GPT-PACK: Can GenAI act as TPACK Peer Tutor for Preservice Teachers? A Comparative Study on Objective TPACK of Humans and ChatGPT 3.5

GPT-PACK: Can GenAI act as TPACK Peer Tutor for Preservice Teachers? A Comparative Study on Objective TPACK of Humans and ChatGPT 3.5 | 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 GPT-PACK: Can GenAI act as TPACK Peer Tutor for Preservice Teachers? A Comparative Study on Objective TPACK of Humans and ChatGPT 3.5 Ferdi Çelik, Ceylan Yangın Ersanlı, Aaron Drummond This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-3388153/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 This study employs a single case-control design to evaluate the objective TPACK performance of ChatGPT by comparing it to human preservice teachers. A purposive sampling method selected ChatGPT 3.5 as a potential GenAI peer tutor and 93 Australian preservice teachers as the normative sample. Objective TPACK performance was measured using an adapted TPACK scale. Bayesian analysis revealed that ChatGPT significantly outperformed the human control group with a substantial effect size. Qualitative data analysis of ChatGPT’s explanations further demonstrated its comprehensive understanding of TPACK dimensions. While certain limitations were identified, including the narrow focus on Australian preservice teachers and the availability of more advanced AI models, the study emphasizes ChatGPT 3.5’s potential as a more knowledgeable other for preservice teachers to enhance their TPACK. This research encourages further investigation into the integration of AI in education and its implications for pedagogical practices. ChatGPT 3.5’s remarkable TPACK performance suggests its role as a valuable peer tutor in TPACK-related domains, complementing human educators while recognizing the irreplaceable human aspects of the learning experience. Figures Figure 1 Figure 2 Figure 3 INTRODUCTION The integration of emerging technologies has become essential for teachers as it enhances student engagement [ 1 – 3 ] facilitates personalized learning [ 4 ], fosters digital literacy [ 5 ], and prepares students for the demands of the 21st century [ 6 ]. Given the increasing integration of digital technologies into almost every aspect of daily life, teachers need to possess in-depth Technological Pedagogical Content Knowledge (TPACK) [ 7 , 8 ], as it enables them to create meaningful and engaging learning experiences that cater to students' diverse needs and prepare them for success in a technology-driven world. Continuous support throughout their training process within Faculties of Education may yield better outcomes for preservice teachers by equipping them with the necessary skills and knowledge to teach effectively [ 9 ]. However, due to the growing student number, limited staffing resources, the demanding workload of academics, and the multi-faceted nature of their responsibilities, it is impractical for academics always to offer availability to the preservice teachers, making peer tutoring a desired option. Technology is evolving so rapidly that many preservice teachers may feel left alone and try to find self-directed ways to improve their TPACK [ 10 ](Max et al., 2023). Indeed, many online resources, such as teaching and technology websites, offer a large amount of data. Nevertheless, they lack a friendly conversation to give a hand whenever the preservice teachers need active guidance. On the other hand, generative AI (GenAI) [ 11 ], empowered by the impressive advancements in natural language processing (NLP) [ 12 ], has become an accessible and transformative tool for educators. While GenAI holds hope-inspiring potential for shaping how we teach and learn, its specific educational applications remain largely unexplored. This study proposes ChatGPT, an NLP-powered GenAI chatbot, as a possible peer tutor for preservice teachers that can offer them 24/7 support at their own pace by promoting self-directed learning, as preservice teachers can access it at any time using a smartphone, computer, or tablet connected to the internet. Grounded in the social constructivist school of thought, the Zone of Proximal Development (ZPD) Theory, developed by Lev Vygotsky [ 13 ], proposes that humans can acquire knowledge and skills through interactions with individuals who possess higher expertise, often referred to as more knowledgeable others [ 14 ]. We believe that GenAI, such as ChatGPT, may act as a supportive partner for preservice teachers, capable of providing constructive feedback [ 15 ], answering preservice teachers' questions, and having them engage in meaningful discussions [ 16 ] related to teaching practices and pedagogical strategies due to its advanced NLP capabilities. According to the principles of peer tutoring, the ideal peer tutor is typically more knowledgeable or experienced than the learner. This dynamic allows for the transfer of knowledge from the more knowledgeable peer to the learner within their ZPD. It fosters effective learning through scaffolded support and guidance provided by the more knowledgeable peer [ 17 ]. Therefore, to consider ChatGPT as a possible peer tutor for preservice teachers to enhance their TPACK, a comprehensive understanding of ChatGPT's objective TPACK performance is crucial compared to human preservice teachers. To this end, the primary aim of this study is to conduct a comprehensive evaluation, juxtaposing ChatGPT against human preservice teachers' Technological Pedagogical Content Knowledge (TPACK). LITERATURE REVIEW Natural Language Processing In recent years, text-based GenAI has experienced considerable development, and NLP is a key contributing factor [ 18 , 19 ]. It might be tempting to question whether NLP is a technology or a scientific field dealing with how humans naturally process a language. In its essence, it combines advanced computational techniques to replicate not just the surface-level mechanics of human language and use but the very essence of our communicative competence [ 20 , 21 ] and performance [ 22 ]. Therefore, NLP, in its current sense, is a technology utilizing advanced techniques to learn from the data for achieving linguistic tasks, and therefore, contributing to the development of GenAI models that are capable of comprehending, interpreting, and generating language in a manner that engages in human-like communication for various objectives. It might be worth noting that the quality of the model is contingent upon the quality of the training data, though. The roots of NLP trace back to scholarly discussions in the 1950s, driven by the need for automated translation after World War II [ 23 ]. Earliest well-known debates include the argument of Turing's “Can Machines Think?” in 1950 [ 24 ] (p. 433) and Chomsky's critique in 1955, exemplified by the ambiguous but grammatically valid sentence “Colorless green ideas sleep furiously” [ 25 ] (p. 242), highlighting the limitations of early NLP approaches. NLP has made significant progress through the collaborative effort of scholars in the fields of computer science, linguistics, psychology, and neuroscience [ 26 – 30 ]. Through several developmental stages, NLP has currently reached a point where transfer learning [ 31 – 33 ], the ability of AI to transfer the pre-trained model's learned features and adapt them to the new task using a smaller dataset, and deep learning, the ability of AI to automatically learn from the data using deep neural networks [ 34 – 36 ] have emerged as key technologies. This, in turn, equips GenAI models, such as ChatGPT, with increasingly human-like communication capabilities, making them potentially more viable peers for conversational interactions on TPACK. ChatGPT ChatGPT is a text-based GenAI developed by OpenAI. While OpenAI offers a subscription plan for users to engage with GPT 4.0 (highlighted by its capability) for a subscription fee, GPT 3.5 (highlighted by its processing speed) is more commonly used worldwide because a free research preview is available online. New Android and IOS applications have been released in less than a year, making GPT more accessible. At least a hundred million users are interacting with ChatGPT, and OpenAI reached a $ 30 billion valuation after the release of ChatGPT. This is because it may be regarded as one of the first open-access GenAI models to offer highly advanced NLP capabilities that provide human-like communication, backed by its massive knowledge base acquired during its training process. ChatGPT has been proposed as a tool in several fields, including but not limited to public health [ 37 ], improving library services [ 38 ](Lund & Wang, 2023), fixing software bugs [ 39 ], translation [ 40 ], clinical practice [ 41 ]. Specifically, in the field of learning and teaching [ 42 ] (Tlili et al., 2023), ChatGPT has been interacted with for self-directed learning for its role as a more knowledgeable other [ 15 ], teacher training[ 43 , 44 ], and improving preservice teachers' historical thinking skills [ 45 ]. Despite its great potential, it has limitations such as overreliance on training data [ 46 ], which is time-bound (currently up to 2021), and inability to access external resources. Additionally, several issues have been raised related to ethics [ 22 , 47 ], safety [ 42 ], misleading information generation [ 48 ], and the potential for racial and gender biases inherent in the models and the content they produce [ 49 ]. Technological Pedagogical Content Knowledge Technological Pedagogical and Content Knowledge, abbreviated as TPACK, is a comprehensive framework to indicate the knowledge of teachers in using technology in education [ 50 ], which was built after the pedagogical content knowledge framework [ 51 ] with the domination of technology in our everyday life. In the past, pedagogical and content knowledge (PCK) has been seen as the two main domains that make up the educational environment. It encompasses the particular area of professional expertise that educators possess, enabling them to successfully teach students content knowledge through a variety of pedagogical techniques. TPACK is a concept that has emerged as a result of Schulman's framework's evolution in the context of modern education to include a new technology dimension. This paradigm shift emphasizes how important technology-supported educational approaches are becoming in contemporary pedagogical practices. TPACK is a model that underscores the relationship between the content, pedagogical, and technological knowledge of teachers and is required for the effective integration of technology into teaching [ 52 ]. According to Harris [ 53 ], TPACK represents the intertwined and context-specific knowledge essential for proficiently incorporating digital tools and resources into curriculum-driven instruction. The framework involves three fundamental, interrelated constructs; Pedagogical Content Knowledge refers to the integration of subject matter proficiency and the mastery of pedagogical skills. It involves the