatGPT in conceptual knowledge. The manuscript underscores the importance of integrating human-centered design and educational psychology theories into GenAI applications to optimize learning outcomes and proposes future research and practical implications. Educational Psychology generative artificial intelligence generative sense making learning teaching self-explaining Figures Figure 1 Figure 2 Introduction Generative Artificial Intelligence (GenAI) has received increasing public interest and research (Kasneci et al., 2023 ; Lorenz et al., 2023 ; Yan et al., 2024 ). The widespread use of GenAI has surged with the release of publicly available large language models such as BERT (Devlin et al., 2019 ), Llama (Touvron et al., 2023 ), OpenAI (Helmore, 2023 ), Claude (Anthopic, 2024 ), and GPT-4 (Edwards, 2023 ). These models generate text, images, audio, and structured synthetic data through user-friendly interfaces, while also producing human-like responses (Cuéllar et al., 2024 ; Schramowski et al., 2022 ). Compared to predictive AI, which can be used for predicting labels or values, GenAI is designed to create new content or ideas and express them in real-time conversations (Qadir, 2023 ). GenAI has taken society by storm, in the history of the internet, no consumer app has grown faster than OpenAI's ChatGPT (Chow, 2023 ). The experience of asking a Chatbot anything and receiving a coherent, face valid answer has been described by users as "mind-blowing", "impressive", and "amazing" (Taecharungroj, 2023 ). This has led many to anticipate that this technology will inevitably revolutionize society, reshaping the ways we work, live, and learn (Lorenz et al., 2023 ). In education this could change how teachers teach, how learners learn and are assessed, and how education is administered (Chiu et al., 2023 ). However, emerging technologies typically progress through different levels of hype, and initial expectations are often exaggerated due to media coverage (Fenn & Blosch, 2020 ). While it can be argued that some emerging technologies such as computers have eventually had a significant influence on education, these changes are usually slower than initially expected and may be detrimental to the learning process (Chandler, 2009 ). In general, the digitalization of society results in challenges and opportunities for learning and education (Fischer et al., 2020 ; Yan et al., 2024 ). A common challenge with the introduction of novel technology is that it is often implemented and examined from a technocentric perspective, without incorporating extensive research and theory from educational psychology and the science of learning (Chandler 2009 ; Brennan, 2015 ; Yan et al., 2024 ). In this study, we argue that it is important to look beyond the "wow" factor of GenAI in education and adopt a learner-centered approach to investigating how the technology can enhance rather than distract from learning. Here, we agree with the perspective that psychological theory is central to understanding how GenAI can be used in this endeavor (e.g. Molenaar, 2022 ; Yan et al., 2024 ). Specifically, we use generative learning theory (Fiorella, 2023 ; Fiorella & Mayer, 2016 ) to design a GenAI educational application that supports human-AI collaboration in promoting generative processing. Generative learning activities (GLAs) are learner-driven actions that foster the active construction of knowledge as they prime key cognitive processes such as selecting, organizing, and integrating knowledge (Fiorella & Mayer, 2015 , 2022 ; Mayer, 1996 ; Wittrock & Farley, 2010 ). By aligning GLAs with the capabilities of GenAI, we explore how integrating human and machine intelligence can enhance learning outcomes, augmenting rather than replacing human abilities (Molenaar, 2022 ). That is, we adopt an augmentation perspective, viewing AI not as a simulation of human intelligence but as a distinct form of intelligence that complements our own, serving as a tool to potentially enhance human capabilities and help us function better as humans (Molenaar, 2022 ). Combining human and AI capabilities through hybrid intelligence has been proposed as a key to advancing the learning sciences in an AI era (Järvelä et al., 2023 ). But systematic research that builds on examining this integration is scarce (Hennessy et al., 2024 ; Roschelle et al., 2020 ). Ideally, hybrid intelligence systems should optimize human strengths while compensating for their weaknesses. In a world where knowledge is abundant and easily accessible, the focus of education could shift from imparting knowledge to nurturing students' ability to ask meaningful questions and critically evaluate information. A meta-analysis on the effect of inducing self-explanation found that having learners generate an explanation is often more effective than presenting them with an explanation (Birsa et al., 2018). Its impact was greatest when learners used new information to revise their explanations. However, while instructor-scripted self-explanation prompts work well for standardized content, they are impractical for highly personalized learning, and the authors recommend investigating computer-generated, content-specific prompts. GenAI makes this possible at scale, and in this study we investigate its potential to foster higher-order cognitive processes, such as conceptual understanding (Anderson & Krathwohl, 2001 ), while also promoting self-efficacy (Bandura, 1977 ), trust (Gulati et al., 2019 ), enjoyment (Davis et al., 1992 ; Huang & Zou, 2024 ), and behavioral intentions (Lee & Choi, 2017 ). Methodological rigor has been recently highlighted as a serious challenge to the research investigating GenAI in education (Yan et al., 2024 ). A criticism is that the majority of research on GenAI in education is conceptual, focusing on literature reviews, analyses of public discourse, or surveys of people's perceptions (Hennessy et al., 2024 , p. 6). Emerging evidence also suggests that the impact on learning and engagement presents a complex picture, such as instances where GenAI may have a positive initial impact on learning outcomes, however, it can act as a crutch and the removal of GenAI may ultimately hinder learning (Darvishi et al., 2024 ; Nie et al., 2024 ). This has prompted calls for more experimental research in authentic educational settings, grounded in robust theoretical frameworks, as well as for longitudinal studies to evaluate GenAI's long-term benefits to human learning by comparing its effectiveness with conventional methods (e.g. Hennessy et al., 2024 ; Yan et al., 2024 ). Furthermore, for GenAI to enhance rather than detract from learning, collaboration among researchers, practitioners, GenAI developers, policymakers, and educators is crucial to ensure its effective and responsible integration into teaching, aligning with educational goals (Yan et al., 2024 ). The first goal of this paper is to use theories from educational psychology and technology-enhanced learning to provide an understanding of how GenAI can effectively be incorporated into a lesson. While many theories could be used as an approach for using GenAI in education, in this article we focus on generative learning theory (Fiorella, 2023 ; Fiorella & Mayer, 2016 ). The second goal of the paper is to empirically test if the use of GenAI in education implemented to facilitate generative learning can increase conceptual knowledge, self-efficacy and trust immediately after a lesson, as well as conceptual knowledge, enjoyment and behavior intentions in a follow-up test four weeks later. In an experiment conducted in an authentic university setting, we compare the use of a theoretically informed GenAI application with a standard GenAI application and a teaching-as-usual condition. Prior to describing this experiment in more detail, we begin with the first objective and provide a theoretical overview of how GenAI can be used to facilitate generative learning. Theoretical Background Generative Learning Generative learning involves "making sense" of learning material by actively organizing and integrating it with existing knowledge (Wittrock, 1989 ). According to Fiorella ( 2023 , p. 2) " a generative learning activity (GLA) is something a learner does to try to make sense of what they are learning ". There is evidence that learners do not spontaneously engage in sense-making when learning from text (e.g. Fiorella & Mayer, 2017 ), visualizations (e.g. Makransky & Petersen, 2021 ) or examples (Renkl, 1997 ). Prompting learners to engage in GLAs is therefore a potential remedy for achieving meaningful and durable learning outcomes (Fiorella, 2023 ). Fiorella and Mayer ( 2015 , 2016 ) identified eight generative learning strategies (GLS) including: summarizing, mapping, drawing, imagining, self-testing, self-explaining, teaching, and enacting. Reviews and meta-analyses generally show that GLS support learning but each is susceptible to potential boundary conditions related to learner characteristics, learning materials, and support levels (Bisra et al., 2018 ; Brod, 2021 ; Dargue et al., 2019 ; Fiorella & Zhang, 2018 ; Lachner et al., 2021 ). That is, some learners need extensive guidance (Fiorella & Zhang, 2018 ), and in this study we investigate if GenAI can be used to provide personalized guidance that can improve educational outcomes. In the current study we are specifically interested in the verbalizing generative learning activities of teaching which incorporates elements of self-explaining. Generating explanations during learning can activate prior knowledge, integrate new information, enhance memory, and encourage reasoning by prompting students to make inferences and revise mental models (Brod, 2021 ). A recent meta-analysis (Bisra et al., 2018 ) found a mean effect size of 0.55 SDs for self-explanation, comparing conditions with and without self-explanation prompts. When considering only studies in which time on task was equated, the effect size was still 0.41 SDs. Generating explanations during learning can be done through self-explanation, which involves generating verbal statements (involving inferences) to clarify the meaning of the learning material to oneself (Fiorella & Mayer, 2022 , p. 341). Fiorella and Mayer's (2015) review found positive effects of self-explaining in 44 of 54 experiments, with a median effect size of 0.61. Self-explaining through teaching involves generating verbal statements (involving inferences) to convey the meaning of the learning material to others (Fiorella & Mayer, 2022 , p. 341). Fiorella and Mayer's (2015) review reported positive effects of learning by teaching in 17 of 19 experiments, with a median effect size of d = .71. A meta-analysis by Kobayashi ( 2019 ) found stronger effects for students who actually taught (g = .56) compared to those who only prepared to teach (g = .35). Explaining to others adds social elements, including a sense of presence and opportunities for interaction, for instance by answering questions. Evidence suggests that explaining is most effective when done aloud, without access to learning materials, when students possess background knowledge, and when they respond to thought-provoking questions (Fiorella & Mayer, 2022 ). Supporting factors include scaffolded peer interactions (King, Staffieri, & Adelgais, 1998 ), focused prompts (Berthold, Eysink, & Renkl, 2009 ), and explicit training (McNamara, 2004). However, most of this research has been conducted without the ability to provide adaptive specific feedback that is now possible with GenAI. In this study, we investigate the GSL of teaching, building on broader research in self-explanation, generative learning, AI-based education (AIEd), among other related fields. How can the use of Theoretically Informed GenAI Increase Generative Sense-Making? It is possible for instructors to use GLA products as formative assessment and adapt instruction and guidance as necessary (van de Pol et al., 2020 ). For instance, when learners' GLA products show knowledge gaps or misconceptions, instructors can give targeted feedback to aid their internal sense making, improve the GLA products, and enhance learning outcomes (Fiorella, 2023 ). The challenge is that assessing knowledge gaps and misconceptions and then providing targeted feedback is time consuming and difficult in most educational settings. The question that we attempt to investigate in this study is if a GenAI Chatbot can be used as a tool to scaffold learning by providing personalized feedback to stimulate and support generative learning activities. According to a taxonomy developed by Holmes and Tuomi ( 2022 ), AIEd applications can be classified into those that are student-focused, teacher-focused, or institution-focused. This study exclusively deals with student-focused AIEd applications that are leveraged to support student learning. Holmes & Tuomi ( 2022 ) also distinguish between AI-tools designed for students and those that are used by students. In this study we compare a theoretically informed GenAI-tool designed for students (ChatTutor) and a GenAI-tool used by students (ChatGPT) to a teaching-as-usual condition. ChatTutor builds on generative learning theory, and technology enhanced learning evidence and theory (e.g., Fiorella, 2023 ; Fiorella & Mayer, 2022 ; Mayer, 2014 ). Figure 1 depicts a framework for the development of a student-focused GenAI educational application where interaction is initiated by a) the learner, b) the AIEd system, and c) the instructor. In this study both ChatTutor and ChatGPT allow for interaction initiated by the learner. However, building on the research from generative learning (Fiorella, 2023 ; Fiorella & Mayer, 2016 ), we have furthermore designed ChatTutor to prompt students to engage in GLA and provide formative feedback. Effective guidance involves adapting learning materials and GLA support to learner's knowledge and beliefs, thereby enhancing their internal sense making and external products (Fiorella, 2023 ). We suggest that GenAI based learning systems such as ChatTutor can use the quality of these products as formative assessments to inform further adjustments of learning materials and GLA support. This is the case because in addition to providing responses to prompts that are initiated by the learner, learners who do not automatically initiate sense-making can systematically be prompted to do so by the GenAI system. We will use existing literature related to GLS to propose how this can be done in the following sections. Using a Theoretically Informed GenAI Chatbot for Prompting and Guiding GLA. There is evidence that for novices, complete discovery without guidance is ineffective; and some level of scaffolding is necessary (Mayer, 2004 ; Pedaste et al., 2015 ). Therefor for student-centered learning to be effective, guided discovery is essential. Ideally, this guidance must be personalized, which can be facilitated through technological advances (De Jong, 2006 ). GenAI systems can provide valuable feedback, and the focus should include the knowledge students possess as well as the questions they ask (Berlyne, 2006 ; Dillenbourg, 1999 ). The challenge is to support students in asking the right questions, finding accurate answers, and critically analyzing the information they receive to further investigate and understand the knowledge (Hmelo-Silver et al., 2007 ). Many guided learning principles including problem-based learning and inquiry learning, extensively use scaffolding to reduce cognitive load, enabling students to learn in complex domains (Hmelo-Silver et al., 2007 ). Fiorella ( 2023 ) suggests that GLA support can be categorized into three broad levels: explicit instruction, scaffolded practice, and independent practice which align with research on GLAs (Fiorella & Zhang, 2018 ; McNamara, 2017 ) and cognitive load theory (Sweller et al., 2019 ), progressing from worked examples to partial problems to problem-solving practice. Explicit instruction includes explanations and demonstrations on selecting and performing GLAs. Explicit prompts help learners to generate internal self-explanations (Fiorella, 2023 ). The literature on worked examples suggests that this is necessary because learners often struggle to detect errors (Van Meter, 2001 ). They may need support in generating internal feedback (e.g., self-explaining why their response was inaccurate) and revising their knowledge (Zhang & Fiorella, 2023 ). Wittwer and Renkl ( 2010 ) describe how effective instruction involves a balanced combination of providing support, such as worked solution steps, and encouraging learner activity, like self-explanation prompts. An experienced instructor can provide this balance by encouraging learners to engage in a GLA such as self-explaining, or teaching, and then help them detect gaps in their knowledge thereby stimulating further inquiry. This requires the ability to engage learners in generative processing, and the knowledge to recognize and highlight gaps in a way that motivates students to engage in further information seeking (Murayama, 2022 ). A GenAI chatbot can be programmed to provide specific self-explanation prompts, trained using a targeted knowledge bank, and interact with learners as a knowledgeable teacher, an overconfident peer, or a curious collaborator. Once learners become more familiar with the course content then GenAI based chatbots can progress to scaffolded practice activities. Scaffolded practice includes providing specific prompts or hints (Berthold & Renkl, 2009 ; Roelle & Berthold, 2017 ) or partial representations for learners to complete (Fiorella et al., 2021 ; Ponce et al., 2020 ; Schwamborn et al., 2010 ). Learners can benefit from scaffolded prompts where they fill in key components of an explanation rather than generating the entire explanation themselves (Bai et al., 2022 ). Self-explaining is more effective when learners receive focused prompts targeting specific relationships or principles rather than open or generic prompts (Berthold & Renkl, 2009 ). Scaffolded practice is most effective when followed by appropriate feedback, which helps learners evaluate and correct their performance (Johnson & Marrafano, 2022). In the current study we build on this literature by using GenAI to prompt students to engage in the generative activities of teaching. More specifically, students are given prompts that ask them to explain key components or explanations. The ChatTutor chatbot engages students in personalized written conversations based on the course slides and texts, and responds to address the specific answers provided by students and correcting any misconceptions or gaps in knowledge. Students are gradually provided with more information, scaffolding their understanding until they can offer a suitable explanation. This approach simulates a student-tutor interaction, where the tutor encourages generative learning by prompting students to explain specific content, intervening only when misconceptions are evident or when students request help. Finally, independent practice involves repeated opportunities to use GLAs in specific learning contexts. This practice is crucial for learners to use GLAs spontaneously (Manalo et al., 2017 ) but is not investigated in this study. It is important to note that we are not proposing that GenAI is necessary or in its current technological iteration optimal for progressing from explicit instruction through scaffolded practice and ultimately to independent practice, however, we do propose that GenAI can be used to scale this process when resources to formatively assess students and provide personalized support is not possible. In the following section we describe how a theoretically informed GenAI chatbot can be used to mitigate some of the documented barriers to GLS. How can a Theoretically Informed GenAI Chatbot Mitigate Known Barriers of GLS? Fiorella ( 2023 ) describes different cognitive, metacognitive and motivational barriers to sense making. Below we describe how a GenAI chatbot could attempt to mitigate these barriers thereby increasing sense-making and ultimately learning and motivational outcomes. Cognitive Barriers : One cognitive barrier to engaging in effective GLS is insufficient background knowledge or available memory capacity to engage in GLAs effectively (Castro-Alonso et al., 2021 ). Learners need sufficient background knowledge to generate the necessary inferences for sense-making (Simonsmeier et al., 2022 ; Willingham, 2008 ). Without it, learning materials won't be effective, regardless of the GLA used. With GenAI, students can ask a personalized chatbot particular questions thereby gaining specific knowledge. In ChatTutor this is implemented by linking directly to the texts and slides from the course., or by asking the LLM to provide simpler alternative explanations such as summaries or explanations. This is relevant because lower-knowledge students often benefit more from GLAs than higher-knowledge students (Fiorella & Mayer, 2015 ). This is because higher-knowledge learners are more likely to engage in sense-making spontaneously (Lombrozo, 2006 ). Metacognitive Barriers : Metacognitive barriers include a lack of strategic knowledge about what GLAs are, how to choose the right ones, and how to use them effectively. Many learners do not use GLAs spontaneously when studying texts (Fiorella & Mayer, 2017 ), visualizations (Renkl & Scheiter, 2017 ), worked examples (Chi et al., 1989 ; Renkl, 1997 ), solving problems (Rellensmann et al., 2022 ), discussing material with peers (Roscoe & Chi, 2007 ), or writing essays (Bereiter & Scardamalia, 2013 ). GenAI could provide explicit instructional support to assist learners to engage in GLAs. Specific prompting and scaffolding have been shown to be more effective compared to generic prompts in explaining, visualizing, or enacting concepts (Berthold & Renkl, 2009 ; Schmeck et al., 2014 ). Moreover, the quality of what learners generate during learning typically predicts their performance on later comprehension and transfer tests (Chi et al., 1989 ; Fiorella & Kuhlmann, 2020 ; Goldin-Meadow et al., 2009 ; Renkl, 1997 ; Schwamborn et al., 2010 ). Motivational barriers : Motivational barriers include beliefs about one’s ability to use GLAs successfully, the perceived value of GLAs for achieving one’s goals, or the perceived cost of GLAs (e.g., (Schukajlow et al., 2022 ). GenAI could increase self-efficacy by providing individualized mastery experiences (Bandura, 2001 ). A GenAI Chatbot can be seen as a more knowledgeable companion, providing positive feedback and enhancing learning by targeting the learner's zone of proximal development (Vygotsky, 1978 ). Can a Theoretically Informed GenAI Chatbot influence other outcomes? In addition to cognitive learning outcomes and self-efficacy, there are other important outcome variables that can ultimately impact the use of GenAI for learning. In this study we selected three: trust, enjoyment, and behavioral intentions. The adoption of GenAI in educational contexts brings forth several ethical challenges, such as transparency, privacy, equality, and beneficence (Khosravi et al., 2022 ). While all of these factors are relevant, building a GenAI chatbot that takes a learner-centered approach to interaction and builds on the curriculum in a course may specifically be relevant for transparency. A recent systematic review by Yan et al. ( 2023 ) found that most GenAI tools (92%) currently employed to support learning are comprehensible only to AI experts. Educators, students, and other key stakeholders often lack the necessary insight into how these tools function. The authors highlight that the transparency gap largely stems from the limited integration of human-in-the-loop approaches in prior research. For instance, educators and students have seldom been actively involved in the design and evaluation of GenAI-based educational technologies. This shortfall highlights the urgent need for learner-centered AI, emphasizing the importance of engaging all stakeholders in the development process to ensure GenAI tools are both effective and meaningful in real-world educational settings. In this study we engage instructors, researchers, learners, and a GenAI development start-up in developing the ChatTutor system, and assess students’ trust (Gulati et al., 2019 ) in using the system in our experiment. It can also be useful to measure the enjoyment of using AI technology in education because enjoyment is a powerful motivational factor that directly impacts learners' engagement and long-term adoption of these tools (Huang & Zou, 2024 ). According to the Control-Value Theory of Achievement Emotions, enjoyment plays a critical role in fostering intrinsic motivation and engagement by enhancing learners' perceptions of control over their learning and the value they assign to educational tasks (Pekrun, 2006 ). This positive emotional experience can reinforce satisfaction with the learning process and increase the intention to continue using AI-enhanced educational platforms (Venkatesh & Bala, 2008 ), leading to sustained engagement and ultimately more widespread adoption (Dai et al., 2024 ). We therefore measure trust, enjoyment, and behavioral intentions in this study. Current study In the current study we address some of the gaps in the literature by designing a GenAI application building on a learner-centered theory of learning and instruction. Our main research question (RQ 1) is: What are the effects of using a theoretically informed GenAI chatbot (ChatTutor) to stimulate generative sense making on learning outcomes? We pre-registered two main hypotheses regarding this research question: • Hypothesis 1 (H1): Learners in the theoretically informed GenAI condition (ChatTutor) will achieve higher scores on a conceptual knowledge test immediately after the lesson when compared to learners in the teaching-as-usual (H1a) and ChatGPT-4 (H1b) conditions. • Hypothesis 2 (H2): Learners in the theoretically informed GenAI condition (ChatTutor) will achieve higher scores on a conceptual knowledge test administered four weeks after the lesson when compared to learners in the teaching-as-usual (H2a) and ChatGPT-4 (H2b) conditions. Our secondary research question (RQ 2) is: Does the use of a theoretically informed GenAI chatbot (ChatTutor) lead to higher (1) self-efficacy compared to the teaching-as-usual condition and the ChatGPT condition, and more (2) trust, (3) enjoyment, and (4) intentions to use the technology in the future compared to the ChatGPT condition? Methods Study design The study adopted a between-subject design consisting of one control group and two distinct GenAI conditions: ChatGPT and ChatTutor. The design, hypotheses, data collection, and analysis plan for this study were pre-registered on AsPredicted in September 2024, prior to the commencement of data collection. The anonymized preregistration is available at: https://aspredicted.org/w5mk-4vx4.pdf . The study was conducted at a large European University. Based on the self-assessment, the department’s local ethical board accepted the research project. Procedure The project was initiated by psychology university instructors who were interested in investigating if GenAI could improve conceptual knowledge retention, self-efficacy, trust, enjoyment, and behavioral intentions in a cognitive psychology university course on the topic of Sternberg’s serial memory scanning experiment (Sternberg, 1966 , 1969 ) which is a challenging topic for most students. The cognitive psychology course consists of a weekly lecture with over 240 psychology students, followed by both a seminar and an exercise class. At the beginning of the semester, students were randomly assigned to one of seven cognitive psychology classes with approximately 35 students in each. In accordance with local study board regulations, attendance in the lecture or the classes was not mandatory. The experiment took place in the first week of October, 2024, when the topic in the course was short-term memory and working memory. The follow-up took place as the first part of the lesson four weeks later. The Sternberg task was first introduced at the end of the seminar classes. In the following exercise class, students received a more in-depth overview and performed the task themselves. The exercise class consisted of three 45 minute blocks with 15 minute breaks between. The first block started with a wrap-up of the previous week, followed by a more in-depth introduction into the short-term memory task. In the second teaching block students took part in the Sternberg experiment themselves. After running the experiment, the students analyzed their logfile created at the end of the experiment, this task was otherwise unrelated to the current study. The third block started with a 20-minute power-point presentation by the instructor which was standardized across the different classes. This was followed by a 15 minute follow-up activity which was the focus of the experiment. In this between-subjects pre-registered experiment, two exercise classes were assigned to using the theoretically informed GenAI system (ChatTutor), two classes were assigned to the standard ChatGPT-4 (standard GenAI) condition which had the same interface as the ChatTutor condition, and the remaining three classes were assigned to the teaching-as-usual condition. Therefore, the difference between the groups was isolated to the generative learning activity which took approximately 15 minutes (see Fig. 2 ). The final 10 minutes of the class consisted of a post-test including measures of conceptual knowledge, self-efficacy, and trust in AI. Four weeks after the intervention students used 10 minutes at the beginning of the exercise class to take a follow-up test that included another conceptual knowledge test as well as measures of enjoyment and behavioral intentions. Students were informed by their instructors that the assessment was an effort to investigate how to improve the teaching methods used in the course, but the students were not informed of the differences between instructor classes. At the end of the semester, students were informed of the results of the experiment and students in all treatment conditions were given access to the ChatTutor system. Participants A total of 175 university students who were enrolled in the cognitive psychology course and attended their exercise class, agreed to participate in the study. Of these, 75 participants were in the teaching-as-usual condition, 49 were in the ChatGPT condition, and 51 were in the ChatTutor condition. The topic of the lesson was centered on the Sternberg Experiment. The experiment was integrated into classes’ schedules and conducted during regular teaching hours. Participants provided their informed consent prior to the commencement of the experiment. As outlined in the preregistration, data from participants who chose not to engage in the experiment (< 5) were excluded. To our knowledge there were no additional students who opted not to use the AI program, and we did not exclude any students due to technical issues while using the GenAI program. A total of 124 students who completed the initial test, also completed the delayed follow-up test four weeks after the intervention. Among these, 54 participants belonged to the teaching-as-usual condition, 32 to the ChatGPT condition, and 38 to the ChatTutor condition. Conditions The experiment was isolated to the follow up activity which lasted approximately 15 minutes. All students received the similar instructions regarding the follow-up activity. Those in the teaching-as-usual-condition had 15 minutes to reflect over what they had learned and could discuss this with their classmates and had the opportunity to ask their instructor any questions which they were in doubt about regarding the topics that they had learned that day. The students in the ChatGPT condition had 15 minutes to reflect over what they had learned and had the opportunity to ask the GenAI any questions which they were in doubt of regarding the topics that they had learned that day. The students in the ChatTutor condition also had 15 minutes to engage in the ChatTutor system as described above. 1. Teaching-as-usual : Students in this condition were able to engage with the material and discuss with their fellow-students and could ask the instructor any relevant questions related to the material. Moreover, they were free to use any existing learning materials and internet sources. 2. Generative Sense Making (ChatTutor) condition : Students in this condition got access to ChatTutor website, which was designed to encourage generative processing as their learning tool. Students were encouraged to engage with the tool which actively prompted students to partake in the generative learning activity of teaching with scaffolded feedback. The ChatTutor system allowed learners to escalate a question to a human instructor if they encountered confusing or inaccurate information. 3. Standard GenAI (ChatGPT) condition : Students in this condition got access to a website interface that was identical to ChatTutor. However, the system linked directly to ChatGPT-4. Students were encouraged to engage with the tool with the same general instructions as in ChatTutor condition but the system responded based on the current technological capabilities of ChatGPT-4. However, the system also allowed learners to escalate a question to a human instructor if they encountered confusing or inaccurate information. Measures The complete list of items is included in the supplementary materials. Conceptual Knowledge (post-test and follow-up) : A test assessing participants' conceptual knowledge, with questions related to the memory scanning experiment and Sternberg’s original hypotheses (Sternberg, 1969 ) was administered immediately after the intervention/at the end of the lesson and again after four weeks. The test consisted of 10 multiple choice items, however, one was removed from the analyses and not included in the follow-up test because it accidentally contained more than one correct answer, resulting in a total of nine items. The test was designed to measure a broad range of topics covered with a focus on comprehension and not factual knowledge. That is, we prioritized content validity by having a broad range of questions rather than focusing on a narrow range of content which could have potentially increased internal reliability. The reliability coefficient for the nine-item conceptual knowledge test was α = .55 in the post-test and α = .68 in the follow-up. The test-retest reliability was r = .41. New Conceptual Knowledge items (follow-up) : In addition to the nine items, four new conceptual knowledge items were administered in the follow-up to ensure that we could measure conceptual knowledge while accounting for testing effect (that students would remember the answers to the items because they had seen them before). The reliability coefficient for the new conceptual knowledge items was α = .40. Self-efficacy (post-test) : We measured self-efficacy using three items adapted from (Pintrich et al., 1991 ): 1. “I’m confident I can understand the concepts of serial and parallel memory search”; 2. “I’m confident I can understand the concepts of serial exhaustive search and serial self-terminating search”; 3. “I’m confident I could explain the Sternberg memory search graphs to a friend”. The reliability coefficient of the self-efficacy measure was α = .82. Trust (post-test) : We measured trust using four items adapted from Gulati et al. ( 2019 ): 1. “I feel I must be cautious when using ChatTutor”; 2. “I believe that ChatTutor will act in my best interest”; 3: “I think that ChatTutor is competent and effective in providing teaching assistance”; 4. “I can trust the information presented to me by ChatTutor”. The reliability coefficient of the trust measure was α = .64. Enjoyment (follow-up) : We measured enjoyment using three items adapted from Huang & Zou ( 2024 ), originally from (Davis et al., 1992 ): 1) "I find it satisfying to use ChatTutor," 2) "I find it rewarding to use ChatTutor," and 3) "I find it pleasant to use ChatTutor." The reliability coefficient for the enjoyment measure was α = .80. Behavioral intentions (follow-up) : We measured behavioral intentions with three items adapted from Lee and Choi ( 2017 ), originally from Davis et al. ( 1992 ). 1. “I intend to use ChatTutor again”; 2. “In the future, I will use ChatTutor to support my learning processes”; 3. “I would recommend ChatTutor to others”. The reliability coefficient for behavioral intentions was α = .55. All items in the self-efficacy, trust, enjoyment, and behavioral intentions were measured on a 5-point scale (Strongly disagree:1. Disagree: 2. Neither agree nor disagree: 3. Agree: 4. Strongly agree: 5). Data analysis and deviations from pre-registration The analyses for the study were performed using SPSS Statistics (IBM Corp., 2023). We examined our hypotheses using independent samples t-tests rather than ANOVA as the hypotheses were framed as pairwise comparisons and because trust, enjoyment and behavioral intentions were only assessed in the two GenAI groups. A sensitivity power analysis using G*power (Faul et al., 2007 ) indicated that with an α = .05, and power of .80 we could expect to find an effect size d = .45 for the difference between the ChatTutor and teaching-as-usual conditions and an effect size of d = .50 between the ChatTutor and ChatGPT conditions. This value increased to d = .53, and d = .60 respectively in the follow-up due to drop-out. Values below these levels should therefore be interpreted with caution. Please note that we only pre-registered our measures of conceptual knowledge, self-efficacy, and trust, but did not pre-register the measures of enjoyment or behavioral intentions, which were initially left out of the post-test due to time constraints but we were later able to include them in the follow-up. Results The main results of the study can be seen in Table 1. The results are organized based on the research questions below. Table 1: Means, standard deviations, sample sizes and significance levels for the outcome variables used in the study. Results Related to the Main Research Question (RQ 1): The main research question in this study investigated the effects of using a theoretically informed GenAI chatbot (ChatTutor) to stimulate generative sense making on learning outcomes immediately after the intervention and in a delayed follow-up test four weeks after the intervention. The top row of Table 1 illustrates that the results partially support Hypothesis 1. More specifically, the theoretically informed GenAI ChatTutor condition resulted in significantly higher conceptual knowledge compared to the ChatGPT condition in the immediate post-test. However, while the ChatTutor condition mean score was higher than the teaching-as-usual condition, this difference was not statistically significant in the immediate post-test. The next row of Table 1 illustrates that the results also partially support Hypothesis 2. The theoretically informed GenAI ChatTutor condition resulted in significantly higher conceptual knowledge compared to the teaching-as-usual condition in the follow-up test four weeks after the intervention. However, while the ChatTutor condition mean score was higher than the standard ChatGPT condition, this difference was not statistically significant in the follow-up test. Because it was not possible to limit the students from using the ChatTutor interface between the lesson and the follow-up test, we asked students to report on a five-point scale how often they had used ChatTutor system since the lesson (1. Not at all; 2. Seldom; 3. Sometimes; 4. Often; 5. Very often). All of the students in the ChatGPT condition reported not using it at all, and all but six students in the ChatTutor condition reported not using it at all (four reported using it seldomly, and two reported using it sometimes). We re-ran all the analyses for the follow-up test, discarding these students as a robustness check, and found that this did not change any of the findings. Results Related to the Secondary Research Question (RQ 2): The second research question in this study was if a theoretically informed GenAI chatbot (ChatTutor) leads to higher (1) self-efficacy compared to the teaching-as-usual condition and the ChatGPT condition, and more (2) trust, (3) enjoyment, and (4) intentions to use the technology in the future compared to the ChatGPT condition. The results in Table 1 illustrate that the ChatTutor group (M = 3.64) scored significantly higher than the ChatGPT group (M = 3.29) on trust t = 2.955; p = .002. The ChatTutor group (M = 3.65) also scored significantly higher than the ChatGPT group (M = 2.99) on enjoyment t = 3.304; p < .001. Furthermore, the ChatTutor group (M = 3.28) scored significantly higher than the ChatGPT group (M = 2.73) on behavioral intentions t = 2.188; p = .016. However, the ChatTutor group (M = 4.09) scored only marginally higher than the ChatGPT group (M = 3.82) on self-efficacy t = 1.661; p = .050. Finally, there were no significant differences on self-efficacy between the ChatTutor and the teaching-as-usual conditions t = .167; p = .434. Discussion The first goal of this manuscript was to propose how GenAI could be used to support generative sense-making, providing a theoretical account of a learner-centered approach to using GenAI in education. Adopting an augmentation perspective, we use generative sense making research and theory to describe how GenAI can be used enhance lasting learning outcomes. More specifically, we propose how GenAI can be used to prime generative processing through enhancing personalized feedback when using teaching as a generative learning strategy. The second goal of this manuscript was to empirically investigate whether a theoretically informed GenAI chatbot (in this case, ChatTutor) could facilitate generative learning in education. Specifically, we aimed to test its impact on conceptual knowledge, self-efficacy, and trust immediately after a lesson, as well as its influence on conceptual knowledge, enjoyment, and behavioral intentions in a follow-up test four weeks later. Empirical Contributions The results of our study indicated that using ChatTutor resulted in significantly higher trust, enjoyment, and intentions to use the technology, and marginally higher self-efficacy compared to using ChatGPT. Here it is important to note that the students were unaware of the different conditions, as they were blind to the study's design and all used the ChatTutor interface. The difference was that the ChatTutor group obtained a theoretically informed GenAI learning intervention that builds on generative learning research and theory. Regarding conceptual knowledge the results indicate that the ChatTutor condition outperformed the ChatGPT condition in the immediate post-test but not in the delayed follow-up, and conversely the ChatTutor condition outperformed the teaching-as-usual condition in the follow-up but not the immediate post-test. The results related to conceptual knowledge are therefore mixed. Delving into the mixed results related to conceptual knowledge in more depth, it is important to note that the groups only had 15 minutes on the follow-up generative activity task. This should be seen in the context of the entire course where learners were exposed to the topic of the Sternberg experiment through a seminar, lab activities and a specific lecture about the topic prior to the intervention. Therefore, it is likely that many students gained fundamental knowledge prior to engaging in the follow-up activity. While it would have been optimal to allow students more time to use ChatTutor we wanted to keep the amount of time across conditions constant as this has been highlighted as an important methodological consideration in previous studies (Brod, 2021 ). Furthermore, it is a strength that the study took place in an authentic learning setting where the value of GenAI for learning should ultimately be tested. When delving into the literature on GLA there are some other factors to be aware of regarding our mixed results. Previous research on teaching as a GLA suggests that oral explanations may be more effective than written ones, as they can evoke a sense of social presence, leading to higher levels of arousal (Hoogerheide et al., 2019 ). A recent meta-analysis found a large positive effect of AI chatbots on learning outcomes (Wu & Yu, 2024 ). The authors recommend that future designers and educators further enhance these outcomes by incorporating human-like avatars, gamification elements, and emotional intelligence into AI chatbots. In this study we did not attempt to increase feelings of social presence through any of these methods which provide avenues for future research. Furthermore, there is evidence that the GLA of teaching is most effective when the learning material is not available (Koh et al., 2018 ). This can be the case because it requires learners to retrieve information from long-term memory rather than relying on the lesson, aiding in knowledge consolidation and future accessibility (e.g., Roediger & Karpicke, 2006 ). In the current study the material was visually available for the students through the ChatTutor system. Therefore, it would have been possible to interact with ChatTutor by finding relevant passages in the text and paraphrasing them in the chat-box. Previous research has found that only students who generate an explanation exhibit improved understanding (Fiorella & Mayer, 2013 ; Hoogerheide et al., 2016 ). Therefore, future research should investigate the availability of learning content when using GenAI for GLS. Theoretical Implications The leading theoretical paradigms guiding research on learning have evolved over the past 100 years, reflecting paradigm shifts from response strengthening to information acquisition, knowledge construction, and knowledge co-construction (Mayer, 2012 ). Only time will tell if we are on the brink of a paradigm shift to a human-computer co-construction of knowledge. While it is unquestionable that existing educational psychology theory and research should guide and iteratively improve the development of GenAI-based learning interventions, as argued in this manuscript, a fundamental question remains: if and how educational psychology research must adapt to the new possibilities GenAI brings to education. The fundamental advantage of using GenAI lies in its ability to scale individualized learning experiences through realistic, dynamic conversations tailored to learners' needs. This approach shifts the focus from static, one-size-fits-all interventions to methods for delivering just-in-time support, aligning with advancements in learning analytics and self-regulated learning research (Azevedo & Gašević, 2019 ; Molenaar & Järvelä, 2014 ). In the context of generative sense-making, this could involve optimizing real-time feedback to support the creation and refinement of generative learning products, thereby enhancing engagement and understanding. Practical Implications One of the main challenges associated with the recent rapid advancements in GenAI is its disruptive impact. As formulated in Holmes et al. ( 2019 ), p 3: “If you can search, or have an intelligent agent find, anything, why learn anything? What is truly worth learning?” Hence, it is useful to discuss how GenAI-enhanced learning fits (or does not fit) with traditional views of learning outcomes and processes. The goal of most education is for learners to be able to apply what they have learned in a relevant context. This is referred to as meaningful learning or deep knowledge acquisition which is indicated by performance on transfer tests, which involves being able to use the learned material in new situations (Mayer, 2024 ). Kasneci et. al. ( 2023 ) highlight some of the opportunities and challenges of using GenAI in education. The opportunities include personalized learning experiences, and challenges include the need for critical thinking and strategies for fact checking because GenAI agents have a tendency to hallucinate and provide inaccurate information. In a society where knowledge is abundant and easily accessible through internet searches or conversations with GenAI chatbots, the goal of education cannot be to simply impart knowledge but rather to engage students to ask the right questions and develop the critical thinking skills needed to evaluate the information they receive. This would suggest the need to shift from remembering and focusing on higher order cognitive process dimensions such as understanding, applying, analyzing, evaluating, and creating from Anderson and Krathwohl's (2001) taxonomy. One challenge is that GenAI may make students regulate less (Yan et al., 2024 ). For example, while GenAI enhances the efficiency of information processing and retrieval, it poses a risk of fluency bias, where learners may overestimate their understanding due to the ease of processing information. Additionally, relying on GenAI for creative and problem-solving tasks could undermine these essential skills, fostering dependency and potentially hindering innovation and original thought (Rafner et al., 2023 ; Scneiderman, 2020). From a practical perspective this highlights the importance of thinking of GenAI as support rather than using it as a replacement as we have attempted to do in this manuscript. Another important practical implication is the need to educate stakeholders on the differences between using generic GenAI tools like ChatGPT, educational GenAI, and educational GenAI designed with educational psychology theory. Our study results clearly show that an educational GenAI chatbot that builds on theory, such as ChatTutor, fosters significantly greater trust, enjoyment, and intentions to use the technology in the future compared to a generic GenAI system like ChatGPT. Another important practical consideration is the challenge teachers face in defining their role in the use of GenAI for learning. Yan et al. ( 2024 ) suggest incorporating human-in-the-loop elements, such as fostering partnerships among researchers, practitioners, and policymakers. This approach was used in our study, where we collaborated with instructors, researchers, and a GenAI development startup to design the ChatTutor intervention. Furthermore, the intervention included human-in-the-loop elements, as the ChatTutor system allowed learners to escalate a question to a human instructor if they encountered confusing or inaccurate information. This feature is crucial given that GenAI chatbots are known to 'hallucinate,' that is, provide false information. Limitations and Future Research Our study had several limitations. The primary limitation was the short duration learners had to interact with the GenAI chatbot, compared to the extensive time spent on the topic through a seminar, exercise course lectures, and lab activities. This setup was not ideal for measuring the tool's impact on learning outcomes but was necessary given the context of an authentic cognitive psychology course. Such an approach is typical of a value-added study, where the objective is to assess the additional benefit of an educational intervention. Future research should explore the value of using GenAI to facilitate GLS when students have more time to engage with a theoretically informed GenAI chatbot. Additionally, future studies should examine different age groups, as previous research has shown varying levels of GLS effectiveness across age groups (Broad, 2021). Another limitation of our study concerns the reliability coefficients of some outcome measures, including conceptual knowledge and trust, which fell below acceptable levels. We prioritized content validity in our conceptual knowledge measure by including a broad range of items assessing different components of the learning material. However, future research would benefit from incorporating measures that are both valid and reliable. The goal of the learning intervention in this study was conceptual knowledge understanding. Future research could explore the value of theoretically informed GenAI for higher-order learning outcomes, such as applying, analyzing, evaluating, and creating, where relevant to learning interventions. Furthermore, this study focused on developing a GenAI intervention based on generative sense-making research and theory. Future research should explore how other educational psychology theories can inform the development of GenAI interventions, given the scarcity of such studies. Conclusion Generative Artificial Intelligence (GenAI) has captured significant public and scholarly attention and could challenge education by transforming traditional methods of teaching, learning, assessment, and administration. However, a recurring challenge with novel technologies lies in the tendency to implement and evaluate them from a technocentric perspective, often neglecting insights from educational psychology and learning science. This manuscript had two primary goals: first, to extend generative learning theory by proposing how GenAI can be utilized to support generative sense-making in education; and second, to empirically examine its impact on the outcomes of conceptual knowledge, self-efficacy, trust, enjoyment, and behavioral intentions. To achieve these objectives, we developed a theoretically informed GenAI chatbot (ChatTutor) designed to scaffold learning by prompting students to engage in the generative learning strategy of teaching a GenAI chatbot. Through a pre-registered experiment conducted in an authentic university setting, we compared ChatTutor to ChatGPT and a teaching-as-usual condition. The findings demonstrated that ChatTutor significantly enhanced trust, enjoyment, and behavioral intentions compared to ChatGPT, but not self-efficacy. Regarding learning outcomes, ChatTutor led to higher immediate conceptual knowledge scores than ChatGPT but did not outperform teaching-as-usual. In the delayed follow-up test, ChatTutor led to significantly higher conceptual knowledge than teaching-as-usual, but the differences with ChatGPT were no longer significant. This study underscores the importance of adapting educational psychology theories to harness the unique capabilities of GenAI in supporting meaningful learning. The findings contribute to the growing call for theory-driven, learner-centered integration of GenAI into education, moving beyond the initial fascination with the technology. Declarations Acknowledgements We would like to thank Gustav B. Petersen, Benjamin B. Christensen, and Sidsel Drejer for helping develop the content of the lesson used in the experiment. Furthermore, we would like to thank Gustav B. Petersen, Marius Corson, Mikkel Marfelt and Kurt G. Nielsen for helping organize the experiment. Finally, we would like to thank Sidsel Drejer, Benjamin B. Christensen, Rubina F. Gogolu, Rebecca G. Lachmann, Alexander T. Ysbæk-Nielsen, Monique S. Damberg and Daniel Jensen who were the instructors who make the experiment possible. References Anderson, L. W., & Krathwohl, D. R. (2001). A Taxonomy for learning, teaching, and assesing. A taxonomy for learning, teaching and assessing: A revision of Bloom’s taxonomy . Longman Publishing, 1–336. Anthopic. (2024). Claude’s character . Anthopic. https://doi.org/https://www.anthropic.com/research/claude-character Azevedo, R., & Gašević, D. (2019). Analyzing multimodal multichannel data about self-regulated learning with advanced learning technologies: Issues and challenges. Computers in Human Behavior , 96 , 207–210. https://doi.org/10.1016/j.chb.2019.03.025 Bai, Y., Jones, A., Ndousse, K., Askell, A., Chen, A., DasSarma, N., Drain, D., Fort, S., Ganguli, D., Henighan, T., Joseph, N., Kadavath, S., Kernion, J., Conerly, T., El-Showk, S., Elhage, N., Hatfield-Dodds, Z., Hernandez, D., Hume, T., … Kaplan, J. (2022). Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback , 1–74. https://doi.org/10.48550/arXiv.2204.05862 Bandura, A. (1977). Self-efficacy: Toward a unifying theory of behavioral change. Psychological Review , 84 (2), 191–214. https://doi.org/10.1037/0033-295X.84.2.191 Bandura, A. (2001). Social cognitive theory: An agentic perspective. In Annual Review of Psychology, 52, 1–26. https://doi.org/10.1146/annurev.psych.52.1.1 Benbasat, I., & Wang, W. (2005). Trust In and Adoption of Online Recommendation Agents. Journal of the Association for Information Systems , 6 (3), 72–101. https://doi.org/10.17705/1jais.00065 Bereiter, C., & Scardamalia, M. (2013). The psychology of written composition. In The Psychology of Written Composition , 1–389. https://doi.org/10.4324/9780203812310 Berlyne, D. E. (2006). Conflict, arousal, and curiosity. In Conflict, arousal, and curiosity, 1–366. https://doi.org/10.1037/11164-000 Berthold, K., & Renkl, A. (2009). Instructional Aids to Support a Conceptual Understanding of Multiple Representations. Journal of Educational Psychology , 101 (1), 70–87. https://doi.org/10.1037/a0013247 Berthold, K., Eysink, T. H., & Renkl, A. (2009). Assisting selfexplanation prompts are more effective than open prompts when learning fromm multiple representations. Instructional Science, 37, 345–363. https://doi.org/10.1007/s11251-008-9051-z Bisra, K., Liu, Q., Nesbit, J. C., Salimi, F., & Winne, P. H. (2018). Inducing Self-Explanation: a Meta-Analysis. Educational Psychology Review , 30 (3), 703–725. https://doi.org/10.1007/s10648-018-9434-x Brennan, K. (2015). Beyond technocentrism: Supporting constructionism in the classroom. Constructivist Foundations , 10 (3), 289–286. https://constructivist.info/10/3/289.brennan.pdf Brod, G. (2021). Generative Learning: Which Strategies for What Age? In Educational Psychology Review . 33 (4), 1295–1318. https://doi.org/10.1007/s10648-020-09571-9 Burkhart, C., Lachner, A., & Nückles, M. (2021). Using Spatial Contiguity and Signaling to Optimize Visual Feedback on Students’ Written Explanations. Journal of Educational Psychology , 113 (5), 998–1023. https://doi.org/10.1037/edu0000607 Castro-Alonso, J. C., de Koning, B. B., Fiorella, L., & Paas, F. (2021). Five Strategies for Optimizing Instructional Materials: Instructor- and Learner-Managed Cognitive Load. In Educational Psychology Review , 33 (4), 1379–1407. https://doi.org/10.1007/s10648-021-09606-9 Chandler, P. (2009). Dynamic visualisations and hypermedia: Beyond the “Wow” factor. Computers in Human Behavior , 25 (2), 389–392. https://doi.org/10.1016/j.chb.2008.12.018 Chi, M. T. H., Bassok, M., Lewis, M. W., Reimann, P., & Glaser, R. (1989). Self-explanations: How students study and use examples in learning to solve problems. Cognitive Science , 13 (2), 145–182. https://doi.org/10.1016/0364-0213(89)90002-5 Chiu, T. K. F., Moorhouse, B. L., Chai, C. S., & Ismailov, M. (2023). Teacher support and student motivation to learn with Artificial Intelligence (AI) based chatbot. Interactive Learning Environments, 32 (7), 1–17. https://doi.org/10.1080/10494820.2023.2172044 Chow, A. (2023). How ChatGPT Managed to Grow Faster Than TikTok or Instagram. TIME . Retrieved https://time.com/6253615/chatgpt-fastest-growing/ Cuéllar, M. F., Larsen, B., Lee, Y. S., & Webb, M. (2024). Does Information About AI Regulation Change Manager Evaluation of Ethical Concerns and Intent to Adopt AI? The Journal of Law, Economics, and Organization , 40 (1), 34–75. https://doi.org/10.1093/jleo/ewac004 Dai, J., Zhang, X., & Wang, C. (2024). A meta-analysis of learners’ continuance intention toward online education platforms. Education and Information Technologies, 29, 1–36. https://doi.org/10.1007/s10639-024-12654-7 Dargue, N., Sweller, N., & Jones, M. P. (2019). When our hands help us understand: A meta-analysis into the effects of gesture on comprehension. Psychological Bulletin , 145 (8), 765–784. https://doi.org/10.1037/bul0000202 Darvishi, A., Khosravi, H., Sadiq, S., Gašević, D., & Siemens, G. (2024). Impact of AI assistance on student agency. Computers and Education , 210, 1–18. https://doi.org/10.1016/j.compedu.2023.104967 Davis, F. D., Bagozzi, R. P., & Warshaw, P. R. (1992). Extrinsic and Intrinsic Motivation to Use Computers in the Workplace. Journal of Applied Social Psychology , 22 (14), 1111–1132. https://doi.org/10.1111/j.1559-1816.1992.tb00945.x De Jong, T. (2006). Technological advances in inquiry learning. In Science, 312 (5773), 532–533. https://doi.org/10.1126/science.1127750 Devlin, J., Chang, M. W., Lee, K., & Toutanova, K. (2019). BERT: Pre-training of deep bidirectional transformers for language understanding. In proceedings of NAACL HLT 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, (pp. 1–16) . https://doi.org/10.48550/arXiv.1810.04805 Dillenbourg, P. (1999). Collaborative Learning: Cognitive and Computational Approaches. Advances in Learning and Instruction Series. Elsevier Science , 1–246. https://eric.ed.gov/?id=ED437928 Edwards, B. (2023). OpenAI’s GPT-4 exhibits “human-level performance” on professional benchmarks. Ars Technica. Retrieved from: https://arstechnica.com/information-technology/2023/03/openai-announces-gpt-4-its-next-generation-ai-language-model/ Eshuis, E. H., ter Vrugte, J., Anjewierden, A., & de Jong, T. (2022). Expert examples and prompted reflection in learning with self-generated concept maps. Journal of Computer Assisted Learning , 38 (2), 350–365. https://doi.org/10.1111/jcal.12615 Faul, F., Erdfelder, E., Lang, A.-G., & Buchner, A. (2007). G*Power 3: A flexible statistical power analysis program for the social, behavioral, and biomedical sciences. Behavior Research Methods , 39 , 175–191. https://doi.org/10.3758/BF03193146 Fenn, J., & Blosch, M. (2020). Understanding Gartner’s Hype Cycles. Gartner Research. https://www.gartner.com/en/documents/3887767 Fiorella, L. (2023). Making Sense of Generative Learning. In Educational Psychology Review, 35 (2), 1–42. https://doi.org/10.1007/s10648-023-09769-7 Fiorella, L., & Kuhlmann, S. (2020). Creating Drawings Enhances Learning by Teaching. Journal of Educational Psychology , 112 (4), 811–822. https://doi.org/10.1037/edu0000392 Fiorella, L., & Mayer, R. E. (2013). The relative benefits of learning by teaching and teaching expectancy. Contemporary Educational Psychology, 38 (4), 281–288. https://doi.org/10.1016/j.cedpsych.2013.06.001 Fiorella, L., & Mayer, R. E. (2015). Learning as a generative activity: Eight learning strategies that promote understanding. 1–218. https://doi.org/10.1017/CBO9781107707085 Fiorella, L., & Mayer, R. E. (2016). Eight Ways to Promote Generative Learning. In Educational Psychology Review , 28 (4), 717–741. https://doi.org/10.1007/s10648-015-9348-9 Fiorella, L., & Mayer, R. E. (2017). Spontaneous spatial strategy use in learning from scientific text. Contemporary Educational Psychology , 49 , 66–79. https://doi.org/10.1016/j.cedpsych.2017.01.002 Fiorella, L., & Mayer, R. E. (2022). The Generative Activity Principle in Multimedia Learning, 339–350. In The Cambridge Handbook of Multimedia Learning . https://doi.org/10.1017/9781108894333.036 Fiorella, L., Yoon, S. Y., Atit, K., Power, J. R., Panther, G., Sorby, S., Uttal, D. H., & Veurink, N. (2021). Validation of the Mathematics Motivation Questionnaire (MMQ) for secondary school students. International Journal of STEM Education , 8 (1), 1–14. https://doi.org/10.1186/s40594-021-00307-x Fiorella, L., & Zhang, Q. (2018). Drawing Boundary Conditions for Learning by Drawing. In Educational Psychology Review, 30 (3), 1115–1137. https://doi.org/10.1007/s10648-018-9444-8 Fischer, G., Lundin, J., & Lindberg, J. O. (2020). Rethinking and reinventing learning, education and collaboration in the digital age—from creating technologies to transforming cultures. International Journal of Information and Learning Technology , 37 (5), 241–252. https://doi.org/10.1108/IJILT-04-2020-0051 Goldin-Meadow, S., Cook, S. W., & Mitchell, Z. A. (2009). Gesturing gives children new ideas about math. Psychological Science , 20 (3), 267–272. https://doi.org/10.1111/j.1467-9280.2009.02297.x Gulati, S., Sousa, S., & Lamas, D. (2019). Design, development and evaluation of a human-computer trust scale. Behaviour and Information Technology , 38 (10), 1004–1015. https://doi.org/10.1080/0144929X.2019.1656779 Helmore, E. (2023). We are a little bit scared’: OpenAI CEO warns of risks of artificial intelligence. TheGuardian . Retrieved from: https://www.theguardian.com/technology/2023/mar/17/openai-sam-altman-artificial-intelligence-warning-gpt4 Hennessy, S., Cukurova, M., Lewin, C., Mavrikis, M., & Major, L. (2024). BJET Editorial 2024: A call for research rigour. In British Journal of Educational Technology , 55 (1), 5–9. https://doi.org/10.1111/bjet.13426 Hmelo-Silver, C. E., Duncan, R. G., & Chinn, C. A. (2007). Scaffolding and achievement in problem-based and inquiry learning: A response to Kirschner, Sweller, and Clark (2006). In Educational Psychologist , 42 (2), 99–107. https://doi.org/10.1080/00461520701263368 Hoogerheide, V., Deijkers, L., Loyens, S. M., Heijltjes, A., & van Gog, T. (2016). Gaining from explaining: Learning improves from explaining to fictitious others on video, not from writing to them. Contemporary Educational Psychology, 44, 95–106. https://doi.org/10.1016/j.cedpsych.2016.02.005 Hoogerheide, V., Renkl, A., Fiorella, L., Paas, F., & Van Gog, T. (2019). Enhancing example-based learning: Teaching on video increases arousal and improves problem-solving performance. Journal of Educational Psychology, 111 (1), 45–56. https://doi.org/10.1037/edu0000272 Holmes, W., Fadel, C., & Bialik, M. (2019). Artificial intelligence in education: Promises and implications for teaching and learning. Journal of Computer Assisted Learning , 14 (4). https://www.researchgate.net/publication/332180327_Artificial_Intelligence_in_Education_Promise_and_Implications_for_Teaching_and_Learning Holmes, W., & Tuomi, I. (2022). State of the art and practice in AI in education. European Journal of Education , 57 (4), 542–570. https://doi.org/10.1111/ejed.12533 Huang, F., & Zou, B. (2024). English speaking with artificial intelligence (AI): The roles of enjoyment, willingness to communicate with AI, and innovativeness. Computers in Human Behavior , 159, 2–8. https://doi.org/10.1016/j.chb.2024.108355 IBM Corp. (2023). IBM SPSS Statistics for Windows. https://doi.org/https://www.ibm.com/products/spss-statistics Järvelä, S., Nguyen, A., Vuorenmaa, E., Malmberg, J., & Järvenoja, H. (2023). Predicting regulatory activities for socially shared regulation to optimize collaborative learning. Computers in Human Behavior , 144, 1–10. https://doi.org/10.1016/j.chb.2023.107737 Johnson, C. I., & Marrafno, M. D. (2022). The feedback principle in multimedia learning. In R. E. Mayer & L. Fiorella (Eds.), The Cambridge handbook of multimedia learning (3rd ed., pp. 286– 295). Cambridge University Press. https://doi.org/10.1017/CBO9781139547369.023 Kasneci, E., Sessler, K., Küchemann, S., Bannert, M., Dementieva, D., Fischer, F., Gasser, U., Groh, G., Günnemann, S., Hüllermeier, E., Krusche, S., Kutyniok, G., Michaeli, T., Nerdel, C., Pfeffer, J., Poquet, O., Sailer, M., Schmidt, A., Seidel, T., … Kasneci, G. (2023). ChatGPT for good? On opportunities and challenges of large language models for education. In Learning and Individual Differences , 103, 1–9 . https://doi.org/10.1016/j.lindif.2023.102274 Khosravi, H., Shum, S. B., Chen, G., Conati, C., Tsai, Y. S., Kay, J., Knight, S., Martinez-Maldonado, R., Sadiq, S., & Gašević, D. (2022). Explainable Artificial Intelligence in education. Computers and Education: Artificial Intelligence , 3, 1–22. https://doi.org/10.1016/j.caeai.2022.100074 King, A., Staffieri, A., & Adelgais, A. (1998). Mutual peer tutoring: Effects of structuring tutorial interaction to scaffold peer learning. Journal of Educational Psychology, 90 , 134–152. https://doi.org/10.1037/0022-0663.90.1.134 Kobayashi, K. (2019). Learning by preparing-to-teach and teaching: A meta-analysis. Japanese Psychological Research, 61 (3), 192–203. https://doi.org/10.1111/jpr.12221 Koh, A. W. L., Lee, S. C., & Lim, S. W. H. (2018). The learning benefits of teaching: A retrieval practice hypothesis. Applied Cognitive Psychology, 32 (3), 401–410. https://doi.org/10.1002/acp.3410 Lachner, A., Jacob, L., & Hoogerheide, V. (2021). Learning by writing explanations: Is explaining to a fictitious student more effective than self-explaining? Learning and Instruction , 74, 1–13. https://doi.org/10.1016/j.learninstruc.2020.101438 Lachner, A., & Neuburg, C. (2019). Learning by writing explanations: computer-based feedback about the explanatory cohesion enhances students’ transfer. Instructional Science , 47 (1), 19–37. https://doi.org/10.1007/s11251-018-9470-4 Lee, S. Y., & Choi, J. (2017). Enhancing user experience with conversational agent for movie recommendation: Effects of self-disclosure and reciprocity. International Journal of Human Computer Studies , 103, 95–105. https://doi.org/10.1016/j.ijhcs.2017.02.005 Lombrozo, T. (2006). The structure and function of explanations. Trends in Cognitive Sciences , 10 (10), 464-470. https://doi.org/10.1016/j.tics.2006.08.004 Lorenz, P., Perset, K., & Berryhill, J. (2023). Initial policy considerations for generative artificial intelligence. OECD Artificial intelligence papers , 1, 1–40 . Retrieved from: https://www.oecd.org/en/publications/initial-policy-considerations-for-generative-artificial-intelligence_fae2d1e6-en.html Makransky, G., & Petersen, G. B. (2021). The Cognitive Affective Model of Immersive Learning (CAMIL): a Theoretical Research-Based Model of Learning in Immersive Virtual Reality. Educational Psychology Review, 33 (3), 937–958. https://doi.org/10.1007/s10648-020-09586-2 Manalo, E., Uesaka, Y., & Chinn, C. A. (2017). Promoting Spontaneous Use of Learning and Reasoning Strategies. In Promoting Spontaneous Use of Learning and Reasoning Strategies, 1–166. https://doi.org/10.4324/9781315564029 Mayer, R. E. (1996). Learning strategies for making sense out of expository text: the soi model for guiding three cognitive processes in knowledge construction. Educational Psychology Review , 8 (4), 357–371. https://doi.org/10.1007/BF01463939 Mayer, R. E. (2004). Should There Be a Three-Strikes Rule Against Pure Discovery Learning? American Psychologist , 59 (1), 14–19 . https://doi.org/10.1037/0003-066x.59.1.14 Mayer, R. E. (2012). Getting Started on the Road to Applying the Science of Learning. In Applied Cognitive Psychology , 26 (2), 330–331. https://doi.org/10.1002/acp.1829 Mayer, R. E. (2014). Cognitive Theory of Multimedia Learning (2nd ed). Cambridge University Press, 43–71. https://doi.org/10.1017/CBO9781139547369.005 Mayer, R. E. (2024). The Past, Present, and Future of the Cognitive Theory of Multimedia Learning. Educational Psychology Review , 36 (1), 1–25. https://doi.org/10.1007/s10648-023-09842-1 McNamara, D. S. (2017). Self-Explanation and Reading Strategy Training (SERT) Improves Low-Knowledge Students’ Science Course Performance. Discourse Processes , 54 (7), 479–492. https://doi.org/10.1080/0163853X.2015.1101328 Molenaar, I. (2022). Towards hybrid human-AI learning technologies. European Journal of Education , 57 (4), 632–645. https://doi.org/10.1111/ejed.12527 Molenaar, I., & Järvelä, S. (2014). Sequential and temporal characteristics of self and socially regulated learning. Metacognition and Learning , 9 , 75–85. https://doi.org/10.1007/s11409-014-9114-2 Murayama, K. (2022). A reward-learning framework of knowledge acquisition: An integrated account of curiosity, interest, and intrinsic–extrinsic rewards. Psychological Review , 129 (1), 175. https://doi.org/10.1037/rev0000349 Nie, A., Chandak, Y., Suzara, M., Ali, M., Woodrow, J., Peng, M., Sahami, M., Brunskill, E., & Piech, C. (2024). The GPT Surprise: Offering Large Language Model Chat in a Massive Coding Class Reduced Engagement but Increased Adopters Exam Performances , arXiv, 1–33. http://arxiv.org/abs/2407.09975 Pedaste, M., Mäeots, M., Siiman, L. A., de Jong, T., van Riesen, S. A. N., Kamp, E. T., Manoli, C. C., Zacharia, Z. C., & Tsourlidaki, E. (2015). Phases of inquiry-based learning: Definitions and the inquiry cycle. Educational Research Review , 14, 47–61. https://doi.org/10.1016/j.edurev.2015.02.003 Pekrun, R. (2006). The control-value theory of achievement emotions: Assumptions, corollaries, and implications for educational research and practice. Educational psychology review , 18 , 315–341. https://doi.org/10.1007/s10648-006-9029-9 Pintrich, P. R. R., Smith, D., Garcia, T., & McKeachie, W. (1991). A manual for the use of the Motivated Strategies for Learning Questionnaire (MSLQ). ERIC, 1–76. Retrieved from: https://files.eric.ed.gov/fulltext/ED338122.pdf Ponce, H. R., Mayer, R. E., Loyola, M. S., & López, M. J. (2020). Study Activities That Foster Generative Learning: Notetaking, Graphic Organizer, and Questioning. Journal of Educational Computing Research , 58 (2), 1–22. https://doi.org/10.1177/0735633119865554 Qadir, J. (2023). Engineering Education in the Era of ChatGPT: Promise and Pitfalls of Generative AI for Education. In IEEE Global Engineering Education Conference, EDUCON , 1–9. https://doi.org/10.1109/EDUCON54358.2023.10125121 Rafner, J., Beaty, R. E., Kaufman, J. C., Lubart, T., & Sherson, J. (2023). Creativity in the age of generative AI. Nature Human Behaviour , 7 (11), 1836–1838. https://doi.org/10.1038/s41562-023-01751-1 Rellensmann, J., Schukajlow, S., Blomberg, J., & Leopold, C. (2022). Effects of drawing instructions and strategic knowledge on mathematical modeling performance: Mediated by the use of the drawing strategy. Applied Cognitive Psychology , 36 (2), 402–417. https://doi.org/10.1002/acp.3930 Renkl, A. (1997). Learning from worked-out examples: A study on individual differences. Cognitive Science , 21 (1), 1–29. https://doi.org/10.1207/s15516709cog2101_1 Renkl, A., & Scheiter, K. (2017). Studying Visual Displays: How to Instructionally Support Learning. Educational Psychology Review , 29 (3), 599–621. https://doi.org/10.1007/s10648-015-9340-4 Roediger III, H. L., & Karpicke, J. D. (2006). Test-enhanced learning: Taking memory tests improves long-term retention. Psychological Science, 17 (3), 249–255. https://doi.org/10.1111/j.1467-9280.2006.01693.x Roelle, J., & Berthold, K. (2017). Effects of incorporating retrieval into learning tasks: The complexity of the tasks matters. Learning and Instruction , 49, 142–156. https://doi.org/10.1016/j.learninstruc.2017.01.008 Roschelle, J., Lester, J., Fusco, J., Safir, A., Johnstun, K., Trettin, S., Chhin, C., Metz, E., Digital Promise colleagues, O., Cator, K., Means, B., Bellin, M., & Van Ostrand, K. (2020). AI and the Future of Learning: Expert Panel Report. Digital Promise, 1–27. Retrived from: https://circls.org/wp-content/uploads/2020/11/CIRCLS-AI-Report-Nov2020.pdf Roscoe, R. D., & Chi, M. T. H. (2007). Understanding tutor learning: Knowledge-building and knowledge-telling in peer tutors’ explanations and questions. Review of Educational Research , 77 (4), 534–574. https://doi.org/10.3102/0034654307309920 Schmeck, A., Mayer, R. E., Opfermann, M., Pfeiffer, V., & Leutner, D. (2014). Drawing pictures during learning from scientific text: testing the generative drawing effect and the prognostic drawing effect. Contemporary Educational Psychology , 39 (4), 275–286. https://doi.org/10.1016/j.cedpsych.2014.07.003 Shneiderman, B. (2020). Human-centered artificial intelligence: reliable, safe & trustworthy. Int. J. Hum. Comput. Interact. 