effective synthesis of content expertise and instructional methods. Technological Content Knowledge focuses on how the utilization of technology influences the ways content is presented. This involves recognizing how technology alters content representation and learning dynamics. Technological Pedagogical Knowledge encompasses a teacher's awareness of the array of technological tools available and their capacity to effectively select and employ these technologies to enhance the desired learning outcomes. It entails the strategic integration of technology within pedagogical contexts to optimize educational experiences. The TPACK theoretical framework is frequently employed for the analysis of educators' integrated knowledge encompassing technology, pedagogy, and content. This approach makes it easier to examine how teachers use technology to present, organize, and change subject information in a way that improves students' understanding of it. METHOD Research Design A single case-control design was chosen for this study to conduct a robust comparison of the objective TPACK performance of ChatGPT to that of human preservice teachers [ 54 ]. A single case-control design is a research approach that involves the comparison of a single case (in the present study, GPT 3.5) with a control group (human preservice teachers) to draw meaningful conclusions about the case's performance [ 55 ]. Thus, it was a good fit for the present study. As this design provided a structured framework for examining the performance of a single case compared to a normative or control sample through Bayesian statistics [ 54 , 56 , 57 ], it contributed to a well-rounded understanding of GPT 3.5's potential as a more knowledgeable other . Sampling This study employed a purposive sampling method [ 58 ]. We chose GPT 3.5 as a possible GenAI peer tutor due to its advanced NLP capabilities, accessibility, language processing speed, user-friendly interface, sophisticated knowledge base, and free-to-use feature. We also utilized the objective TPACK performance of 93 Australian preservice teachers (32 males) [ 59 ] as normative data considering the aim of the present study. Instruments There were a variety of scales measuring the TPACK levels of the preservice teachers, such as TPACK-deep, TPACK-21 scale [ 60 ], TPACK SAS [ 61 ], and TPACK-G [ 62 ]. However, these measures heavily relied on self-reports provided by preservice teachers. Using the data collected via these scales could introduce issues related to social desirability bias, in which preservice teachers might have provided responses they found socially acceptable instead of reflecting their true understanding or skills. Additionally, asking GPT to answer self-report items might not be desirably reliable. After a rigorous literature review and based on our evaluation as scholars, we decided to use the objective TPACK test. This was also because the correlation between how preservice teachers subjectively view their proficiency in TPACK (TPACK-deep) [ 63 ] and the concrete retention of that knowledge (objective-TPACK) was relatively weak [ 59 ]. In addition to collecting GPT 3.5's true and false responses to the objective-TPACK test, we had GPT 3.5 provide explanations for each response to comprehensively understand its TPACK performance by analyzing this qualitative data. This would also enhance the credibility and validity of findings by triangulating the data [ 64 ]. To sum up, in the present study, we collected both quantitative and qualitative data. We collected the quantitative data through the objective-TPACK scale and qualitative data by triggering GPT's explanations for each item in the scale. [ 65 ], which reveals whether the correct responses are given by chance or based on knowledge. As the case that would take the objective-TPACK test was a GenAI, we needed to investigate all the items in the scale carefully and adapt them as necessary. A few changes had been made: string “(following you)” was added after the word “students” in the 8th item, and the string “to be taught to students” was added at the end of the 11th item for clarification after consulting with the developers of the objective-TPACK test. As the original scale was developed in 2013 and published in 2016, we had to scrutinize whether the items were still valid regarding the TPACK performance of participants. We concluded that the items were valid and that there was no need for further changes. TPACK Objective Dataset As a base performance of human preservice teachers, we analyzed the TPACK Objective dataset [ 66 ], a publicly available dataset containing responses of 93 Australian preservice teachers (mean age 22.7) to the objective-TPACK scale. The original dataset included dedicated columns for d' prime and c index scores. We recalculated the scores in the dataset to verify the scores and ensure the computing steps to apply the consistent analysis to analyze the TPACK performance of GPT 3.5 and compare it. Procedure After having the necessary permissions to use the scale and TPACK Objective dataset, an account was created on the official OpenAI website to access ChatGPT (version 3.5). We used “Can you be an English language teacher?” as the first prompt. Then, we used the “Say true or false: The design of interactive teaching materials supports different learning theories” prompt, which contained the first item in the Objective TPACK scale. As the first prompt triggered GPT to generate explanations, we did not need to ask again. As for