36, 495–504. https://doi.org/10.1080/10447318.2020.1741118 Schramowski, P., Turan, C., Andersen, N., Rothkopf, C. A., & Kersting, K. (2022). Large pre-trained language models contain human-like biases of what is right and wrong to do. Nature Machine Intelligence , 4 (3), 258–268. https://doi.org/10.1038/s42256-022-00458-8 Schukajlow, S., Krawitz, J., Kanefke, J., & Rakoczy, K. (2022). Interest and performance in solving open modeling problems and closed real-world problems. PME, 3. Retrieved from: https://ivv5hpp.uni-muenster.de/u/sschu_12/pdf/Publikationen/Schukajlow_etal_2022_PME45.pdf Schwamborn, A., Mayer, R. E., Thillmann, H., Leopold, C., & Leutner, D. (2010). Drawing as a Generative Activity and Drawing as a Prognostic Activity. Journal of Educational Psychology , 102 (4), 872–879. https://doi.org/10.1037/a0019640 Simonsmeier, B. A., Flaig, M., Deiglmayr, A., Schalk, L., & Schneider, M. (2022). Domain-specific prior knowledge and learning: A meta-analysis. Educational Psychologist , 57 (1), 31–54. https://doi.org/10.1080/00461520.2021.1939700 Sternberg, S. (1966). High-speed scanning in human memory. Science, 153 (3736), 652–654. https://doi.org/10.1126/science.153.3736.652 Sternberg, S. (1969). Memory-scanning: Mental processes revealed by reaction-time experiments. American Scientist, 57(4), 421–457. Retrieved from: https://www.sas.upenn.edu/~saul/am.scientist69.pdf Sweller, J., van Merriënboer, J. J. G., & Paas, F. (2019). Cognitive Architecture and Instructional Design: 20 Years Later. In Educational Psychology Review , 31 (2), 261–292. https://doi.org/10.1007/s10648-019-09465-5 Taecharungroj, V. (2023). “What Can ChatGPT Do?” Analyzing Early Reactions to the Innovative AI Chatbot on Twitter. Big Data and Cognitive Computing , 7 (1), 1–10. https://doi.org/10.3390/bdcc7010035 Touvron, H., Lavril, T., Izacard, G., Martinet, X., Lachaux, M., Lacroix, T., Rozière, B., Goyal, N., Hambro, E., Azhar, F., Rodriguez, A., Joulin, A., Grave, E., & Lample, G. (2023). LLAMA: Open and Efficient Foundation Language Models. ArXi, 1–17. https://doi.org/https://doi.org/10.48550/arXiv.2302.13971 van de Pol, J., van Loon, M., van Gog, T., Braumann, S., & de Bruin, A. (2020). Mapping and Drawing to Improve Students’ and Teachers’ Monitoring and Regulation of Students’ Learning from Text: Current Findings and Future Directions. In Educational Psychology Review , 32 (4), 951–977. https://doi.org/10.1007/s10648-020-09560-y Van Meter, P. (2001). Drawing construction as a strategy for learning from text. Journal of Educational Psychology , 93 (1), 129–977. https://doi.org/10.1037/0022-0663.93.1.129 Venkatesh, V., & Bala, H. (2008). Technology acceptance model 3 and a research agenda on interventions. Decision sciences , 39 (2), 273–315. https://doi.org/10.1111/j.1540-5915.2008.00192.x Vygotsky, L. S. (1978). Mind and Society: The Development of Higher Psychological Processes. In Harvard University Press, 1–174. https://doi.org/10.2307/j.ctvjf9vz4 Willingham, D. T. (2008). Critical Thinking: Why Is It So Hard to Teach? Arts Education Policy Review , 109 (4), 21–32. https://doi.org/10.3200/AEPR.109.4.21-32 Wittrock, M. C. (1989). Generative Processes of Comprehension. Educational Psychologist , 24 (4), 345–275. https://doi.org/10.1207/s15326985ep2404_2 Wittrock, M. C., & Farley, F. (2010). Learning as a generative process. Educational Psychologist , 45 (1), 40–45. https://doi.org/10.1080/00461520903433554 Wittwer, J., & Renkl, A. (2010). How Effective are Instructional Explanations in Example-Based Learning? A Meta-Analytic Review. In Educational Psychology Review, 22 (4), 393–409. https://doi.org/10.1007/s10648-010-9136-5 Wu, R., & Yu, Z. (2024). Do AI chatbots improve students learning outcomes? Evidence from a meta‐analysis. British Journal of Educational Technology , 55 (1), 10–33. https://doi.org/10.1111/bjet.13334 Yan, L., Greiff, S., Teuber, Z., & Gašević, D. (2024). Promises and challenges of generative artificial intelligence for human learning. Nature Human Behavior, 8, 1839–1850. https://doi.org/10.1038/s41562-024-02004-5 Yan, L., Sha, L., Zhao, L., Li, Y., Martinez-Maldonado, R., Chen, G., Li, X., Jin, Y., & Gašević, D. (2023). Practical and ethical challenges of large language models in education: A systematic scoping review. In British Journal of Educational Technology , 55 (1), 90–112. https://doi.org/10.1111/bjet.13370 Zhang, Q., & Fiorella, L. (2023). An integrated model of learning from errors. Educational Psychologist , 58 (1), 18–34. https://doi.org/10.1080/00461520.2022.2149525 Table Table 1 is available in the Supplementary Files section Additional Declarations The authors declare potential competing interests as follows: Tue Herlau is a Researcher at The Danish Technical University and Founder of ChatTutor Supplementary Files SupplamentaryMaterial.docx Table1.docx Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-5622133","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":388924953,"identity":"43ef221d-bdf1-4cbd-b13d-a3ecf536b9dc","order_by":0,"name":"Guido Makransky","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAApUlEQVRIiWNgGAWjYFAC5gMg0gCIjYnVwpZAshYeAxK1yIed+fbwZxuDscEB5s0GRGkxvJ273Zi3jcHM4ABbcQJxWmbnbpNmbGOwMTjAY3yASC05zyR/kqRFXjqHTQLiMB5j4hxmIJ1mbsxzTsJY8jBbMXHel5+d/OzhjzIbw77jzZsliLPlAAMbAyMbUDEzUepBtjQAtTD8IVb5KBgFo2AUjEgAAATGKb3kaRUpAAAAAElFTkSuQmCC","orcid":"https://orcid.org/0000-0003-1862-7824","institution":"University of Copenhagen","correspondingAuthor":true,"prefix":"","firstName":"Guido","middleName":"","lastName":"Makransky","suffix":""},{"id":388924954,"identity":"719c9cbb-e4c6-425d-9a59-368667c10ab9","order_by":1,"name":"Ban M. Shiwalia","email":"","orcid":"https://orcid.org/0000-0001-8057-5303","institution":"University of Copenhagen","correspondingAuthor":false,"prefix":"","firstName":"Ban","middleName":"M.","lastName":"Shiwalia","suffix":""},{"id":388924955,"identity":"f188c6e6-ef67-4a30-85fe-2e13bfcf6363","order_by":2,"name":"Tue Herlau","email":"","orcid":"https://orcid.org/0000-0001-7288-6953","institution":"Danish Technical University","correspondingAuthor":false,"prefix":"","firstName":"Tue","middleName":"","lastName":"Herlau","suffix":""},{"id":388924956,"identity":"4e19e5f1-41c3-43fc-8fec-5286a8efb3b5","order_by":3,"name":"Steven Blurton","email":"","orcid":"https://orcid.org/0000-0003-4311-0202","institution":"University of Copenhagen","correspondingAuthor":false,"prefix":"","firstName":"Steven","middleName":"","lastName":"Blurton","suffix":""}],"badges":[],"createdAt":"2024-12-11 07:56:42","currentVersionCode":1,"declarations":{"humanSubjects":true,"vertebrateSubjects":false,"conflictsOfInterestStatement":true,"humanSubjectEthicalGuidelines":true,"humanSubjectConsent":true,"humanSubjectClinicalTrial":false,"humanSubjectCaseReport":false,"vertebrateSubjectEthicalGuidelines":false},"doi":"10.21203/rs.3.rs-5622133/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-5622133/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":71197155,"identity":"0074fc37-3dcf-47e2-9472-8658c74845e3","added_by":"auto","created_at":"2024-12-12 05:27:51","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":65137,"visible":true,"origin":"","legend":"\u003cp\u003eFramework for the development of a student-focused GenAI educational application where interaction is initiated by a) the learner, b) the AIEd system, and c) the instructor, adapted from the taxonomy developed by Holmes and Tuomi (2022).\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-5622133/v1/6ad52688989dffa1d50c987c.png"},{"id":71197154,"identity":"e08f768d-2964-4068-af46-c1094e8117c1","added_by":"auto","created_at":"2024-12-12 05:27:51","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":91827,"visible":true,"origin":"","legend":"\u003cp\u003eFigure of experimental procedure\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-5622133/v1/68bcfb542c13133cb8779fbd.png"},{"id":71200274,"identity":"14115f71-cfc2-45ed-8b64-98ae956f38fb","added_by":"auto","created_at":"2024-12-12 06:05:47","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":625385,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-5622133/v1/5ad2e07c-acc6-4dc6-a58a-a6193fff19f8.pdf"},{"id":71197156,"identity":"963f820b-5bcf-4520-a43d-8231bec71377","added_by":"auto","created_at":"2024-12-12 05:27:51","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":1861735,"visible":true,"origin":"","legend":"","description":"","filename":"SupplamentaryMaterial.docx","url":"https://assets-eu.researchsquare.com/files/rs-5622133/v1/e2a9485cea83cdcd8bc4544b.docx"},{"id":71197391,"identity":"cef0c1a3-cabe-4ddc-ab4e-c40b5bb32d57","added_by":"auto","created_at":"2024-12-12 05:35:51","extension":"docx","order_by":2,"title":"","display":"","copyAsset":false,"role":"supplement","size":16507,"visible":true,"origin":"","legend":"","description":"","filename":"Table1.docx","url":"https://assets-eu.researchsquare.com/files/rs-5622133/v1/a94a362108f9695aa2a47029.docx"}],"financialInterests":"The authors declare potential competing interests as follows: Tue Herlau is a Researcher at The Danish Technical University and Founder of ChatTutor","formattedTitle":"\u003cp\u003e\u003cstrong\u003eBeyond the \"Wow\" factor: Using Generative AI for Increasing Generative Sense-Making\u003c/strong\u003e\u003c/p\u003e","fulltext":[{"header":"Introduction","content":"\u003cp\u003eGenerative Artificial Intelligence (GenAI) has received increasing public interest and research (Kasneci et al., \u003cspan class=\"CitationRef\"\u003e2023\u003c/span\u003e; Lorenz et al., \u003cspan class=\"CitationRef\"\u003e2023\u003c/span\u003e; Yan et al., \u003cspan class=\"CitationRef\"\u003e2024\u003c/span\u003e). The widespread use of GenAI has surged with the release of publicly available large language models such as BERT (Devlin et al., \u003cspan class=\"CitationRef\"\u003e2019\u003c/span\u003e), Llama (Touvron et al., \u003cspan class=\"CitationRef\"\u003e2023\u003c/span\u003e), OpenAI (Helmore, \u003cspan class=\"CitationRef\"\u003e2023\u003c/span\u003e), Claude (Anthopic, \u003cspan class=\"CitationRef\"\u003e2024\u003c/span\u003e), and GPT-4 (Edwards, \u003cspan class=\"CitationRef\"\u003e2023\u003c/span\u003e). These models generate text, images, audio, and structured synthetic data through user-friendly interfaces, while also producing human-like responses (Cu\u0026eacute;llar et al., \u003cspan class=\"CitationRef\"\u003e2024\u003c/span\u003e; Schramowski et al., \u003cspan class=\"CitationRef\"\u003e2022\u003c/span\u003e). Compared to predictive AI, which can be used for predicting labels or values, GenAI is designed to create new content or ideas and express them in real-time conversations (Qadir, \u003cspan class=\"CitationRef\"\u003e2023\u003c/span\u003e). GenAI has taken society by storm, in the history of the internet, no consumer app has grown faster than OpenAI\u0026apos;s ChatGPT (Chow, \u003cspan class=\"CitationRef\"\u003e2023\u003c/span\u003e). The experience of asking a Chatbot anything and receiving a coherent, face valid answer has been described by users as \u0026quot;mind-blowing\u0026quot;, \u0026quot;impressive\u0026quot;, and \u0026quot;amazing\u0026quot; (Taecharungroj, \u003cspan class=\"CitationRef\"\u003e2023\u003c/span\u003e). This has led many to anticipate that this technology will inevitably revolutionize society, reshaping the ways we work, live, and learn (Lorenz et al., \u003cspan class=\"CitationRef\"\u003e2023\u003c/span\u003e). In education this could change how teachers teach, how learners learn and are assessed, and how education is administered (Chiu et al., \u003cspan class=\"CitationRef\"\u003e2023\u003c/span\u003e). However, emerging technologies typically progress through different levels of hype, and initial expectations are often exaggerated due to media coverage (Fenn \u0026amp; Blosch, \u003cspan class=\"CitationRef\"\u003e2020\u003c/span\u003e). While it can be argued that some emerging technologies such as computers have eventually had a significant influence on education, these changes are usually slower than initially expected and may be detrimental to the learning process (Chandler, \u003cspan class=\"CitationRef\"\u003e2009\u003c/span\u003e). In general, the digitalization of society results in challenges and opportunities for learning and education (Fischer et al., \u003cspan class=\"CitationRef\"\u003e2020\u003c/span\u003e; Yan et al., \u003cspan class=\"CitationRef\"\u003e2024\u003c/span\u003e).\u003c/p\u003e\n\u003cp\u003eA common challenge with the introduction of novel technology is that it is often implemented and examined from a technocentric perspective, without incorporating extensive research and theory from educational psychology and the science of learning (Chandler \u003cspan class=\"CitationRef\"\u003e2009\u003c/span\u003e; Brennan, \u003cspan class=\"CitationRef\"\u003e2015\u003c/span\u003e; Yan et al., \u003cspan class=\"CitationRef\"\u003e2024\u003c/span\u003e). In this study, we argue that it is important to look beyond the \u0026quot;wow\u0026quot; factor of GenAI in education and adopt a learner-centered approach to investigating how the technology can enhance rather than distract from learning. Here, we agree with the perspective that psychological theory is central to understanding how GenAI can be used in this endeavor (e.g. Molenaar, \u003cspan class=\"CitationRef\"\u003e2022\u003c/span\u003e; Yan et al., \u003cspan class=\"CitationRef\"\u003e2024\u003c/span\u003e). Specifically, we use generative learning theory (Fiorella, \u003cspan class=\"CitationRef\"\u003e2023\u003c/span\u003e; Fiorella \u0026amp; Mayer, \u003cspan class=\"CitationRef\"\u003e2016\u003c/span\u003e) to design a GenAI educational application that supports human-AI collaboration in promoting generative processing. Generative learning activities (GLAs) are learner-driven actions that foster the active construction of knowledge as they prime key cognitive processes such as selecting, organizing, and integrating knowledge (Fiorella \u0026amp; Mayer, \u003cspan class=\"CitationRef\"\u003e2015\u003c/span\u003e, \u003cspan class=\"CitationRef\"\u003e2022\u003c/span\u003e; Mayer, \u003cspan class=\"CitationRef\"\u003e1996\u003c/span\u003e; Wittrock \u0026amp; Farley, \u003cspan class=\"CitationRef\"\u003e2010\u003c/span\u003e). By aligning GLAs with the capabilities of GenAI, we explore how integrating human and machine intelligence can enhance learning outcomes, augmenting rather than replacing human abilities (Molenaar, \u003cspan class=\"CitationRef\"\u003e2022\u003c/span\u003e). That is, we adopt an augmentation perspective, viewing AI not as a simulation of human intelligence but as a distinct form of intelligence that complements our own, serving as a tool to potentially enhance human capabilities and help us function better as humans (Molenaar, \u003cspan class=\"CitationRef\"\u003e2022\u003c/span\u003e). Combining human and AI capabilities through hybrid intelligence has been proposed as a key to advancing the learning sciences in an AI era (J\u0026auml;rvel\u0026auml; et al., \u003cspan class=\"CitationRef\"\u003e2023\u003c/span\u003e). But systematic research that builds on examining this integration is scarce (Hennessy et al., \u003cspan class=\"CitationRef\"\u003e2024\u003c/span\u003e; Roschelle et al., \u003cspan class=\"CitationRef\"\u003e2020\u003c/span\u003e). Ideally, hybrid intelligence systems should optimize human strengths while compensating for their weaknesses.\u003c/p\u003e\n\u003cp\u003eIn a world where knowledge is abundant and easily accessible, the focus of education could shift from imparting knowledge to nurturing students\u0026apos; ability to ask meaningful questions and critically evaluate information. A meta-analysis on the effect of inducing self-explanation found that having learners generate an explanation is often more effective than presenting them with an explanation (Birsa et al., 2018). Its impact was greatest when learners used new information to revise their explanations. However, while instructor-scripted self-explanation prompts work well for standardized content, they are impractical for highly personalized learning, and the authors recommend investigating computer-generated, content-specific prompts. GenAI makes this possible at scale, and in this study we investigate its potential to foster higher-order cognitive processes, such as conceptual understanding (Anderson \u0026amp; Krathwohl, \u003cspan class=\"CitationRef\"\u003e2001\u003c/span\u003e), while also promoting self-efficacy (Bandura, \u003cspan class=\"CitationRef\"\u003e1977\u003c/span\u003e), trust (Gulati et al., \u003cspan class=\"CitationRef\"\u003e2019\u003c/span\u003e), enjoyment (Davis et al., \u003cspan class=\"CitationRef\"\u003e1992\u003c/span\u003e; Huang \u0026amp; Zou, \u003cspan class=\"CitationRef\"\u003e2024\u003c/span\u003e), and behavioral intentions (Lee \u0026amp; Choi, \u003cspan class=\"CitationRef\"\u003e2017\u003c/span\u003e).\u003c/p\u003e\n\u003cp\u003eMethodological rigor has been recently highlighted as a serious challenge to the research investigating GenAI in education (Yan et al., \u003cspan class=\"CitationRef\"\u003e2024\u003c/span\u003e). A criticism is that the majority of research on GenAI in education is conceptual, focusing on literature reviews, analyses of public discourse, or surveys of people\u0026apos;s perceptions (Hennessy et al., \u003cspan class=\"CitationRef\"\u003e2024\u003c/span\u003e, p. 6). Emerging evidence also suggests that the impact on learning and engagement presents a complex picture, such as instances where GenAI may have a positive initial impact on learning outcomes, however, it can act as a crutch and the removal of GenAI may ultimately hinder learning (Darvishi et al., \u003cspan class=\"CitationRef\"\u003e2024\u003c/span\u003e; Nie et al., \u003cspan class=\"CitationRef\"\u003e2024\u003c/span\u003e). This has prompted calls for more experimental research in authentic educational settings, grounded in robust theoretical frameworks, as well as for longitudinal studies to evaluate GenAI\u0026apos;s long-term benefits to human learning by comparing its effectiveness with conventional methods (e.g. Hennessy et al., \u003cspan class=\"CitationRef\"\u003e2024\u003c/span\u003e; Yan et al., \u003cspan class=\"CitationRef\"\u003e2024\u003c/span\u003e). Furthermore, for GenAI to enhance rather than detract from learning, collaboration among researchers, practitioners, GenAI developers, policymakers, and educators is crucial to ensure its effective and responsible integration into teaching, aligning with educational goals (Yan et al., \u003cspan class=\"CitationRef\"\u003e2024\u003c/span\u003e).\u003c/p\u003e\n\u003cp\u003eThe first goal of this paper is to use theories from educational psychology and technology-enhanced learning to provide an understanding of how GenAI can effectively be incorporated into a lesson. While many theories could be used as an approach for using GenAI in education, in this article we focus on generative learning theory (Fiorella, \u003cspan class=\"CitationRef\"\u003e2023\u003c/span\u003e; Fiorella \u0026amp; Mayer, \u003cspan class=\"CitationRef\"\u003e2016\u003c/span\u003e). The second goal of the paper is to empirically test if the use of GenAI in education implemented to facilitate generative learning can increase conceptual knowledge, self-efficacy and trust immediately after a lesson, as well as conceptual knowledge, enjoyment and behavior intentions in a follow-up test four weeks later. In an experiment conducted in an authentic university setting, we compare the use of a theoretically informed GenAI application with a standard GenAI application and a teaching-as-usual condition. Prior to describing this experiment in more detail, we begin with the first objective and provide a theoretical overview of how GenAI can be used to facilitate generative learning.\u003c/p\u003e"},{"header":"Theoretical Background","content":"\u003ch2\u003eGenerative Learning\u003c/h2\u003e\n\u003cp\u003eGenerative learning involves \u0026quot;making sense\u0026quot; of learning material by actively organizing and integrating it with existing knowledge (Wittrock, \u003cspan class=\"CitationRef\"\u003e1989\u003c/span\u003e). According to Fiorella (\u003cspan class=\"CitationRef\"\u003e2023\u003c/span\u003e, p. 2) \u0026quot;\u003cem\u003ea generative learning activity (GLA) is something a learner does to try to make sense of what they are learning\u003c/em\u003e\u0026quot;. There is evidence that learners do not spontaneously engage in sense-making when learning from text (e.g. Fiorella \u0026amp; Mayer, \u003cspan class=\"CitationRef\"\u003e2017\u003c/span\u003e), visualizations (e.g. Makransky \u0026amp; Petersen, \u003cspan class=\"CitationRef\"\u003e2021\u003c/span\u003e) or examples (Renkl, \u003cspan class=\"CitationRef\"\u003e1997\u003c/span\u003e). Prompting learners to engage in GLAs is therefore a potential remedy for achieving meaningful and durable learning outcomes (Fiorella, \u003cspan class=\"CitationRef\"\u003e2023\u003c/span\u003e). Fiorella and Mayer (\u003cspan class=\"CitationRef\"\u003e2015\u003c/span\u003e, \u003cspan class=\"CitationRef\"\u003e2016\u003c/span\u003e) identified eight generative learning strategies (GLS) including: summarizing, mapping, drawing, imagining, self-testing, self-explaining, teaching, and enacting. Reviews and meta-analyses generally show that GLS support learning but each is susceptible to potential boundary conditions related to learner characteristics, learning materials, and support levels (Bisra et al., \u003cspan class=\"CitationRef\"\u003e2018\u003c/span\u003e; Brod, \u003cspan class=\"CitationRef\"\u003e2021\u003c/span\u003e; Dargue et al., \u003cspan class=\"CitationRef\"\u003e2019\u003c/span\u003e; Fiorella \u0026amp; Zhang, \u003cspan class=\"CitationRef\"\u003e2018\u003c/span\u003e; Lachner et al., \u003cspan class=\"CitationRef\"\u003e2021\u003c/span\u003e). That is, some learners need extensive guidance (Fiorella \u0026amp; Zhang, \u003cspan class=\"CitationRef\"\u003e2018\u003c/span\u003e), and in this study we investigate if GenAI can be used to provide personalized guidance that can improve educational outcomes.\u003c/p\u003e\n\u003cp\u003eIn the current study we are specifically interested in the verbalizing generative learning activities of teaching which incorporates elements of self-explaining. Generating explanations during learning can activate prior knowledge, integrate new information, enhance memory, and encourage reasoning by prompting students to make inferences and revise mental models (Brod, \u003cspan class=\"CitationRef\"\u003e2021\u003c/span\u003e). A recent meta-analysis (Bisra et al., \u003cspan class=\"CitationRef\"\u003e2018\u003c/span\u003e) found a mean effect size of 0.55 SDs for self-explanation, comparing conditions with and without self-explanation prompts. When considering only studies in which time on task was equated, the effect size was still 0.41 SDs. Generating explanations during learning can be done through self-explanation, which involves generating verbal statements (involving inferences) to clarify the meaning of the learning material to oneself (Fiorella \u0026amp; Mayer, \u003cspan class=\"CitationRef\"\u003e2022\u003c/span\u003e, p. 341). Fiorella and Mayer\u0026apos;s (2015) review found positive effects of self-explaining in 44 of 54 experiments, with a median effect size of 0.61.\u003c/p\u003e\n\u003cp\u003eSelf-explaining through teaching involves generating verbal statements (involving inferences) to convey the meaning of the learning material to others (Fiorella \u0026amp; Mayer, \u003cspan class=\"CitationRef\"\u003e2022\u003c/span\u003e, p. 341). Fiorella and Mayer\u0026apos;s (2015) review reported positive effects of learning by teaching in 17 of 19 experiments, with a median effect size of d\u0026thinsp;=\u0026thinsp;.71. A meta-analysis by Kobayashi (\u003cspan class=\"CitationRef\"\u003e2019\u003c/span\u003e) found stronger effects for students who actually taught (g\u0026thinsp;=\u0026thinsp;.56) compared to those who only prepared to teach (g\u0026thinsp;=\u0026thinsp;.35). Explaining to others adds social elements, including a sense of presence and opportunities for interaction, for instance by answering questions.\u003c/p\u003e\n\u003cp\u003eEvidence suggests that explaining is most effective when done aloud, without access to learning materials, when students possess background knowledge, and when they respond to thought-provoking questions (Fiorella \u0026amp; Mayer, \u003cspan class=\"CitationRef\"\u003e2022\u003c/span\u003e). Supporting factors include scaffolded peer interactions (King, Staffieri, \u0026amp; Adelgais, \u003cspan class=\"CitationRef\"\u003e1998\u003c/span\u003e), focused prompts (Berthold, Eysink, \u0026amp; Renkl, \u003cspan class=\"CitationRef\"\u003e2009\u003c/span\u003e), and explicit training (McNamara, 2004). However, most of this research has been conducted without the ability to provide adaptive specific feedback that is now possible with GenAI. In this study, we investigate the GSL of teaching, building on broader research in self-explanation, generative learning, AI-based education (AIEd), among other related fields.\u003c/p\u003e\n\u003cp\u003e\u003cspan type=\"Underline\" class=\"Underline\" name=\"Emphasis\"\u003eHow can the use of Theoretically Informed GenAI Increase Generative Sense-Making?\u003c/span\u003e\u003c/p\u003e\n\u003cp\u003eIt is possible for instructors to use GLA products as formative assessment and adapt instruction and guidance as necessary (van de Pol et al., \u003cspan class=\"CitationRef\"\u003e2020\u003c/span\u003e). For instance, when learners\u0026apos; GLA products show knowledge gaps or misconceptions, instructors can give targeted feedback to aid their internal sense making, improve the GLA products, and enhance learning outcomes (Fiorella, \u003cspan class=\"CitationRef\"\u003e2023\u003c/span\u003e). The challenge is that assessing knowledge gaps and misconceptions and then providing targeted feedback is time consuming and difficult in most educational settings. The question that we attempt to investigate in this study is if a GenAI Chatbot can be used as a tool to scaffold learning by providing personalized feedback to stimulate and support generative learning activities.\u003c/p\u003e\n\u003cp\u003eAccording to a taxonomy developed by Holmes and Tuomi (\u003cspan class=\"CitationRef\"\u003e2022\u003c/span\u003e), AIEd applications can be classified into those that are student-focused, teacher-focused, or institution-focused. This study exclusively deals with student-focused AIEd applications that are leveraged to support student learning. Holmes \u0026amp; Tuomi (\u003cspan class=\"CitationRef\"\u003e2022\u003c/span\u003e) also distinguish between AI-tools designed \u003cem\u003efor\u003c/em\u003e students and those that are used \u003cem\u003eby\u003c/em\u003e students. In this study we compare a theoretically informed GenAI-tool designed for students (ChatTutor) and a GenAI-tool used by students (ChatGPT) to a teaching-as-usual condition. ChatTutor builds on generative learning theory, and technology enhanced learning evidence and theory (e.g., Fiorella, \u003cspan class=\"CitationRef\"\u003e2023\u003c/span\u003e; Fiorella \u0026amp; Mayer, \u003cspan class=\"CitationRef\"\u003e2022\u003c/span\u003e; Mayer, \u003cspan class=\"CitationRef\"\u003e2014\u003c/span\u003e). Figure\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003e depicts a framework for the development of a student-focused GenAI educational application where interaction is initiated by a) the learner, b) the AIEd system, and c) the instructor. In this study both ChatTutor and ChatGPT allow for interaction initiated by the learner. However, building on the research from generative learning (Fiorella, \u003cspan class=\"CitationRef\"\u003e2023\u003c/span\u003e; Fiorella \u0026amp; Mayer, \u003cspan class=\"CitationRef\"\u003e2016\u003c/span\u003e), we have furthermore designed ChatTutor to prompt students to engage in GLA and provide formative feedback. Effective guidance involves adapting learning materials and GLA support to learner\u0026apos;s knowledge and beliefs, thereby enhancing their internal sense making and external products (Fiorella, \u003cspan class=\"CitationRef\"\u003e2023\u003c/span\u003e). We suggest that GenAI based learning systems such as ChatTutor can use the quality of these products as formative assessments to inform further adjustments of learning materials and GLA support. This is the case because in addition to providing responses to prompts that are initiated by the learner, learners who do not automatically initiate sense-making can systematically be prompted to do so by the GenAI system. We will use existing literature related to GLS to propose how this can be done in the following sections.\u003c/p\u003e\n\u003cp\u003e\u003cspan type=\"Underline\" class=\"Underline\" name=\"Emphasis\"\u003eUsing a Theoretically Informed GenAI Chatbot for Prompting and Guiding GLA.\u003c/span\u003e\u003c/p\u003e\n\u003cp\u003eThere is evidence that for novices, complete discovery without guidance is ineffective; and some level of scaffolding is necessary (Mayer, \u003cspan class=\"CitationRef\"\u003e2004\u003c/span\u003e; Pedaste et al., \u003cspan class=\"CitationRef\"\u003e2015\u003c/span\u003e). Therefor for student-centered learning to be effective, guided discovery is essential. Ideally, this guidance must be personalized, which can be facilitated through technological advances (De Jong, \u003cspan class=\"CitationRef\"\u003e2006\u003c/span\u003e). GenAI systems can provide valuable feedback, and the focus should include the knowledge students possess as well as the questions they ask (Berlyne, \u003cspan class=\"CitationRef\"\u003e2006\u003c/span\u003e; Dillenbourg, \u003cspan class=\"CitationRef\"\u003e1999\u003c/span\u003e). The challenge is to support students in asking the right questions, finding accurate answers, and critically analyzing the information they receive to further investigate and understand the knowledge (Hmelo-Silver et al., \u003cspan class=\"CitationRef\"\u003e2007\u003c/span\u003e). Many guided learning principles including problem-based learning and inquiry learning, extensively use scaffolding to reduce cognitive load, enabling students to learn in complex domains (Hmelo-Silver et al., \u003cspan class=\"CitationRef\"\u003e2007\u003c/span\u003e). Fiorella (\u003cspan class=\"CitationRef\"\u003e2023\u003c/span\u003e) suggests that GLA support can be categorized into three broad levels: explicit instruction, scaffolded practice, and independent practice which align with research on GLAs (Fiorella \u0026amp; Zhang, \u003cspan class=\"CitationRef\"\u003e2018\u003c/span\u003e; McNamara, \u003cspan class=\"CitationRef\"\u003e2017\u003c/span\u003e) and cognitive load theory (Sweller et al., \u003cspan class=\"CitationRef\"\u003e2019\u003c/span\u003e), progressing from worked examples to partial problems to problem-solving practice.\u003c/p\u003e\n\u003cp\u003eExplicit instruction includes explanations and demonstrations on selecting and performing GLAs. Explicit prompts help learners to generate internal self-explanations (Fiorella, \u003cspan class=\"CitationRef\"\u003e2023\u003c/span\u003e). The literature on worked examples suggests that this is necessary because learners often struggle to detect errors (Van Meter, \u003cspan class=\"CitationRef\"\u003e2001\u003c/span\u003e). They may need support in generating internal feedback (e.g., self-explaining why their response was inaccurate) and revising their knowledge (Zhang \u0026amp; Fiorella, \u003cspan class=\"CitationRef\"\u003e2023\u003c/span\u003e). Wittwer and Renkl (\u003cspan class=\"CitationRef\"\u003e2010\u003c/span\u003e) describe how effective instruction involves a balanced combination of providing support, such as worked solution steps, and encouraging learner activity, like self-explanation prompts. An experienced instructor can provide this balance by encouraging learners to engage in a GLA such as self-explaining, or teaching, and then help them detect gaps in their knowledge thereby stimulating further inquiry. This requires the ability to engage learners in generative processing, and the knowledge to recognize and highlight gaps in a way that motivates students to engage in further information seeking (Murayama, \u003cspan class=\"CitationRef\"\u003e2022\u003c/span\u003e). A GenAI chatbot can be programmed to provide specific self-explanation prompts, trained using a targeted knowledge bank, and interact with learners as a knowledgeable teacher, an overconfident peer, or a curious collaborator.\u003c/p\u003e\n\u003cp\u003eOnce learners become more familiar with the course content then GenAI based chatbots can progress to scaffolded practice activities. Scaffolded practice includes providing specific prompts or hints (Berthold \u0026amp; Renkl, \u003cspan class=\"CitationRef\"\u003e2009\u003c/span\u003e; Roelle \u0026amp; Berthold, \u003cspan class=\"CitationRef\"\u003e2017\u003c/span\u003e) or partial representations for learners to complete (Fiorella et al., \u003cspan class=\"CitationRef\"\u003e2021\u003c/span\u003e; Ponce et al., \u003cspan class=\"CitationRef\"\u003e2020\u003c/span\u003e; Schwamborn et al., \u003cspan class=\"CitationRef\"\u003e2010\u003c/span\u003e). Learners can benefit from scaffolded prompts where they fill in key components of an explanation rather than generating the entire explanation themselves (Bai et al., \u003cspan class=\"CitationRef\"\u003e2022\u003c/span\u003e). Self-explaining is more effective when learners receive focused prompts targeting specific relationships or principles rather than open or generic prompts (Berthold \u0026amp; Renkl, \u003cspan class=\"CitationRef\"\u003e2009\u003c/span\u003e). Scaffolded practice is most effective when followed by appropriate feedback, which helps learners evaluate and correct their performance (Johnson \u0026amp; Marrafano, 2022). In the current study we build on this literature by using GenAI to prompt students to engage in the generative activities of teaching. More specifically, students are given prompts that ask them to explain key components or explanations. The ChatTutor chatbot engages students in personalized written conversations based on the course slides and texts, and responds to address the specific answers provided by students and correcting any misconceptions or gaps in knowledge. Students are gradually provided with more information, scaffolding their understanding until they can offer a suitable explanation. This approach simulates a student-tutor interaction, where the tutor encourages generative learning by prompting students to explain specific content, intervening only when misconceptions are evident or when students request help.\u003c/p\u003e\n\u003cp\u003eFinally, independent practice involves repeated opportunities to use GLAs in specific learning contexts. This practice is crucial for learners to use GLAs spontaneously (Manalo et al., \u003cspan class=\"CitationRef\"\u003e2017\u003c/span\u003e) but is not investigated in this study.\u003c/p\u003e\n\u003cp\u003eIt is important to note that we are not proposing that GenAI is necessary or in its current technological iteration optimal for progressing from explicit instruction through scaffolded practice and ultimately to independent practice, however, we do propose that GenAI can be used to scale this process when resources to formatively assess students and provide personalized support is not possible. In the following section we describe how a theoretically informed GenAI chatbot can be used to mitigate some of the documented barriers to GLS.\u003c/p\u003e\n\u003ch3\u003eHow can a Theoretically Informed GenAI Chatbot Mitigate Known Barriers of GLS?\u003c/h3\u003e\n\u003cp\u003eFiorella (\u003cspan class=\"CitationRef\"\u003e2023\u003c/span\u003e) describes different cognitive, metacognitive and motivational barriers to sense making. Below we describe how a GenAI chatbot could attempt to mitigate these barriers thereby increasing sense-making and ultimately learning and motivational outcomes.\u003c/p\u003e\n\u003cp\u003e\u003cspan type=\"Underline\" class=\"Underline\" name=\"Emphasis\"\u003eCognitive Barriers\u003c/span\u003e: One cognitive barrier to engaging in effective GLS is insufficient background knowledge or available memory capacity to engage in GLAs effectively (Castro-Alonso et al., \u003cspan class=\"CitationRef\"\u003e2021\u003c/span\u003e). Learners need sufficient background knowledge to generate the necessary inferences for sense-making (Simonsmeier et al., \u003cspan class=\"CitationRef\"\u003e2022\u003c/span\u003e; Willingham, \u003cspan class=\"CitationRef\"\u003e2008\u003c/span\u003e). Without it, learning materials won\u0026apos;t be effective, regardless of the GLA used. With GenAI, students can ask a personalized chatbot particular questions thereby gaining specific knowledge. In ChatTutor this is implemented by linking directly to the texts and slides from the course., or by asking the LLM to provide simpler alternative explanations such as summaries or explanations. This is relevant because lower-knowledge students often benefit more from GLAs than higher-knowledge students (Fiorella \u0026amp; Mayer, \u003cspan class=\"CitationRef\"\u003e2015\u003c/span\u003e). This is because higher-knowledge learners are more likely to engage in sense-making spontaneously (Lombrozo, \u003cspan class=\"CitationRef\"\u003e2006\u003c/span\u003e).\u003c/p\u003e\n\u003cp\u003e\u003cspan type=\"Underline\" class=\"Underline\" name=\"Emphasis\"\u003eMetacognitive Barriers\u003c/span\u003e: Metacognitive barriers include a lack of strategic knowledge about what GLAs are, how to choose the right ones, and how to use them effectively. Many learners do not use GLAs spontaneously when studying texts (Fiorella \u0026amp; Mayer, \u003cspan class=\"CitationRef\"\u003e2017\u003c/span\u003e), visualizations (Renkl \u0026amp; Scheiter, \u003cspan class=\"CitationRef\"\u003e2017\u003c/span\u003e), worked examples (Chi et al., \u003cspan class=\"CitationRef\"\u003e1989\u003c/span\u003e; Renkl, \u003cspan class=\"CitationRef\"\u003e1997\u003c/span\u003e), solving problems (Rellensmann et al., \u003cspan class=\"CitationRef\"\u003e2022\u003c/span\u003e), discussing material with peers (Roscoe \u0026amp; Chi, \u003cspan class=\"CitationRef\"\u003e2007\u003c/span\u003e), or writing essays (Bereiter \u0026amp; Scardamalia, \u003cspan class=\"CitationRef\"\u003e2013\u003c/span\u003e). GenAI could provide explicit instructional support to assist learners to engage in GLAs. Specific prompting and scaffolding have been shown to be more effective compared to generic prompts in explaining, visualizing, or enacting concepts (Berthold \u0026amp; Renkl, \u003cspan class=\"CitationRef\"\u003e2009\u003c/span\u003e; Schmeck et al., \u003cspan class=\"CitationRef\"\u003e2014\u003c/span\u003e). Moreover, the quality of what learners generate during learning typically predicts their performance on later comprehension and transfer tests (Chi et al., \u003cspan class=\"CitationRef\"\u003e1989\u003c/span\u003e; Fiorella \u0026amp; Kuhlmann, \u003cspan class=\"CitationRef\"\u003e2020\u003c/span\u003e; Goldin-Meadow et al., \u003cspan class=\"CitationRef\"\u003e2009\u003c/span\u003e; Renkl, \u003cspan class=\"CitationRef\"\u003e1997\u003c/span\u003e; Schwamborn et al., \u003cspan class=\"CitationRef\"\u003e2010\u003c/span\u003e).\u003c/p\u003e\n\u003cp\u003e\u003cspan type=\"Underline\" class=\"Underline\" name=\"Emphasis\"\u003eMotivational barriers\u003c/span\u003e: Motivational barriers include beliefs about one\u0026rsquo;s ability to use GLAs successfully, the perceived value of GLAs for achieving one\u0026rsquo;s goals, or the perceived cost of GLAs (e.g., (Schukajlow et al., \u003cspan class=\"CitationRef\"\u003e2022\u003c/span\u003e). GenAI could increase self-efficacy by providing individualized mastery experiences (Bandura, \u003cspan class=\"CitationRef\"\u003e2001\u003c/span\u003e). A GenAI Chatbot can be seen as a more knowledgeable companion, providing positive feedback and enhancing learning by targeting the learner\u0026apos;s zone of proximal development (Vygotsky, \u003cspan class=\"CitationRef\"\u003e1978\u003c/span\u003e).\u003c/p\u003e\n\u003ch3\u003eCan a Theoretically Informed GenAI Chatbot influence other outcomes?\u003c/h3\u003e\n\u003cp\u003eIn addition to cognitive learning outcomes and self-efficacy, there are other important outcome variables that can ultimately impact the use of GenAI for learning. In this study we selected three: trust, enjoyment, and behavioral intentions.\u003c/p\u003e\n\u003cp\u003eThe adoption of GenAI in educational contexts brings forth several ethical challenges, such as transparency, privacy, equality, and beneficence (Khosravi et al., \u003cspan class=\"CitationRef\"\u003e2022\u003c/span\u003e). While all of these factors are relevant, building a GenAI chatbot that takes a learner-centered approach to interaction and builds on the curriculum in a course may specifically be relevant for transparency. A recent systematic review by Yan et al. (\u003cspan class=\"CitationRef\"\u003e2023\u003c/span\u003e) found that most GenAI tools (92%) currently employed to support learning are comprehensible only to AI experts. Educators, students, and other key stakeholders often lack the necessary insight into how these tools function. The authors highlight that the transparency gap largely stems from the limited integration of human-in-the-loop approaches in prior research. For instance, educators and students have seldom been actively involved in the design and evaluation of GenAI-based educational technologies. This shortfall highlights the urgent need for learner-centered AI, emphasizing the importance of engaging all stakeholders in the development process to ensure GenAI tools are both effective and meaningful in real-world educational settings. In this study we engage instructors, researchers, learners, and a GenAI development start-up in developing the ChatTutor system, and assess students\u0026rsquo; trust (Gulati et al., \u003cspan class=\"CitationRef\"\u003e2019\u003c/span\u003e) in using the system in our experiment.\u003c/p\u003e\n\u003cp\u003eIt can also be useful to measure the enjoyment of using AI technology in education because enjoyment is a powerful motivational factor that directly impacts learners\u0026apos; engagement and long-term adoption of these tools (Huang \u0026amp; Zou, \u003cspan class=\"CitationRef\"\u003e2024\u003c/span\u003e). According to the Control-Value Theory of Achievement Emotions, enjoyment plays a critical role in fostering intrinsic motivation and engagement by enhancing learners\u0026apos; perceptions of control over their learning and the value they assign to educational tasks (Pekrun, \u003cspan class=\"CitationRef\"\u003e2006\u003c/span\u003e). This positive emotional experience can reinforce satisfaction with the learning process and increase the intention to continue using AI-enhanced educational platforms (Venkatesh \u0026amp; Bala, \u003cspan class=\"CitationRef\"\u003e2008\u003c/span\u003e), leading to sustained engagement and ultimately more widespread adoption (Dai et al., \u003cspan class=\"CitationRef\"\u003e2024\u003c/span\u003e). We therefore measure trust, enjoyment, and behavioral intentions in this study.\u003c/p\u003e\n\u003ch3\u003eCurrent study\u003c/h3\u003e\n\u003cp\u003eIn the current study we address some of the gaps in the literature by designing a GenAI application building on a learner-centered theory of learning and instruction. Our main research question (RQ 1) is: What are the effects of using a theoretically informed GenAI chatbot (ChatTutor) to stimulate generative sense making on learning outcomes? We pre-registered two main hypotheses regarding this research question:\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u0026bull; Hypothesis 1\u0026nbsp;\u003c/strong\u003e(H1): Learners in the theoretically informed GenAI condition (ChatTutor) will achieve higher scores on a conceptual knowledge test immediately after the lesson when compared to learners in the teaching-as-usual (H1a) and ChatGPT-4 (H1b) conditions.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u0026bull; Hypothesis 2\u003c/strong\u003e (H2): Learners in the theoretically informed GenAI condition (ChatTutor) will achieve higher scores on a conceptual knowledge test administered four weeks after the lesson when compared to learners in the teaching-as-usual (H2a) and ChatGPT-4 (H2b) conditions.\u003c/p\u003e\n\u003cp\u003eOur secondary research question (RQ 2) is: Does the use of a theoretically informed GenAI chatbot (ChatTutor) lead to higher (1) self-efficacy compared to the teaching-as-usual condition and the ChatGPT condition, and more (2) trust, (3) enjoyment, and (4) intentions to use the technology in the future compared to the ChatGPT condition?\u003c/p\u003e"},{"header":"Methods","content":"\u003cdiv id=\"Sec8\" class=\"Section2\"\u003e\n \u003ch2\u003eStudy design\u003c/h2\u003e\n \u003cp\u003eThe study adopted a between-subject design consisting of one control group and two distinct GenAI conditions: ChatGPT and ChatTutor. The design, hypotheses, data collection, and analysis plan for this study were pre-registered on AsPredicted in September 2024, prior to the commencement of data collection. The anonymized preregistration is available at: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://aspredicted.org/w5mk-4vx4.pdf\u003c/span\u003e\u003c/span\u003e. The study was conducted at a large European University. Based on the self-assessment, the department\u0026rsquo;s local ethical board accepted the research project.