the following prompts, we entered the test items respectively. The outputs of GPT 3.5 were exported and saved in a text document for analysis. Data Analysis In order to assess GPT 3.5's TPACK performance compared to human preservice teachers, we analyzed the quantitative data acquired through objective TPACK scale. The quantitative data was initially analyzed based on the signal detection theory, a mathematical theory commonly used by neuroscientists [ 67 , 68 ]. After entering the GPT 3.5's answers to an MS Excel file, we summed its correct answers to true items (hits) and incorrect answers to false items (false alarms). A flattening constant was applied to prevent failing the calculation of d' and c scores by subtracting 0.5 from the perfect scores and adding 0.5 to the scores of 0 while they were still the lowest/highest possible score. Next, we computed the hit rate by dividing the number of corrected hits by the total count of true items, and we calculated the false alarm rate by dividing the number of corrected false alarms by the total number of false items. Then, both the false alarm rate and hit rate needed to be transformed to a normalized Z value, and d' score is calculated by subtracting the normalized hit rate from the normalized false alarm rate. After acquiring the TPACK d' and c scores of GPT 3.5, we compared it to that of human preservice teachers by using the Bayesian analysis method of [ 57 ], which was developed to compare a single case's performance with that of a normative or control sample. This was because Bayesian analysis would provide a more intuitive way to handle small sample sizes, which was often the case when assessing AI systems. Common statistical computing software used in social sciences (e.g. IBM's SPSS v26) did not have a built-in feature for this analysis. Therefore, we used SingleBayes_ES (a dedicated PC software built based on the original manuscript [ 54 , 57 ]. Using Bayesian methods produced probability distributions, which enabled us to quantify uncertainty and variability in our conclusions rather than solely relying on point estimates. This ensured that our findings were not only statistically robust but also reflective of the inherent uncertainties in our data. Lastly, for the qualitative data, a thematic analysis was conducted [ 69 ]. RESULTS The present study aimed to evaluate GPT 3.5's TPACK performance compared to 93 human preservice teachers to propose GPT as a peer tutor. A meticulous analysis compared human preservice teachers and the possible AI-driven peer tutor, ChatGPT. Quantitative Data The normative sample, comprising 93 human preservice teachers, exhibited a mean TPACK score of 0.85 and a standard deviation of 0.57. These parameters provided a contextual foundation for the subsequent comparisons. Upon calculating ChatGPT's d' score, we have, intriguingly, seen that ChatGPT achieved a remarkable TPACK performance score of 2.68 out of a possible 3.07. To determine whether this result was higher than the average performance of the normative sample, we used Bayesian one-sample t-tests employing default priors to compare GPT 3.5's scores with the control sample. This analysis yielded definitive evidence that ChatGPT's score (2.68) was higher than the average performance of the normative sample ( M = 0.85, SD = 0.57), BF 10 = 4.41 x 10 46 , δ = 3.11 95% CI [2.747, 3.525]. The Bayes factor suggests that it is 4.41 x 10 46 times more likely that there is a difference between ChatGPT's performance and the performance of the normative sample with a considerable effect size. Figure 1 shows the prior and posterior effect sizes for the difference between ChatGPT and the control sample. We utilized the Bayesian method of Crawford-Garthwaite [57] to scrutinize the proportion of control sample scores that would fall below ChatGPT 3.5's performance and control sample scores using SingleBayes ES software. This analysis suggested that approximately 99.9037% of the control population would obtain a score lower than ChatGPT. Figure 2 shows robustness checks for the difference between ChatGPT and the control sample based on different Cauchy prior widths. The priors adopted within a Bayesian analysis can impact whether the evidence for an effect is considered strong enough to increase our confidence in our priors – for instance, if we expect a large effect, default priors might overestimate the amount of evidence a given analysis yields because the default priors are smaller than what was expected to be observed. Our analysis used uninformed default priors with a Cauchy prior width of 0.707. As shown in Figure 2, the analysis is robust to a wide range of priors, with evidence remaining definitive that ChatGPT outperformed the normative sample irrespective of the priors chosen for this analysis. Figure 3 shows a graph of how evidence accumulates across a sequential analysis of the data. Although Bayesian analyses are more robust to outliers and small n samples, outliers, and extreme scores can sometimes impact the results as they do in a frequentist analysis. Figure 3 shows that as more cases are included in the analysis, a steady accumulation of evidence occurs, suggesting that no particular score or scores greatly influence the results. This analysis increases our confidence that the findings accurately represent the data. Ideally, a perfect responder will have a c score of 0, indicating no bias in their response. Positive c scores indicate a tendency to report false answers to the questions