\u003c/p\u003e\n\u003c/div\u003e\n\u003ch3\u003eProcedure\u003c/h3\u003e\n\u003cp\u003eThe project was initiated by psychology university instructors who were interested in investigating if GenAI could improve conceptual knowledge retention, self-efficacy, trust, enjoyment, and behavioral intentions in a cognitive psychology university course on the topic of Sternberg\u0026rsquo;s serial memory scanning experiment (Sternberg, \u003cspan class=\"CitationRef\"\u003e1966\u003c/span\u003e, \u003cspan class=\"CitationRef\"\u003e1969\u003c/span\u003e) which is a challenging topic for most students. The cognitive psychology course consists of a weekly lecture with over 240 psychology students, followed by both a seminar and an exercise class. At the beginning of the semester, students were randomly assigned to one of seven cognitive psychology classes with approximately 35 students in each. In accordance with local study board regulations, attendance in the lecture or the classes was not mandatory.\u003c/p\u003e\n\u003cp\u003eThe experiment took place in the first week of October, 2024, when the topic in the course was short-term memory and working memory. The follow-up took place as the first part of the lesson four weeks later. The Sternberg task was first introduced at the end of the seminar classes. In the following exercise class, students received a more in-depth overview and performed the task themselves. The exercise class consisted of three 45 minute blocks with 15 minute breaks between. The first block started with a wrap-up of the previous week, followed by a more in-depth introduction into the short-term memory task. In the second teaching block students took part in the Sternberg experiment themselves. After running the experiment, the students analyzed their logfile created at the end of the experiment, this task was otherwise unrelated to the current study. The third block started with a 20-minute power-point presentation by the instructor which was standardized across the different classes. This was followed by a 15 minute follow-up activity which was the focus of the experiment. In this between-subjects pre-registered experiment, two exercise classes were assigned to using the theoretically informed GenAI system (ChatTutor), two classes were assigned to the standard ChatGPT-4 (standard GenAI) condition which had the same interface as the ChatTutor condition, and the remaining three classes were assigned to the teaching-as-usual condition. Therefore, the difference between the groups was isolated to the generative learning activity which took approximately 15 minutes (see Fig. \u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003e). The final 10 minutes of the class consisted of a post-test including measures of conceptual knowledge, self-efficacy, and trust in AI. Four weeks after the intervention students used 10 minutes at the beginning of the exercise class to take a follow-up test that included another conceptual knowledge test as well as measures of enjoyment and behavioral intentions. Students were informed by their instructors that the assessment was an effort to investigate how to improve the teaching methods used in the course, but the students were not informed of the differences between instructor classes. At the end of the semester, students were informed of the results of the experiment and students in all treatment conditions were given access to the ChatTutor system.\u003c/p\u003e\n\u003ch3\u003eParticipants\u003c/h3\u003e\n\u003cp\u003eA total of 175 university students who were enrolled in the cognitive psychology course and attended their exercise class, agreed to participate in the study. Of these, 75 participants were in the teaching-as-usual condition, 49 were in the ChatGPT condition, and 51 were in the ChatTutor condition. The topic of the lesson was centered on the Sternberg Experiment. The experiment was integrated into classes\u0026rsquo; schedules and conducted during regular teaching hours. Participants provided their informed consent prior to the commencement of the experiment. As outlined in the preregistration, data from participants who chose not to engage in the experiment (\u0026lt;\u0026thinsp;5) were excluded. To our knowledge there were no additional students who opted not to use the AI program, and we did not exclude any students due to technical issues while using the GenAI program. A total of 124 students who completed the initial test, also completed the delayed follow-up test four weeks after the intervention. Among these, 54 participants belonged to the teaching-as-usual condition, 32 to the ChatGPT condition, and 38 to the ChatTutor condition.\u003c/p\u003e\n\u003cdiv id=\"Sec11\" class=\"Section2\"\u003e\n \u003ch2\u003eConditions\u003c/h2\u003e\n \u003cp\u003eThe experiment was isolated to the follow up activity which lasted approximately 15 minutes. All students received the similar instructions regarding the follow-up activity. Those in the teaching-as-usual-condition had 15 minutes to reflect over what they had learned and could discuss this with their classmates and had the opportunity to ask their instructor any questions which they were in doubt about regarding the topics that they had learned that day. The students in the ChatGPT condition had 15 minutes to reflect over what they had learned and had the opportunity to ask the GenAI any questions which they were in doubt of regarding the topics that they had learned that day. The students in the ChatTutor condition also had 15 minutes to engage in the ChatTutor system as described above.\u003c/p\u003e\u003cspan\u003e\n \u003cp\u003e\u003cstrong\u003e1. Teaching-as-usual\u003c/strong\u003e: Students in this condition were able to engage with the material and discuss with their fellow-students and could ask the instructor any relevant questions related to the material. Moreover, they were free to use any existing learning materials and internet sources.\u003c/p\u003e\n \u003c/span\u003e \u003cspan\u003e\n \u003cp\u003e\u003cstrong\u003e2. Generative Sense Making (ChatTutor) condition\u003c/strong\u003e: Students in this condition got access to ChatTutor website, which was designed to encourage generative processing as their learning tool. Students were encouraged to engage with the tool which actively prompted students to partake in the generative learning activity of teaching with scaffolded feedback. The ChatTutor system allowed learners to escalate a question to a human instructor if they encountered confusing or inaccurate information.\u003c/p\u003e\n \u003c/span\u003e \u003cspan\u003e\n \u003cp\u003e\u003cstrong\u003e3. Standard GenAI (ChatGPT) condition\u003c/strong\u003e: Students in this condition got access to a website interface that was identical to ChatTutor. However, the system linked directly to ChatGPT-4. Students were encouraged to engage with the tool with the same general instructions as in ChatTutor condition but the system responded based on the current technological capabilities of ChatGPT-4. However, the system also allowed learners to escalate a question to a human instructor if they encountered confusing or inaccurate information.\u003c/p\u003e\n \u003c/span\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec12\" class=\"Section2\"\u003e\n \u003ch2\u003eMeasures\u003c/h2\u003e\n \u003cp\u003eThe complete list of items is included in the supplementary materials.\u003c/p\u003e\n \u003cp\u003e\u003cspan type=\"Underline\" class=\"Underline\" name=\"Emphasis\"\u003eConceptual Knowledge (post-test and follow-up)\u003c/span\u003e: A test assessing participants\u0026apos; conceptual knowledge, with questions related to the memory scanning experiment and Sternberg\u0026rsquo;s original hypotheses (Sternberg, \u003cspan class=\"CitationRef\"\u003e1969\u003c/span\u003e) was administered immediately after the intervention/at the end of the lesson and again after four weeks. The test consisted of 10 multiple choice items, however, one was removed from the analyses and not included in the follow-up test because it accidentally contained more than one correct answer, resulting in a total of nine items. The test was designed to measure a broad range of topics covered with a focus on comprehension and not factual knowledge. That is, we prioritized content validity by having a broad range of questions rather than focusing on a narrow range of content which could have potentially increased internal reliability. The reliability coefficient for the nine-item conceptual knowledge test was \u0026alpha;\u0026thinsp;=\u0026thinsp;.55 in the post-test and \u0026alpha;\u0026thinsp;=\u0026thinsp;.68 in the follow-up. The test-retest reliability was r\u0026thinsp;=\u0026thinsp;.41.\u003c/p\u003e\n \u003cp\u003e\u003cspan type=\"Underline\" class=\"Underline\" name=\"Emphasis\"\u003eNew Conceptual Knowledge items (follow-up)\u003c/span\u003e: In addition to the nine items, four new conceptual knowledge items were administered in the follow-up to ensure that we could measure conceptual knowledge while accounting for testing effect (that students would remember the answers to the items because they had seen them before). The reliability coefficient for the new conceptual knowledge items was \u0026alpha;\u0026thinsp;=\u0026thinsp;.40.\u003c/p\u003e\n \u003cp\u003e\u003cspan type=\"Underline\" class=\"Underline\" name=\"Emphasis\"\u003eSelf-efficacy (post-test)\u003c/span\u003e: We measured self-efficacy using three items adapted from (Pintrich et al., \u003cspan class=\"CitationRef\"\u003e1991\u003c/span\u003e): 1. \u0026ldquo;I\u0026rsquo;m confident I can understand the concepts of serial and parallel memory search\u0026rdquo;; 2. \u0026ldquo;I\u0026rsquo;m confident I can understand the concepts of serial exhaustive search and serial self-terminating search\u0026rdquo;; 3. \u0026ldquo;I\u0026rsquo;m confident I could explain the Sternberg memory search graphs to a friend\u0026rdquo;. The reliability coefficient of the self-efficacy measure was \u0026alpha;\u0026thinsp;=\u0026thinsp;.82.\u003c/p\u003e\n \u003cp\u003e\u003cspan type=\"Underline\" class=\"Underline\" name=\"Emphasis\"\u003eTrust (post-test)\u003c/span\u003e: We measured trust using four items adapted from Gulati et al. (\u003cspan class=\"CitationRef\"\u003e2019\u003c/span\u003e): 1. \u0026ldquo;I feel I must be cautious when using ChatTutor\u0026rdquo;; 2. \u0026ldquo;I believe that ChatTutor will act in my best interest\u0026rdquo;; 3: \u0026ldquo;I think that ChatTutor is competent and effective in providing teaching assistance\u0026rdquo;; 4. \u0026ldquo;I can trust the information presented to me by ChatTutor\u0026rdquo;. The reliability coefficient of the trust measure was \u0026alpha;\u0026thinsp;=\u0026thinsp;.64.\u003c/p\u003e\n \u003cp\u003e\u003cspan type=\"Underline\" class=\"Underline\" name=\"Emphasis\"\u003eEnjoyment (follow-up)\u003c/span\u003e: We measured enjoyment using three items adapted from Huang \u0026amp; Zou (\u003cspan class=\"CitationRef\"\u003e2024\u003c/span\u003e), originally from (Davis et al., \u003cspan class=\"CitationRef\"\u003e1992\u003c/span\u003e): 1) \u0026quot;I find it satisfying to use ChatTutor,\u0026quot; 2) \u0026quot;I find it rewarding to use ChatTutor,\u0026quot; and 3) \u0026quot;I find it pleasant to use ChatTutor.\u0026quot; The reliability coefficient for the enjoyment measure was \u0026alpha;\u0026thinsp;=\u0026thinsp;.80.\u003c/p\u003e\n \u003cp\u003e\u003cspan type=\"Underline\" class=\"Underline\" name=\"Emphasis\"\u003eBehavioral intentions (follow-up)\u003c/span\u003e: We measured behavioral intentions with three items adapted from Lee and Choi (\u003cspan class=\"CitationRef\"\u003e2017\u003c/span\u003e), originally from Davis et al. (\u003cspan class=\"CitationRef\"\u003e1992\u003c/span\u003e). 1. \u0026ldquo;I intend to use ChatTutor again\u0026rdquo;; 2. \u0026ldquo;In the future, I will use ChatTutor to support my learning processes\u0026rdquo;; 3. \u0026ldquo;I would recommend ChatTutor to others\u0026rdquo;. The reliability coefficient for behavioral intentions was \u0026alpha;\u0026thinsp;=\u0026thinsp;.55.\u003c/p\u003e\n \u003cp\u003eAll items in the self-efficacy, trust, enjoyment, and behavioral intentions were measured on a 5-point scale (Strongly disagree:1. Disagree: 2. Neither agree nor disagree: 3. Agree: 4. Strongly agree: 5).\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec13\" class=\"Section2\"\u003e\n \u003ch2\u003eData analysis and deviations from pre-registration\u003c/h2\u003e\n \u003cp\u003eThe analyses for the study were performed using SPSS Statistics (IBM Corp., 2023). We examined our hypotheses using independent samples t-tests rather than ANOVA as the hypotheses were framed as pairwise comparisons and because trust, enjoyment and behavioral intentions were only assessed in the two GenAI groups. A sensitivity power analysis using G*power (Faul et al., \u003cspan class=\"CitationRef\"\u003e2007\u003c/span\u003e) indicated that with an \u0026alpha;\u0026thinsp;=\u0026thinsp;.05, and power of .80 we could expect to find an effect size d\u0026thinsp;=\u0026thinsp;.45 for the difference between the ChatTutor and teaching-as-usual conditions and an effect size of d\u0026thinsp;=\u0026thinsp;.50 between the ChatTutor and ChatGPT conditions. This value increased to d\u0026thinsp;=\u0026thinsp;.53, and d\u0026thinsp;=\u0026thinsp;.60 respectively in the follow-up due to drop-out. Values below these levels should therefore be interpreted with caution. Please note that we only pre-registered our measures of conceptual knowledge, self-efficacy, and trust, but did not pre-register the measures of enjoyment or behavioral intentions, which were initially left out of the post-test due to time constraints but we were later able to include them in the follow-up.\u003c/p\u003e\n\u003c/div\u003e"},{"header":"Results","content":"\u003cp\u003eThe main results of the study can be seen in Table 1. The results are organized based on the research questions below.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eTable 1: Means, standard deviations, sample sizes and significance levels for the outcome variables used in the study.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cu\u003eResults Related to the Main Research Question (RQ 1):\u003c/u\u003e The main research question in this study investigated the effects of using a theoretically informed GenAI chatbot (ChatTutor) to stimulate generative sense making on learning outcomes immediately after the intervention and in a delayed follow-up test four weeks after the intervention. The top row of Table 1 illustrates that the results partially support Hypothesis 1. More specifically, the theoretically informed GenAI ChatTutor condition resulted in significantly higher conceptual knowledge compared to the ChatGPT condition in the immediate post-test. However, while the ChatTutor condition mean score was higher than the teaching-as-usual condition, this difference was not statistically significant in the immediate post-test.\u003c/p\u003e\n\u003cp\u003eThe next row of Table 1 illustrates that the results also partially support Hypothesis 2.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThe theoretically informed GenAI ChatTutor condition resulted in significantly higher conceptual knowledge compared to the teaching-as-usual condition in the follow-up test four weeks after the intervention. However, while the ChatTutor condition mean score was higher than the standard ChatGPT condition, this difference was not statistically significant in the follow-up test. Because it was not possible to limit the students from using the ChatTutor interface between the lesson and the follow-up test, we asked students to report on a five-point scale how often they had used ChatTutor system since the lesson (1. Not at all; 2. Seldom; 3. Sometimes; 4. Often; 5. Very often). All of the students in the ChatGPT condition reported not using it at all, and all but six students in the ChatTutor condition reported not using it at all (four reported using it seldomly, and two reported using it sometimes). We re-ran all the analyses for the follow-up test, discarding these students as a robustness check, and found that this did not change any of the findings.\u003c/p\u003e\n\u003cp\u003e\u003cu\u003eResults Related to the Secondary Research Question (RQ 2):\u003c/u\u003e The second research question in this study was if a theoretically informed GenAI chatbot (ChatTutor) leads to higher (1) self-efficacy compared to the teaching-as-usual condition and the ChatGPT condition, and more (2) trust, (3) enjoyment, and (4) intentions to use the technology in the future compared to the ChatGPT condition. The results in Table 1 illustrate that the ChatTutor group (M = 3.64) scored significantly higher than the ChatGPT group (M = 3.29) on trust \u003cem\u003et\u003c/em\u003e = 2.955; \u003cem\u003ep\u003c/em\u003e = .002. The ChatTutor group (M = 3.65) also scored significantly higher than the ChatGPT group (M = 2.99) on enjoyment \u003cem\u003et\u003c/em\u003e = 3.304; \u003cem\u003ep\u003c/em\u003e \u0026lt; .001. Furthermore, the ChatTutor group (M = 3.28) scored significantly higher than the ChatGPT group (M = 2.73) on behavioral intentions \u003cem\u003et\u003c/em\u003e = 2.188; \u003cem\u003ep\u003c/em\u003e = .016. However, the ChatTutor group (M = 4.09) scored only marginally higher than the ChatGPT group (M = 3.82) on self-efficacy \u003cem\u003et\u003c/em\u003e = 1.661; \u003cem\u003ep\u003c/em\u003e = .050. Finally, there were no significant differences on self-efficacy between the ChatTutor and the teaching-as-usual conditions \u003cem\u003et\u003c/em\u003e = .167; \u003cem\u003ep\u003c/em\u003e = .434.\u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eThe first goal of this manuscript was to propose how GenAI could be used to support generative sense-making, providing a theoretical account of a learner-centered approach to using GenAI in education. Adopting an augmentation perspective, we use generative sense making research and theory to describe how GenAI can be used enhance lasting learning outcomes. More specifically, we propose how GenAI can be used to prime generative processing through enhancing personalized feedback when using teaching as a generative learning strategy. The second goal of this manuscript was to empirically investigate whether a theoretically informed GenAI chatbot (in this case, ChatTutor) could facilitate generative learning in education. Specifically, we aimed to test its impact on conceptual knowledge, self-efficacy, and trust immediately after a lesson, as well as its influence on conceptual knowledge, enjoyment, and behavioral intentions in a follow-up test four weeks later.\u003c/p\u003e \u003cdiv id=\"Sec16\" class=\"Section2\"\u003e \u003ch2\u003eEmpirical Contributions\u003c/h2\u003e \u003cp\u003eThe results of our study indicated that using ChatTutor resulted in significantly higher trust, enjoyment, and intentions to use the technology, and marginally higher self-efficacy compared to using ChatGPT. Here it is important to note that the students were unaware of the different conditions, as they were blind to the study's design and all used the ChatTutor interface. The difference was that the ChatTutor group obtained a theoretically informed GenAI learning intervention that builds on generative learning research and theory. Regarding conceptual knowledge the results indicate that the ChatTutor condition outperformed the ChatGPT condition in the immediate post-test but not in the delayed follow-up, and conversely the ChatTutor condition outperformed the teaching-as-usual condition in the follow-up but not the immediate post-test. The results related to conceptual knowledge are therefore mixed.\u003c/p\u003e \u003cp\u003eDelving into the mixed results related to conceptual knowledge in more depth, it is important to note that the groups only had 15 minutes on the follow-up generative activity task. This should be seen in the context of the entire course where learners were exposed to the topic of the Sternberg experiment through a seminar, lab activities and a specific lecture about the topic prior to the intervention. Therefore, it is likely that many students gained fundamental knowledge prior to engaging in the follow-up activity. While it would have been optimal to allow students more time to use ChatTutor we wanted to keep the amount of time across conditions constant as this has been highlighted as an important methodological consideration in previous studies (Brod, \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). Furthermore, it is a strength that the study took place in an authentic learning setting where the value of GenAI for learning should ultimately be tested.\u003c/p\u003e \u003cp\u003eWhen delving into the literature on GLA there are some other factors to be aware of regarding our mixed results. Previous research on teaching as a GLA suggests that oral explanations may be more effective than written ones, as they can evoke a sense of social presence, leading to higher levels of arousal (Hoogerheide et al., \u003cspan citationid=\"CR49\" class=\"CitationRef\"\u003e2019\u003c/span\u003e). A recent meta-analysis found a large positive effect of AI chatbots on learning outcomes (Wu \u0026amp; Yu, \u003cspan citationid=\"CR111\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). The authors recommend that future designers and educators further enhance these outcomes by incorporating human-like avatars, gamification elements, and emotional intelligence into AI chatbots. In this study we did not attempt to increase feelings of social presence through any of these methods which provide avenues for future research.\u003c/p\u003e \u003cp\u003eFurthermore, there is evidence that the GLA of teaching is most effective when the learning material is not available (Koh et al., \u003cspan citationid=\"CR61\" class=\"CitationRef\"\u003e2018\u003c/span\u003e). This can be the case because it requires learners to retrieve information from long-term memory rather than relying on the lesson, aiding in knowledge consolidation and future accessibility (e.g., Roediger \u0026amp; Karpicke, \u003cspan citationid=\"CR88\" class=\"CitationRef\"\u003e2006\u003c/span\u003e). In the current study the material was visually available for the students through the ChatTutor system. Therefore, it would have been possible to interact with ChatTutor by finding relevant passages in the text and paraphrasing them in the chat-box. Previous research has found that only students who generate an explanation exhibit improved understanding (Fiorella \u0026amp; Mayer, \u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e2013\u003c/span\u003e; Hoogerheide et al., \u003cspan citationid=\"CR48\" class=\"CitationRef\"\u003e2016\u003c/span\u003e). Therefore, future research should investigate the availability of learning content when using GenAI for GLS.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec17\" class=\"Section2\"\u003e \u003ch2\u003eTheoretical Implications\u003c/h2\u003e \u003cp\u003eThe leading theoretical paradigms guiding research on learning have evolved over the past 100 years, reflecting paradigm shifts from response strengthening to information acquisition, knowledge construction, and knowledge co-construction (Mayer, \u003cspan citationid=\"CR71\" class=\"CitationRef\"\u003e2012\u003c/span\u003e). Only time will tell if we are on the brink of a paradigm shift to a human-computer co-construction of knowledge. While it is unquestionable that existing educational psychology theory and research should guide and iteratively improve the development of GenAI-based learning interventions, as argued in this manuscript, a fundamental question remains: if and how educational psychology research must adapt to the new possibilities GenAI brings to education. The fundamental advantage of using GenAI lies in its ability to scale individualized learning experiences through realistic, dynamic conversations tailored to learners' needs. This approach shifts the focus from static, one-size-fits-all interventions to methods for delivering just-in-time support, aligning with advancements in learning analytics and self-regulated learning research (Azevedo \u0026amp; Gašević, \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e2019\u003c/span\u003e; Molenaar \u0026amp; J\u0026auml;rvel\u0026auml;, \u003cspan citationid=\"CR76\" class=\"CitationRef\"\u003e2014\u003c/span\u003e). In the context of generative sense-making, this could involve optimizing real-time feedback to support the creation and refinement of generative learning products, thereby enhancing engagement and understanding.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec18\" class=\"Section2\"\u003e \u003ch2\u003ePractical Implications\u003c/h2\u003e \u003cp\u003eOne of the main challenges associated with the recent rapid advancements in GenAI is its disruptive impact. As formulated in Holmes et al. (\u003cspan citationid=\"CR50\" class=\"CitationRef\"\u003e2019\u003c/span\u003e), p 3: \u0026ldquo;If you can search, or have an intelligent agent find, anything, why learn anything? What is truly worth learning?\u0026rdquo; Hence, it is useful to discuss how GenAI-enhanced learning fits (or does not fit) with traditional views of learning outcomes and processes. The goal of most education is for learners to be able to apply what they have learned in a relevant context. This is referred to as meaningful learning or deep knowledge acquisition which is indicated by performance on transfer tests, which involves being able to use the learned material in new situations (Mayer, \u003cspan citationid=\"CR73\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). Kasneci et. al. (\u003cspan citationid=\"CR57\" class=\"CitationRef\"\u003e2023\u003c/span\u003e) highlight some of the opportunities and challenges of using GenAI in education. The opportunities include personalized learning experiences, and challenges include the need for critical thinking and strategies for fact checking because GenAI agents have a tendency to hallucinate and provide inaccurate information. In a society where knowledge is abundant and easily accessible through internet searches or conversations with GenAI chatbots, the goal of education cannot be to simply impart knowledge but rather to engage students to ask the right questions and develop the critical thinking skills needed to evaluate the information they receive. This would suggest the need to shift from remembering and focusing on higher order cognitive process dimensions such as understanding, applying, analyzing, evaluating, and creating from Anderson and Krathwohl's (2001) taxonomy. One challenge is that GenAI may make students regulate less (Yan et al., \u003cspan citationid=\"CR113\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). For example, while GenAI enhances the efficiency of information processing and retrieval, it poses a risk of fluency bias, where learners may overestimate their understanding due to the ease of processing information. Additionally, relying on GenAI for creative and problem-solving tasks could undermine these essential skills, fostering dependency and potentially hindering innovation and original thought (Rafner et al., \u003cspan citationid=\"CR84\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Scneiderman, 2020). From a practical perspective this highlights the importance of thinking of GenAI as support rather than using it as a replacement as we have attempted to do in this manuscript.\u003c/p\u003e \u003cp\u003eAnother important practical implication is the need to educate stakeholders on the differences between using generic GenAI tools like ChatGPT, educational GenAI, and educational GenAI designed with educational psychology theory. Our study results clearly show that an educational GenAI chatbot that builds on theory, such as ChatTutor, fosters significantly greater trust, enjoyment, and intentions to use the technology in the future compared to a generic GenAI system like ChatGPT. Another important practical consideration is the challenge teachers face in defining their role in the use of GenAI for learning. Yan et al. (\u003cspan citationid=\"CR113\" class=\"CitationRef\"\u003e2024\u003c/span\u003e) suggest incorporating human-in-the-loop elements, such as fostering partnerships among researchers, practitioners, and policymakers. This approach was used in our study, where we collaborated with instructors, researchers, and a GenAI development startup to design the ChatTutor intervention. Furthermore, the intervention included human-in-the-loop elements, as the ChatTutor system allowed learners to escalate a question to a human instructor if they encountered confusing or inaccurate information. This feature is crucial given that GenAI chatbots are known to 'hallucinate,' that is, provide false information.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec19\" class=\"Section2\"\u003e \u003ch2\u003eLimitations and Future Research\u003c/h2\u003e \u003cp\u003eOur study had several limitations. The primary limitation was the short duration learners had to interact with the GenAI chatbot, compared to the extensive time spent on the topic through a seminar, exercise course lectures, and lab activities. This setup was not ideal for measuring the tool's impact on learning outcomes but was necessary given the context of an authentic cognitive psychology course. Such an approach is typical of a value-added study, where the objective is to assess the additional benefit of an educational intervention. Future research should explore the value of using GenAI to facilitate GLS when students have more time to engage with a theoretically informed GenAI chatbot. Additionally, future studies should examine different age groups, as previous research has shown varying levels of GLS effectiveness across age groups (Broad, 2021).\u003c/p\u003e \u003cp\u003eAnother limitation of our study concerns the reliability coefficients of some outcome measures, including conceptual knowledge and trust, which fell below acceptable levels. We prioritized content validity in our conceptual knowledge measure by including a broad range of items assessing different components of the learning material. However, future research would benefit from incorporating measures that are both valid and reliable. The goal of the learning intervention in this study was conceptual knowledge understanding. Future research could explore the value of theoretically informed GenAI for higher-order learning outcomes, such as applying, analyzing, evaluating, and creating, where relevant to learning interventions. Furthermore, this study focused on developing a GenAI intervention based on generative sense-making research and theory. Future research should explore how other educational psychology theories can inform the development of GenAI interventions, given the scarcity of such studies.\u003c/p\u003e \u003c/div\u003e"},{"header":"Conclusion","content":"\u003cp\u003eGenerative Artificial Intelligence (GenAI) has captured significant public and scholarly attention and could challenge education by transforming traditional methods of teaching, learning, assessment, and administration. However, a recurring challenge with novel technologies lies in the tendency to implement and evaluate them from a technocentric perspective, often neglecting insights from educational psychology and learning science. This manuscript had two primary goals: first, to extend generative learning theory by proposing how GenAI can be utilized to support generative sense-making in education; and second, to empirically examine its impact on the outcomes of conceptual knowledge, self-efficacy, trust, enjoyment, and behavioral intentions. To achieve these objectives, we developed a theoretically informed GenAI chatbot (ChatTutor) designed to scaffold learning by prompting students to engage in the generative learning strategy of teaching a GenAI chatbot. Through a pre-registered experiment conducted in an authentic university setting, we compared ChatTutor to ChatGPT and a teaching-as-usual condition. The findings demonstrated that ChatTutor significantly enhanced trust, enjoyment, and behavioral intentions compared to ChatGPT, but not self-efficacy. Regarding learning outcomes, ChatTutor led to higher immediate conceptual knowledge scores than ChatGPT but did not outperform teaching-as-usual. In the delayed follow-up test, ChatTutor led to significantly higher conceptual knowledge than teaching-as-usual, but the differences with ChatGPT were no longer significant. This study underscores the importance of adapting educational psychology theories to harness the unique capabilities of GenAI in supporting meaningful learning. The findings contribute to the growing call for theory-driven, learner-centered integration of GenAI into education, moving beyond the initial fascination with the technology.\u003c/p\u003e"},{"header":"Declarations","content":"\u003ch2\u003eAcknowledgements\u003c/h2\u003e \u003cp\u003eWe would like to thank Gustav B. Petersen, Benjamin B. Christensen, and Sidsel Drejer for helping develop the content of the lesson used in the experiment. Furthermore, we would like to thank Gustav B. Petersen, Marius Corson, Mikkel Marfelt and Kurt G. Nielsen for helping organize the experiment. Finally, we would like to thank Sidsel Drejer, Benjamin B. Christensen, Rubina F. Gogolu, Rebecca G. Lachmann, Alexander T. Ysb\u0026aelig;k-Nielsen, Monique S. Damberg and Daniel Jensen who were the instructors who make the experiment possible.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n \u003cli\u003eAnderson, L. W., \u0026amp; Krathwohl, D. R. (2001). A Taxonomy for learning, teaching, and assesing. \u003cem\u003eA taxonomy for learning, teaching and assessing: A revision of Bloom\u0026rsquo;s taxonomy\u003c/em\u003e. Longman Publishing, 1\u0026ndash;336.\u003c/li\u003e\n \u003cli\u003eAnthopic. (2024). \u003cem\u003eClaude\u0026rsquo;s character\u003c/em\u003e. Anthopic. https://doi.org/https://www.anthropic.com/research/claude-character\u003c/li\u003e\n \u003cli\u003eAzevedo, R., \u0026amp; Ga\u0026scaron;ević, D. (2019). Analyzing multimodal multichannel data about self-regulated learning with advanced learning technologies: Issues and challenges. \u003cem\u003eComputers in Human Behavior\u003c/em\u003e, \u003cem\u003e96\u003c/em\u003e, 207\u0026ndash;210. https://doi.org/10.1016/j.chb.2019.03.025\u003c/li\u003e\n \u003cli\u003eBai, Y., Jones, A., Ndousse, K., Askell, A., Chen, A., DasSarma, N., Drain, D., Fort, S., Ganguli, D., Henighan, T., Joseph, N., Kadavath, S., Kernion, J., Conerly, T., El-Showk, S., Elhage, N., Hatfield-Dodds, Z., Hernandez, D., Hume, T., \u0026hellip; Kaplan, J. (2022). \u003cem\u003eTraining a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback\u003c/em\u003e, 1\u0026ndash;74. https://doi.org/10.48550/arXiv.2204.05862\u003c/li\u003e\n \u003cli\u003eBandura, A. (1977). Self-efficacy: Toward a unifying theory of behavioral change. \u003cem\u003ePsychological Review\u003c/em\u003e, \u003cem\u003e84\u003c/em\u003e(2), 191\u0026ndash;214. https://doi.org/10.1037/0033-295X.84.2.191\u003c/li\u003e\n \u003cli\u003eBandura, A. (2001). Social cognitive theory: An agentic perspective. In \u003cem\u003eAnnual Review of Psychology, 52,\u0026nbsp;\u003c/em\u003e1\u0026ndash;26. https://doi.org/10.1146/annurev.psych.52.1.1\u003c/li\u003e\n \u003cli\u003eBenbasat, I., \u0026amp; Wang, W. (2005). Trust In and Adoption of Online Recommendation Agents. \u003cem\u003eJournal of the Association for Information Systems\u003c/em\u003e, \u003cem\u003e6\u003c/em\u003e(3), 72\u0026ndash;101. https://doi.org/10.17705/1jais.00065\u003c/li\u003e\n \u003cli\u003eBereiter, C., \u0026amp; Scardamalia, M. (2013). The psychology of written composition. In \u003cem\u003eThe Psychology of Written Composition\u003c/em\u003e, 1\u0026ndash;389. https://doi.org/10.4324/9780203812310\u003c/li\u003e\n \u003cli\u003eBerlyne, D. E. (2006). Conflict, arousal, and curiosity. In \u003cem\u003eConflict, arousal, and curiosity,\u003c/em\u003e 1\u0026ndash;366. \u003cem\u003e\u0026nbsp;\u003c/em\u003ehttps://doi.org/10.1037/11164-000\u003c/li\u003e\n \u003cli\u003eBerthold, K., \u0026amp; Renkl, A. (2009). Instructional Aids to Support a Conceptual Understanding of Multiple Representations. \u003cem\u003eJournal of Educational Psychology\u003c/em\u003e, \u003cem\u003e101\u003c/em\u003e(1), 70\u0026ndash;87. https://doi.org/10.1037/a0013247\u003c/li\u003e\n \u003cli\u003eBerthold, K., Eysink, T. H., \u0026amp; Renkl, A. (2009). Assisting selfexplanation prompts are more effective than open prompts when learning fromm multiple representations. \u003cem\u003eInstructional Science, 37,\u003c/em\u003e 345\u0026ndash;363. https://doi.org/10.1007/s11251-008-9051-z\u003c/li\u003e\n \u003cli\u003eBisra, K., Liu, Q., Nesbit, J. C., Salimi, F., \u0026amp; Winne, P. H. (2018). Inducing Self-Explanation: a Meta-Analysis. \u003cem\u003eEducational Psychology Review\u003c/em\u003e, \u003cem\u003e30\u003c/em\u003e(3), 703\u0026ndash;725. https://doi.org/10.1007/s10648-018-9434-x\u003c/li\u003e\n \u003cli\u003eBrennan, K. (2015). Beyond technocentrism: Supporting constructionism in the classroom. \u003cem\u003eConstructivist Foundations\u003c/em\u003e, \u003cem\u003e10\u003c/em\u003e(3), 289\u0026ndash;286. https://constructivist.info/10/3/289.brennan.pdf\u003c/li\u003e\n \u003cli\u003eBrod, G. (2021). Generative Learning: Which Strategies for What Age? In \u003cem\u003eEducational Psychology Review\u003c/em\u003e. \u003cem\u003e33\u003c/em\u003e(4), 1295\u0026ndash;1318. https://doi.org/10.1007/s10648-020-09571-9\u003c/li\u003e\n \u003cli\u003eBurkhart, C., Lachner, A., \u0026amp; N\u0026uuml;ckles, M. (2021). Using Spatial Contiguity and Signaling to Optimize Visual Feedback on Students\u0026rsquo; Written Explanations. \u003cem\u003eJournal of Educational Psychology\u003c/em\u003e, \u003cem\u003e113\u003c/em\u003e(5), 998\u0026ndash;1023. https://doi.org/10.1037/edu0000607\u003c/li\u003e\n \u003cli\u003eCastro-Alonso, J. C., de Koning, B. B., Fiorella, L., \u0026amp; Paas, F. (2021). Five Strategies for Optimizing Instructional Materials: Instructor- and Learner-Managed Cognitive Load. In \u003cem\u003eEducational Psychology Review\u003c/em\u003e, \u003cem\u003e33\u003c/em\u003e(4), 1379\u0026ndash;1407. https://doi.org/10.1007/s10648-021-09606-9\u003c/li\u003e\n \u003cli\u003eChandler, P. (2009). Dynamic visualisations and hypermedia: Beyond the \u0026ldquo;Wow\u0026rdquo; factor. \u003cem\u003eComputers in Human Behavior\u003c/em\u003e, \u003cem\u003e25\u003c/em\u003e(2), 389\u0026ndash;392. https://doi.org/10.1016/j.chb.2008.12.018\u003c/li\u003e\n \u003cli\u003eChi, M. T. H., Bassok, M., Lewis, M. W., Reimann, P., \u0026amp; Glaser, R. (1989). Self-explanations: How students study and use examples in learning to solve problems. \u003cem\u003eCognitive Science\u003c/em\u003e, \u003cem\u003e13\u003c/em\u003e(2), 145\u0026ndash;182. https://doi.org/10.1016/0364-0213(89)90002-5\u003c/li\u003e\n \u003cli\u003eChiu, T. K. F., Moorhouse, B. L., Chai, C. S., \u0026amp; Ismailov, M. (2023). Teacher support and student motivation to learn with Artificial Intelligence (AI) based chatbot. \u003cem\u003eInteractive Learning Environments, 32\u003c/em\u003e(7),\u003cem\u003e\u0026nbsp;\u003c/em\u003e1\u0026ndash;17. https://doi.org/10.1080/10494820.2023.2172044\u003c/li\u003e\n \u003cli\u003eChow, A. (2023). How ChatGPT Managed to Grow Faster Than TikTok or Instagram. \u003cem\u003eTIME\u003c/em\u003e. Retrieved https://time.com/6253615/chatgpt-fastest-growing/\u003c/li\u003e\n \u003cli\u003eCu\u0026eacute;llar, M. F., Larsen, B., Lee, Y. S., \u0026amp; Webb, M. (2024). Does Information About AI Regulation Change Manager Evaluation of Ethical Concerns and Intent to Adopt AI? \u003cem\u003eThe\u003c/em\u003e \u003cem\u003eJournal of Law, Economics, and Organization\u003c/em\u003e, \u003cem\u003e40\u003c/em\u003e(1), 34\u0026ndash;75. https://doi.org/10.1093/jleo/ewac004\u003c/li\u003e\n \u003cli\u003eDai, J., Zhang, X., \u0026amp; Wang, C. (2024). A meta-analysis of learners\u0026rsquo; continuance intention toward online education platforms. \u003cem\u003eEducation and Information Technologies, 29,\u003c/em\u003e 1\u0026ndash;36. https://doi.org/10.1007/s10639-024-12654-7\u003c/li\u003e\n \u003cli\u003eDargue, N., Sweller, N., \u0026amp; Jones, M. P. (2019). When our hands help us understand: A meta-analysis into the effects of gesture on comprehension. \u003cem\u003ePsychological Bulletin\u003c/em\u003e, \u003cem\u003e145\u003c/em\u003e(8), 765\u0026ndash;784. https://doi.org/10.1037/bul0000202\u003c/li\u003e\n \u003cli\u003eDarvishi, A., Khosravi, H., Sadiq, S., Ga\u0026scaron;ević, D., \u0026amp; Siemens, G. (2024). Impact of AI assistance on student agency. \u003cem\u003eComputers and Education\u003c/em\u003e, \u003cem\u003e210,\u0026nbsp;\u003c/em\u003e1\u0026ndash;18. https://doi.org/10.1016/j.compedu.2023.104967\u003c/li\u003e\n \u003cli\u003eDavis, F. D., Bagozzi, R. P., \u0026amp; Warshaw, P. R. (1992). Extrinsic and Intrinsic Motivation to Use Computers in the Workplace. \u003cem\u003eJournal of Applied Social Psychology\u003c/em\u003e, \u003cem\u003e22\u003c/em\u003e(14), 1111\u0026ndash;1132. https://doi.org/10.1111/j.1559-1816.1992.tb00945.x\u003c/li\u003e\n \u003cli\u003eDe Jong, T. (2006). Technological advances in inquiry learning. In \u003cem\u003eScience,\u003c/em\u003e \u003cem\u003e312\u003c/em\u003e(5773), 532\u0026ndash;533. https://doi.org/10.1126/science.1127750\u003c/li\u003e\n \u003cli\u003eDevlin, J., Chang, M. W., Lee, K., \u0026amp; Toutanova, K. (2019). BERT: Pre-training of deep bidirectional transformers for language understanding. In \u003cem\u003eproceedings of NAACL HLT 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies,\u0026nbsp;\u003c/em\u003e(pp. 1\u0026ndash;16)\u003cem\u003e.\u0026nbsp;\u003c/em\u003e\u003cem\u003ehttps://doi.org/10.48550/arXiv.1810.04805\u003c/em\u003e\u003c/li\u003e\n \u003cli\u003eDillenbourg, P. (1999). Collaborative Learning: Cognitive and Computational Approaches. Advances in Learning and Instruction Series. \u003cem\u003eElsevier Science\u003c/em\u003e, 1\u0026ndash;246. https://eric.ed.gov/?id=ED437928\u003c/li\u003e\n \u003cli\u003eEdwards, B. (2023). OpenAI\u0026rsquo;s GPT-4 exhibits \u0026ldquo;human-level performance\u0026rdquo; on professional benchmarks. \u003cem\u003eArs Technica.\u0026nbsp;\u003c/em\u003eRetrieved from: https://arstechnica.com/information-technology/2023/03/openai-announces-gpt-4-its-next-generation-ai-language-model/\u003c/li\u003e\n \u003cli\u003eEshuis, E. H., ter Vrugte, J., Anjewierden, A., \u0026amp; de Jong, T. (2022). Expert examples and prompted reflection in learning with self-generated concept maps. \u003cem\u003eJournal of Computer Assisted Learning\u003c/em\u003e, \u003cem\u003e38\u003c/em\u003e(2), 350\u0026ndash;365. https://doi.org/10.1111/jcal.12615\u003c/li\u003e\n \u003cli\u003eFaul, F., Erdfelder, E., Lang, A.-G., \u0026amp; Buchner, A. (2007). G*Power 3: A flexible statistical power analysis program for the social, behavioral, and biomedical sciences. \u003cem\u003eBehavior Research Methods\u003c/em\u003e, \u003cem\u003e39\u003c/em\u003e, 175\u0026ndash;191. https://doi.org/10.3758/BF03193146\u003c/li\u003e\n \u003cli\u003eFenn, J., \u0026amp; Blosch, M. (2020). \u003cem\u003eUnderstanding Gartner\u0026rsquo;s Hype Cycles.\u0026nbsp;\u003c/em\u003eGartner Research. https://www.gartner.com/en/documents/3887767\u003c/li\u003e\n \u003cli\u003eFiorella, L. (2023). Making Sense of Generative Learning. In \u003cem\u003eEducational Psychology Review, 35\u003c/em\u003e(2), 1\u0026ndash;42. https://doi.org/10.1007/s10648-023-09769-7\u003c/li\u003e\n \u003cli\u003eFiorella, L., \u0026amp; Kuhlmann, S. (2020). Creating Drawings Enhances Learning by Teaching. \u003cem\u003eJournal of Educational Psychology\u003c/em\u003e, \u003cem\u003e112\u003c/em\u003e(4), 811\u0026ndash;822. https://doi.org/10.1037/edu0000392\u003c/li\u003e\n \u003cli\u003eFiorella, L., \u0026amp; Mayer, R. E. (2013). The relative benefits of learning by teaching and teaching expectancy. \u003cem\u003eContemporary Educational Psychology, 38\u003c/em\u003e(4), 281\u0026ndash;288. https://doi.org/10.1016/j.cedpsych.2013.06.001\u003c/li\u003e\n \u003cli\u003eFiorella, L., \u0026amp; Mayer, R. E. (2015). Learning as a generative activity: Eight learning strategies that promote understanding. 1\u0026ndash;218. https://doi.org/10.1017/CBO9781107707085\u003c/li\u003e\n \u003cli\u003eFiorella, L., \u0026amp; Mayer, R. E. (2016). Eight Ways to Promote Generative Learning. In \u003cem\u003eEducational Psychology Review\u003c/em\u003e,\u003cem\u003e\u0026nbsp;28\u003c/em\u003e(4), 717\u0026ndash;741. https://doi.org/10.1007/s10648-015-9348-9\u003c/li\u003e\n \u003cli\u003eFiorella, L., \u0026amp; Mayer, R. E. (2017). Spontaneous spatial strategy use in learning from scientific text. \u003cem\u003eContemporary Educational Psychology\u003c/em\u003e, \u003cem\u003e49\u003c/em\u003e, 66\u0026ndash;79. https://doi.org/10.1016/j.cedpsych.2017.01.002\u003c/li\u003e\n \u003cli\u003eFiorella, L., \u0026amp; Mayer, R. E. (2022). The Generative Activity Principle in Multimedia Learning, 339\u0026ndash;350. In \u003cem\u003eThe Cambridge Handbook of Multimedia Learning\u003c/em\u003e. https://doi.org/10.1017/9781108894333.036\u003c/li\u003e\n \u003cli\u003eFiorella, L., Yoon, S. Y., Atit, K., Power, J. R., Panther, G., Sorby, S., Uttal, D. H., \u0026amp; Veurink, N. (2021). Validation of the Mathematics Motivation Questionnaire (MMQ) for secondary school students. \u003cem\u003eInternational Journal of STEM Education\u003c/em\u003e, \u003cem\u003e8\u003c/em\u003e(1), 1\u0026ndash;14. https://doi.org/10.1186/s40594-021-00307-x\u003c/li\u003e\n \u003cli\u003eFiorella, L., \u0026amp; Zhang, Q. (2018). Drawing Boundary Conditions for Learning by Drawing. In \u003cem\u003eEducational Psychology Review, 30\u003c/em\u003e(3), 1115\u0026ndash;1137. https://doi.org/10.1007/s10648-018-9444-8\u003c/li\u003e\n \u003cli\u003eFischer, G., Lundin, J., \u0026amp; Lindberg, J. O. (2020). Rethinking and reinventing learning, education and collaboration in the digital age\u0026mdash;from creating technologies to transforming cultures. \u003cem\u003eInternational Journal of Information and Learning Technology\u003c/em\u003e, \u003cem\u003e37\u003c/em\u003e(5), 241\u0026ndash;252. https://doi.org/10.1108/IJILT-04-2020-0051\u003c/li\u003e\n \u003cli\u003eGoldin-Meadow, S., Cook, S. W., \u0026amp; Mitchell, Z. A. (2009). Gesturing gives children new ideas about math. \u003cem\u003ePsychological Science\u003c/em\u003e, \u003cem\u003e20\u003c/em\u003e(3), 267\u0026ndash;272. https://doi.org/10.1111/j.1467-9280.2009.02297.x\u003c/li\u003e\n \u003cli\u003eGulati, S., Sousa, S., \u0026amp; Lamas, D. (2019). Design, development and evaluation of a human-computer trust scale. \u003cem\u003eBehaviour and Information Technology\u003c/em\u003e, \u003cem\u003e38\u003c/em\u003e(10), 1004\u0026ndash;1015. https://doi.org/10.1080/0144929X.2019.1656779\u003c/li\u003e\n \u003cli\u003eHelmore, E. (2023). We are a little bit scared\u0026rsquo;: OpenAI CEO warns of risks of artificial intelligence. \u003cem\u003eTheGuardian\u003c/em\u003e. Retrieved from: https://www.theguardian.com/technology/2023/mar/17/openai-sam-altman-artificial-intelligence-warning-gpt4\u003c/li\u003e\n \u003cli\u003eHennessy, S., Cukurova, M., Lewin, C., Mavrikis, M., \u0026amp; Major, L. (2024). BJET Editorial 2024: A call for research rigour. In \u003cem\u003eBritish Journal of Educational Technology\u003c/em\u003e, \u003cem\u003e55\u003c/em\u003e(1), 5\u0026ndash;9. https://doi.org/10.1111/bjet.13426\u003c/li\u003e\n \u003cli\u003eHmelo-Silver, C. E., Duncan, R. G., \u0026amp; Chinn, C. A. (2007). Scaffolding and achievement in problem-based and inquiry learning: A response to Kirschner, Sweller, and Clark (2006). In \u003cem\u003eEducational Psychologist\u003c/em\u003e, \u003cem\u003e42\u003c/em\u003e(2), 99\u0026ndash;107. https://doi.org/10.1080/00461520701263368\u003c/li\u003e\n \u003cli\u003eHoogerheide, V., Deijkers, L., Loyens, S. M., Heijltjes, A., \u0026amp; van Gog, T. (2016). Gaining from explaining: Learning improves from explaining to fictitious others on video, not from writing to them. \u003cem\u003eContemporary Educational Psychology, 44,\u0026nbsp;\u003c/em\u003e95\u0026ndash;106. https://doi.org/10.1016/j.cedpsych.2016.02.005\u003c/li\u003e\n \u003cli\u003eHoogerheide, V., Renkl, A., Fiorella, L., Paas, F., \u0026amp; Van Gog, T. (2019). Enhancing example-based learning: Teaching on video increases arousal and improves problem-solving performance. \u003cem\u003eJournal of Educational Psychology, 111\u003c/em\u003e(1), 45\u0026ndash;56. https://doi.org/10.1037/edu0000272\u003c/li\u003e\n \u003cli\u003eHolmes, W., Fadel, C., \u0026amp; Bialik, M. (2019). Artificial intelligence in education: Promises and implications for teaching and learning. \u003cem\u003eJournal of Computer Assisted Learning\u003c/em\u003e, \u003cem\u003e14\u003c/em\u003e(4). https://www.researchgate.net/publication/332180327_Artificial_Intelligence_in_Education_Promise_and_Implications_for_Teaching_and_Learning\u003c/li\u003e\n \u003cli\u003eHolmes, W., \u0026amp; Tuomi, I. (2022). State of the art and practice in AI in education. \u003cem\u003eEuropean Journal of Education\u003c/em\u003e, \u003cem\u003e57\u003c/em\u003e(4), 542\u0026ndash;570. https://doi.org/10.1111/ejed.12533\u003c/li\u003e\n \u003cli\u003eHuang, F., \u0026amp; Zou, B. (2024). English speaking with artificial intelligence (AI): The roles of enjoyment, willingness to communicate with AI, and innovativeness. \u003cem\u003eComputers in Human Behavior\u003c/em\u003e, \u003cem\u003e159,\u0026nbsp;\u003c/em\u003e2\u0026ndash;8. https://doi.org/10.1016/j.chb.2024.108355\u003c/li\u003e\n \u003cli\u003eIBM Corp. (2023). IBM SPSS Statistics for Windows. https://doi.org/https://www.ibm.com/products/spss-statistics\u003c/li\u003e\n \u003cli\u003eJ\u0026auml;rvel\u0026auml;, S., Nguyen, A., Vuorenmaa, E., Malmberg, J., \u0026amp; J\u0026auml;rvenoja, H. (2023). Predicting regulatory activities for socially shared regulation to optimize collaborative learning. \u003cem\u003eComputers in\u0026nbsp;\u003c/em\u003e\u003cem\u003eHuman Behavior\u003c/em\u003e, \u003cem\u003e144,\u0026nbsp;\u003c/em\u003e1\u0026ndash;10. https://doi.org/10.1016/j.chb.2023.107737\u003c/li\u003e\n \u003cli\u003eJohnson, C. I., \u0026amp; Marrafno, M. D. (2022). The feedback principle in multimedia learning. In R. E. Mayer \u0026amp; L. Fiorella (Eds.), \u003cem\u003eThe Cambridge handbook of multimedia learning\u003c/em\u003e (3rd ed., pp. 286\u0026ndash; 295). Cambridge University Press. https://doi.org/10.1017/CBO9781139547369.023\u003c/li\u003e\n \u003cli\u003eKasneci, E., Sessler, K., K\u0026uuml;chemann, S., Bannert, M., Dementieva, D., Fischer, F., Gasser, U., Groh, G., G\u0026uuml;nnemann, S., H\u0026uuml;llermeier, E., Krusche, S., Kutyniok, G., Michaeli, T., Nerdel, C., Pfeffer, J., Poquet, O., Sailer, M., Schmidt, A., Seidel, T., \u0026hellip; Kasneci, G. (2023). ChatGPT for good? On opportunities and challenges of large language models for education. In \u003cem\u003eLearning and Individual Differences\u003c/em\u003e, \u003cem\u003e103,\u0026nbsp;\u003c/em\u003e1\u0026ndash;9\u003cem\u003e.\u003c/em\u003e https://doi.org/10.1016/j.lindif.2023.102274\u003c/li\u003e\n \u003cli\u003eKhosravi, H., Shum, S. B., Chen, G., Conati, C., Tsai, Y. S., Kay, J., Knight, S., Martinez-Maldonado, R., Sadiq, S., \u0026amp; Ga\u0026scaron;ević, D. (2022). Explainable Artificial Intelligence in education. \u003cem\u003eComputers and Education: Artificial Intelligence\u003c/em\u003e, \u003cem\u003e3,\u0026nbsp;\u003c/em\u003e1\u0026ndash;22. https://doi.org/10.1016/j.caeai.2022.100074\u003c/li\u003e\n \u003cli\u003eKing, A., Staffieri, A., \u0026amp; Adelgais, A. (1998). Mutual peer tutoring: Effects of structuring tutorial interaction to scaffold peer learning. \u003cem\u003eJournal of Educational Psychology, 90\u003c/em\u003e, 134\u0026ndash;152. https://doi.org/10.1037/0022-0663.90.1.134\u003c/li\u003e\n \u003cli\u003eKobayashi, K. (2019). Learning by preparing-to-teach and teaching: A meta-analysis. \u003cem\u003eJapanese Psychological Research, 61\u003c/em\u003e(3), 192\u0026ndash;203. https://doi.org/10.1111/jpr.12221\u003c/li\u003e\n \u003cli\u003eKoh, A. W. L., Lee, S. C., \u0026amp; Lim, S. W. H. (2018). The learning benefits of teaching: A retrieval practice hypothesis. \u003cem\u003eApplied Cognitive Psychology, 32\u003c/em\u003e(3), 401\u0026ndash;410. https://doi.org/10.1002/acp.3410\u003c/li\u003e\n \u003cli\u003eLachner, A., Jacob, L., \u0026amp; Hoogerheide, V. (2021). Learning by writing explanations: Is explaining to a fictitious student more effective than self-explaining? \u003cem\u003eLearning and Instruction\u003c/em\u003e, \u003cem\u003e74,\u0026nbsp;\u003c/em\u003e1\u0026ndash;13. https://doi.org/10.1016/j.learninstruc.2020.101438\u003c/li\u003e\n \u003cli\u003eLachner, A., \u0026amp; Neuburg, C. (2019). Learning by writing explanations: computer-based feedback about the explanatory cohesion enhances students\u0026rsquo; transfer. \u003cem\u003eInstructional Science\u003c/em\u003e, \u003cem\u003e47\u003c/em\u003e(1), 19\u0026ndash;37. https://doi.org/10.1007/s11251-018-9470-4\u003c/li\u003e\n \u003cli\u003eLee, S. Y., \u0026amp; Choi, J. (2017). Enhancing user experience with conversational agent for movie recommendation: Effects of self-disclosure and reciprocity. \u003cem\u003eInternational Journal of Human Computer Studies\u003c/em\u003e, \u003cem\u003e103,\u0026nbsp;\u003c/em\u003e95\u0026ndash;105. https://doi.org/10.1016/j.ijhcs.2017.02.005\u003c/li\u003e\n \u003cli\u003eLombrozo, T. (2006). The structure and function of explanations. \u003cem\u003eTrends in Cognitive Sciences\u003c/em\u003e,\u003cem\u003e\u0026nbsp;10\u003c/em\u003e(10), 464-470. https://doi.org/10.1016/j.tics.2006.08.004\u003c/li\u003e\n \u003cli\u003eLorenz, P., Perset, K., \u0026amp; Berryhill, J. (2023). Initial policy considerations for generative artificial intelligence. \u003cem\u003eOECD Artificial intelligence papers\u003c/em\u003e, \u003cem\u003e1,\u0026nbsp;\u003c/em\u003e1\u0026ndash;40\u003cem\u003e.\u0026nbsp;\u003c/em\u003eRetrieved from: https://www.oecd.org/en/publications/initial-policy-considerations-for-generative-artificial-intelligence_fae2d1e6-en.html\u003c/li\u003e\n \u003cli\u003eMakransky, G., \u0026amp; Petersen, G. B. (2021). The Cognitive Affective Model of Immersive Learning (CAMIL): a Theoretical Research-Based Model of Learning in Immersive Virtual Reality. \u003cem\u003eEducational Psychology Review, 33\u003c/em\u003e(3), 937\u0026ndash;958. https://doi.org/10.1007/s10648-020-09586-2\u003c/li\u003e\n \u003cli\u003eManalo, E., Uesaka, Y., \u0026amp; Chinn, C. A. (2017). Promoting Spontaneous Use of Learning and Reasoning Strategies. In \u003cem\u003ePromoting Spontaneous Use of Learning and Reasoning Strategies,\u0026nbsp;\u003c/em\u003e1\u0026ndash;166. https://doi.org/10.4324/9781315564029\u003c/li\u003e\n \u003cli\u003eMayer, R. E. (1996). Learning strategies for making sense out of expository text: the soi model for guiding three cognitive processes in knowledge construction. \u003cem\u003eEducational Psychology Review\u003c/em\u003e, \u003cem\u003e8\u003c/em\u003e(4), 357\u0026ndash;371. https://doi.org/10.1007/BF01463939\u003c/li\u003e\n \u003cli\u003eMayer, R. E. (2004). Should There Be a Three-Strikes Rule Against Pure Discovery Learning? \u003cem\u003eAmerican Psychologist\u003c/em\u003e, \u003cem\u003e59\u003c/em\u003e(1), 14\u0026ndash;19 . https://doi.org/10.1037/0003-066x.59.1.14\u003c/li\u003e\n \u003cli\u003eMayer, R. E. (2012). Getting Started on the Road to Applying the Science of Learning. In \u003cem\u003eApplied Cognitive Psychology\u003c/em\u003e, \u003cem\u003e26\u003c/em\u003e(2), 330\u0026ndash;331. https://doi.org/10.1002/acp.1829\u003c/li\u003e\n \u003cli\u003eMayer, R. E. (2014). \u003cem\u003eCognitive Theory of Multimedia Learning\u003c/em\u003e (2nd ed). Cambridge University Press, 43\u0026ndash;71. https://doi.org/10.1017/CBO9781139547369.005\u003c/li\u003e\n \u003cli\u003eMayer, R. E. (2024). The Past, Present, and Future of the Cognitive Theory of Multimedia Learning. \u003cem\u003eEducational Psychology Review\u003c/em\u003e, \u003cem\u003e36\u003c/em\u003e(1), 1\u0026ndash;25. https://doi.org/10.1007/s10648-023-09842-1\u003c/li\u003e\n \u003cli\u003eMcNamara, D. S. (2017). Self-Explanation and Reading Strategy Training (SERT) Improves Low-Knowledge Students\u0026rsquo; Science Course Performance. \u003cem\u003eDiscourse Processes\u003c/em\u003e, \u003cem\u003e54\u003c/em\u003e(7), 479\u0026ndash;492. https://doi.org/10.1080/0163853X.2015.1101328\u003c/li\u003e\n \u003cli\u003eMolenaar, I. (2022). Towards hybrid human-AI learning technologies. \u003cem\u003eEuropean Journal of Education\u003c/em\u003e, \u003cem\u003e57\u003c/em\u003e(4), 632\u0026ndash;645. https://doi.org/10.1111/ejed.12527\u003c/li\u003e\n \u003cli\u003eMolenaar, I., \u0026amp; J\u0026auml;rvel\u0026auml;, S. (2014). Sequential and temporal characteristics of self and socially regulated learning. \u003cem\u003eMetacognition and Learning\u003c/em\u003e, \u003cem\u003e9\u003c/em\u003e, 75\u0026ndash;85. https://doi.org/10.1007/s11409-014-9114-2\u003c/li\u003e\n \u003cli\u003eMurayama, K. (2022). A reward-learning framework of knowledge acquisition: An integrated account of curiosity, interest, and intrinsic\u0026ndash;extrinsic rewards. \u003cem\u003ePsychological Review\u003c/em\u003e, \u003cem\u003e129\u003c/em\u003e(1), 175. https://doi.org/10.1037/rev0000349\u003c/li\u003e\n \u003cli\u003eNie, A., Chandak, Y., Suzara, M., Ali, M., Woodrow, J., Peng, M., Sahami, M., Brunskill, E., \u0026amp; Piech, C. (2024). The GPT Surprise: Offering Large Language Model Chat in a Massive Coding Class Reduced Engagement but Increased Adopters Exam Performances\u003cem\u003e, arXiv,\u0026nbsp;\u003c/em\u003e1\u0026ndash;33. http://arxiv.org/abs/2407.09975\u003c/li\u003e\n \u003cli\u003ePedaste, M., M\u0026auml;eots, M., Siiman, L. A., de Jong, T., van Riesen, S. A. N., Kamp, E. T., Manoli, C. C., Zacharia, Z. C., \u0026amp; Tsourlidaki, E. (2015). Phases of inquiry-based learning: Definitions and the inquiry cycle. \u003cem\u003eEducational Research Review\u003c/em\u003e, \u003cem\u003e14,\u0026nbsp;\u003c/em\u003e47\u0026ndash;61. https://doi.org/10.1016/j.edurev.2015.02.003\u003c/li\u003e\n \u003cli\u003ePekrun, R. (2006). The control-value theory of achievement emotions: Assumptions, corollaries, and implications for educational research and practice. \u003cem\u003eEducational psychology review\u003c/em\u003e, \u003cem\u003e18\u003c/em\u003e, 315\u0026ndash;341. https://doi.org/10.1007/s10648-006-9029-9\u003c/li\u003e\n \u003cli\u003ePintrich, P. R. R., Smith, D., Garcia, T., \u0026amp; McKeachie, W. (1991). A manual for the use of the Motivated Strategies for Learning Questionnaire (MSLQ). \u003cem\u003eERIC,\u0026nbsp;\u003c/em\u003e1\u0026ndash;76. Retrieved from: https://files.eric.ed.gov/fulltext/ED338122.pdf\u003c/li\u003e\n \u003cli\u003ePonce, H. R., Mayer, R. E., Loyola, M. S., \u0026amp; L\u0026oacute;pez, M. J. (2020). Study Activities That Foster Generative Learning: Notetaking, Graphic Organizer, and Questioning. \u003cem\u003eJournal of Educational Computing Research\u003c/em\u003e, \u003cem\u003e58\u003c/em\u003e(2), 1\u0026ndash;22. https://doi.org/10.1177/0735633119865554\u003c/li\u003e\n \u003cli\u003eQadir, J. (2023). Engineering Education in the Era of ChatGPT: Promise and Pitfalls of Generative AI for Education. In \u003cem\u003eIEEE Global Engineering Education Conference, EDUCON\u003c/em\u003e, 1\u0026ndash;9. https://doi.org/10.1109/EDUCON54358.2023.10125121\u003c/li\u003e\n \u003cli\u003eRafner, J., Beaty, R. E., Kaufman, J. C., Lubart, T., \u0026amp; Sherson, J. (2023). Creativity in the age of generative AI. \u003cem\u003eNature Human Behaviour\u003c/em\u003e, \u003cem\u003e7\u003c/em\u003e(11), 1836\u0026ndash;1838. https://doi.org/10.1038/s41562-023-01751-1\u003c/li\u003e\n \u003cli\u003eRellensmann, J., Schukajlow, S., Blomberg, J., \u0026amp; Leopold, C. (2022). Effects of drawing instructions and strategic knowledge on mathematical modeling performance: Mediated by the use of the drawing strategy. \u003cem\u003eApplied Cognitive Psychology\u003c/em\u003e, \u003cem\u003e36\u003c/em\u003e(2), 402\u0026ndash;417. https://doi.org/10.1002/acp.3930\u003c/li\u003e\n \u003cli\u003eRenkl, A. (1997). Learning from worked-out examples: A study on individual differences. \u003cem\u003eCognitive Science\u003c/em\u003e, \u003cem\u003e21\u003c/em\u003e(1), 1\u0026ndash;29. https://doi.org/10.1207/s15516709cog2101_1\u003c/li\u003e\n \u003cli\u003eRenkl, A., \u0026amp; Scheiter, K. (2017). Studying Visual Displays: How to Instructionally Support Learning. \u003cem\u003eEducational Psychology Review\u003c/em\u003e, \u003cem\u003e29\u003c/em\u003e(3), 599\u0026ndash;621. https://doi.org/10.1007/s10648-015-9340-4\u003c/li\u003e\n \u003cli\u003eRoediger III, H. L., \u0026amp; Karpicke, J. D. (2006). Test-enhanced learning: Taking memory tests improves long-term retention. \u003cem\u003ePsychological Science, 17\u003c/em\u003e(3), 249\u0026ndash;255. https://doi.org/10.1111/j.1467-9280.2006.01693.x\u003c/li\u003e\n \u003cli\u003eRoelle, J., \u0026amp; Berthold, K. (2017). Effects of incorporating retrieval into learning tasks: The complexity of the tasks matters. \u003cem\u003eLearning and Instruction\u003c/em\u003e, \u003cem\u003e49,\u0026nbsp;\u003c/em\u003e142\u0026ndash;156. https://doi.org/10.1016/j.learninstruc.2017.01.008\u003c/li\u003e\n \u003cli\u003eRoschelle, J., Lester, J., Fusco, J., Safir, A., Johnstun, K., Trettin, S., Chhin, C., Metz, E., Digital Promise colleagues, O., Cator, K., Means, B., Bellin, M., \u0026amp; Van Ostrand, K. (2020). AI and the Future of Learning: Expert Panel Report. \u003cem\u003eDigital Promise,\u003c/em\u003e 1\u0026ndash;27. Retrived from: https://circls.org/wp-content/uploads/2020/11/CIRCLS-AI-Report-Nov2020.pdf\u003c/li\u003e\n \u003cli\u003eRoscoe, R. D., \u0026amp; Chi, M. T. H. (2007). Understanding tutor learning: Knowledge-building and knowledge-telling in peer tutors\u0026rsquo; explanations and questions. \u003cem\u003eReview of Educational Research\u003c/em\u003e, \u003cem\u003e77\u003c/em\u003e(4), 534\u0026ndash;574. https://doi.org/10.3102/0034654307309920\u003c/li\u003e\n \u003cli\u003eSchmeck, A., Mayer, R. E., Opfermann, M., Pfeiffer, V., \u0026amp; Leutner, D. (2014). Drawing pictures during learning from scientific text: testing the generative drawing effect and the prognostic drawing effect. \u003cem\u003eContemporary Educational Psychology\u003c/em\u003e, \u003cem\u003e39\u003c/em\u003e(4), 275\u0026ndash;286. https://doi.org/10.1016/j.cedpsych.2014.07.003\u003c/li\u003e\n \u003cli\u003eShneiderman, B. (2020). Human-centered artificial intelligence: reliable, safe \u0026amp; trustworthy. \u003cem\u003eInt. J. Hum. Comput. Interact. 36,\u003c/em\u003e 495\u0026ndash;504. https://doi.org/10.1080/10447318.2020.1741118\u003c/li\u003e\n \u003cli\u003eSchramowski, P., Turan, C., Andersen, N., Rothkopf, C. A., \u0026amp; Kersting, K. (2022). Large pre-trained language models contain human-like biases of what is right and wrong to do. \u003cem\u003eNature Machine Intelligence\u003c/em\u003e, \u003cem\u003e4\u003c/em\u003e(3), 258\u0026ndash;268. https://doi.org/10.1038/s42256-022-00458-8\u003c/li\u003e\n \u003cli\u003eSchukajlow, S., Krawitz, J., Kanefke, J., \u0026amp; Rakoczy, K. (2022). Interest and performance in solving open modeling problems and closed real-world problems. \u003cem\u003ePME, 3.\u0026nbsp;\u003c/em\u003e Retrieved from: https://ivv5hpp.uni-muenster.de/u/sschu_12/pdf/Publikationen/Schukajlow_etal_2022_PME45.pdf\u003c/li\u003e\n \u003cli\u003eSchwamborn, A., Mayer, R. E., Thillmann, H., Leopold, C., \u0026amp; Leutner, D. (2010). Drawing as a Generative Activity and Drawing as a Prognostic Activity. \u003cem\u003eJournal of Educational Psychology\u003c/em\u003e, \u003cem\u003e102\u003c/em\u003e(4), 872\u0026ndash;879. https://doi.org/10.1037/a0019640\u003c/li\u003e\n \u003cli\u003eSimonsmeier, B. A., Flaig, M., Deiglmayr, A., Schalk, L., \u0026amp; Schneider, M. (2022). Domain-specific prior knowledge and learning: A meta-analysis. \u003cem\u003eEducational Psychologist\u003c/em\u003e, \u003cem\u003e57\u003c/em\u003e(1), 31\u0026ndash;54. https://doi.org/10.1080/00461520.2021.1939700\u003c/li\u003e\n \u003cli\u003eSternberg, S. (1966). High-speed scanning in human memory. \u003cem\u003eScience, 153\u003c/em\u003e(3736), 652\u0026ndash;654. https://doi.org/10.1126/science.153.3736.652\u003c/li\u003e\n \u003cli\u003eSternberg, S. (1969). Memory-scanning: Mental processes revealed by reaction-time experiments. American Scientist, 57(4), 421\u0026ndash;457. Retrieved from: https://www.sas.upenn.edu/~saul/am.scientist69.pdf\u003c/li\u003e\n \u003cli\u003eSweller, J., van Merri\u0026euml;nboer, J. J. G., \u0026amp; Paas, F. (2019). Cognitive Architecture and Instructional Design: 20 Years Later. In \u003cem\u003eEducational Psychology Review\u003c/em\u003e,\u003cem\u003e\u0026nbsp;31\u003c/em\u003e(2), 261\u0026ndash;292. https://doi.org/10.1007/s10648-019-09465-5\u003c/li\u003e\n \u003cli\u003eTaecharungroj, V. (2023). \u0026ldquo;What Can ChatGPT Do?\u0026rdquo; Analyzing Early Reactions to the Innovative AI Chatbot on Twitter. \u003cem\u003eBig Data and Cognitive Computing\u003c/em\u003e, \u003cem\u003e7\u003c/em\u003e(1), 1\u0026ndash;10. https://doi.org/10.3390/bdcc7010035\u003c/li\u003e\n \u003cli\u003eTouvron, H., Lavril, T., Izacard, G., Martinet, X., Lachaux, M., Lacroix, T., Rozi\u0026egrave;re, B., Goyal, N., Hambro, E., Azhar, F., Rodriguez, A., Joulin, A., Grave, E., \u0026amp; Lample, G. (2023). LLAMA: Open and Efficient Foundation Language Models. \u003cem\u003eArXi,\u0026nbsp;\u003c/em\u003e1\u0026ndash;17. https://doi.org/https://doi.org/10.48550/arXiv.2302.13971\u003c/li\u003e\n \u003cli\u003evan de Pol, J., van Loon, M., van Gog, T., Braumann, S., \u0026amp; de Bruin, A. (2020). Mapping and Drawing to Improve Students\u0026rsquo; and Teachers\u0026rsquo; Monitoring and Regulation of Students\u0026rsquo; Learning from Text: Current Findings and Future Directions. In \u003cem\u003eEducational Psychology Review\u003c/em\u003e,\u003cem\u003e\u0026nbsp;32\u003c/em\u003e(4), 951\u0026ndash;977. https://doi.org/10.1007/s10648-020-09560-y\u003c/li\u003e\n \u003cli\u003eVan Meter, P. (2001). Drawing construction as a strategy for learning from text. \u003cem\u003eJournal of Educational Psychology\u003c/em\u003e, \u003cem\u003e93\u003c/em\u003e(1), 129\u0026ndash;977. https://doi.org/10.1037/0022-0663.93.1.129\u003c/li\u003e\n \u003cli\u003eVenkatesh, V., \u0026amp; Bala, H. (2008). Technology acceptance model 3 and a research agenda on interventions. \u003cem\u003eDecision sciences\u003c/em\u003e, \u003cem\u003e39\u003c/em\u003e(2), 273\u0026ndash;315. https://doi.org/10.1111/j.1540-5915.2008.00192.x\u003c/li\u003e\n \u003cli\u003eVygotsky, L. S. (1978). Mind and Society: The Development of Higher Psychological Processes. In \u003cem\u003eHarvard University Press,\u0026nbsp;\u003c/em\u003e1\u0026ndash;174. https://doi.org/10.2307/j.ctvjf9vz4\u003c/li\u003e\n \u003cli\u003eWillingham, D. T. (2008). Critical Thinking: Why Is It So Hard to Teach? \u003cem\u003eArts Education Policy Review\u003c/em\u003e, \u003cem\u003e109\u003c/em\u003e(4), 21\u0026ndash;32. https://doi.org/10.3200/AEPR.109.4.21-32\u003c/li\u003e\n \u003cli\u003eWittrock, M. C. (1989). Generative Processes of Comprehension. \u003cem\u003eEducational Psychologist\u003c/em\u003e, \u003cem\u003e24\u003c/em\u003e(4), 345\u0026ndash;275. https://doi.org/10.1207/s15326985ep2404_2\u003c/li\u003e\n \u003cli\u003eWittrock, M. C., \u0026amp; Farley, F. (2010). Learning as a generative process. \u003cem\u003eEducational Psychologist\u003c/em\u003e, \u003cem\u003e45\u003c/em\u003e(1), 40\u0026ndash;45. https://doi.org/10.1080/00461520903433554\u003c/li\u003e\n \u003cli\u003eWittwer, J., \u0026amp; Renkl, A. (2010). How Effective are Instructional Explanations in Example-Based Learning? A Meta-Analytic Review. In \u003cem\u003eEducational Psychology Review, 22\u003c/em\u003e(4), 393\u0026ndash;409. https://doi.org/10.1007/s10648-010-9136-5\u003c/li\u003e\n \u003cli\u003eWu, R., \u0026amp; Yu, Z. (2024). Do AI chatbots improve students learning outcomes? Evidence from a meta‐analysis. \u003cem\u003eBritish Journal of Educational Technology\u003c/em\u003e, \u003cem\u003e55\u003c/em\u003e(1), 10\u0026ndash;33.\u003c/li\u003e\n \u003cli\u003ehttps://doi.org/10.1111/bjet.13334\u003c/li\u003e\n \u003cli\u003eYan, L., Greiff, S., Teuber, Z., \u0026amp; Ga\u0026scaron;ević, D. (2024). Promises and challenges of generative artificial intelligence for human learning. \u003cem\u003eNature Human Behavior, 8,\u003c/em\u003e 1839\u0026ndash;1850. https://doi.org/10.1038/s41562-024-02004-5\u003c/li\u003e\n \u003cli\u003eYan, L., Sha, L., Zhao, L., Li, Y., Martinez-Maldonado, R., Chen, G., Li, X., Jin, Y., \u0026amp; Ga\u0026scaron;ević, D. (2023). Practical and ethical challenges of large language models in education: A systematic scoping review. In \u003cem\u003eBritish Journal of Educational Technology\u003c/em\u003e, \u003cem\u003e55\u003c/em\u003e(1), 90\u0026ndash;112. https://doi.org/10.1111/bjet.13370\u003c/li\u003e\n \u003cli\u003eZhang, Q., \u0026amp; Fiorella, L. (2023). An integrated model of learning from errors. \u003cem\u003eEducational Psychologist\u003c/em\u003e, \u003cem\u003e58\u003c/em\u003e(1), 18\u0026ndash;34. https://doi.org/10.1080/00461520.2022.2149525\u003c/li\u003e\n\u003c/ol\u003e"},{"header":"Table","content":"\u003cp\u003eTable 1 is available in the Supplementary Files section\u003c/p\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":true,"highlight":"","institution":"University of Copenhagen","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":true,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"generative artificial intelligence, generative sense making, learning, teaching, self-explaining","lastPublishedDoi":"10.21203/rs.3.rs-5622133/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-5622133/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eGenerative Artificial Intelligence (GenAI) has emerged as a transformative tool in education, offering scalable, individualized learning experiences. However, there is a notable lack of theoretically informed and methodologically rigorous research on how GenAI can effectively augment learning. This study addresses this gap by investigating the potential of a theoretically informed GenAI chatbot, ChatTutor, to facilitate generative sense-making, leveraging principles from generative learning theory. The study had two primary goals: first, to build on theory to propose how GenAI could be used to support generative sense-making; and second, to empirically test its impact on conceptual knowledge, self-efficacy, and trust immediately after the intervention, as well as on conceptual knowledge, enjoyment, and behavioral intentions in a follow-up test four weeks later. Conducted in an authentic university course, the pre-registered experiment with 175 students compared ChatTutor to a generic GenAI system (ChatGPT) and a teaching-as-usual condition. Results show that ChatTutor significantly enhanced trust, enjoyment, and behavioral intentions but not self-efficacy. While ChatTutor improved conceptual knowledge over ChatGPT immediately after the intervention, it did not outperform teaching-as-usual. Four weeks later, ChatTutor significantly outperformed teaching-as-usual but not ChatGPT in conceptual knowledge. The manuscript underscores the importance of integrating human-centered design and educational psychology theories into GenAI applications to optimize learning outcomes and proposes future research and practical implications.\u003c/p\u003e","manuscriptTitle":"Beyond the \"Wow\" factor: Using Generative AI for Increasing Generative Sense-Making","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2024-12-12 05:27:46","doi":"10.21203/rs.3.rs-5622133/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"31dca42c-6416-4c5a-b2dd-7a163aca6d99","owner":[],"postedDate":"December 12th, 2024","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":41449426,"name":"Educational Psychology"}],"tags":[],"updatedAt":"2025-04-29T19:55:01+00:00","versionOfRecord":[],"versionCreatedAt":"2024-12-12 05:27:46","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-5622133","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-5622133","identity":"rs-5622133","version":["v1"]},"buildId":"qtupq5eGEP_6zYnWcrvyt","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}