irrespective of the correct response. In contrast, negative c scores indicate a tendency to report true to answers irrespective of the correct response. We analyzed GPT 3.5's responses using the c index; the analysis indicated that the model provided relatively well-calibrated answers based on its existing knowledge base rather than random guessing when answering questions (c = -0.19). Although ChatGPT appeared to be slightly more likely to report true than the control sample (which did not differ from 0; M = -0.037, SD = 0.399), the effect size was relatively small, BF 10 = 55.2, δ = 0.369 95% CI [0.160, 0.578], and the absolute deviation from zero was slight. This suggests that GPT 3.5 is well calibrated to the objective TPACK task. Overall, these results shed light on the model's ability to draw from its training to generate responses, emphasizing its capacity to provide contextually relevant information regarding TPACK. However, it was essential to consider other evaluation measures to gain a comprehensive understanding of the model's performance and knowledge utilization. Therefore, we analyzed the qualitative data from GPT 3.5 in the following subsection. Qualitative Data Design The design is a crucial aspect of TPACK within education to create effective learning experiences for the learners. Findings indicated that, in the provided GPT explanations, the design aspect was comprehensively explored. The responses highlighted that interactive teaching materials could align with learning theories and cater to diverse styles. The responses indicated that GPT critically assessed that using the latest technology would not guarantee improved outcomes; it was about thoughtful integration and alignment with content and pedagogy. Additionally, the importance of identifying students' needs was underscored by GPT, which it thought was for enabling tailored technology use. The analysis demonstrated GPT's grasp of how design involves aligning technological tools with pedagogical approaches and content to enhance learning: For example, behaviorist approaches may use interactive materials that offer immediate feedback and reinforcement, while constructivist approaches may use materials that encourage exploration and critical thinking. Ultimately, the goal is to engage learners and enhance their understanding and retention of the subject matter. (Response to #1 item) GPT 3.5 acknowledged the potential benefits of technology while cautioning that improved learning outcomes were not guaranteed. GPT suitably emphasized the importance of effective integration, quality content, and consideration of students' readiness and access. It depicted technology as a tool for enhancing learning rather than a guaranteed solution. Technology should be seen as a tool to enhance and support learning rather than a magic solution that automatically leads to improved learning. It's essential to combine technology with effective pedagogical practices and tailor it to meet the specific needs and goals of the students and the learning objectives. (Response to #2 item) When we asked GPT the item four, “A needs analysis of technologies to be used in the teaching and learning process involves the identification of technological skills that students could benefit from”, GPT 3.5 mistakenly asserted that a needs analysis involved identifying the technological skills that students could benefit from. However, this claim, as written in the initial survey, appeared to be more accurate in the absence of specific aims or requirements necessitating particular technological competencies. While the response correctly described the systematic nature of a needs analysis and its role in aligning technology with student requirements, it overlooked the crucial point that technological skill identification was not a default component of every needs analysis. When we checked the raw data, we saw that it was GPT's only incorrect answer in the objective TPACK test: Yes, you are absolutely right. A needs analysis of technologies to be used in the teaching and learning process involves identifying the technological skills that students could benefit from. (...) In the context of educational technology, a needs analysis may include the following steps: (...) By conducting a comprehensive needs analysis, educators can make informed decisions about integrating appropriate technologies into the teaching and learning process, leading to a more effective and engaging educational experience for students. Overall, the analysis revealed that GPT effectively addressed the design dimension of TPACK. While one response contained a misleading assertion, the overall insights offered a sound understanding of the design dimension, underscored by the importance of informed technology selection based on a holistic understanding of educational needs. Exertion Regarding the exertion theme, we saw that the provided explanations offered insights into how technology could enhance student participation and success. While acknowledging the motivational potential of technology, GPT's explanations also highlighted the importance of balanced and thoughtful integration. It recognized that technology alone was not a guaranteed motivator and emphasized the need for alignment with learning objectives and instructional strategies. While technology can be a powerful motivational tool, it's essential to use it thoughtfully and in conjunction with effective pedagogical practices. The success of technology in motivating students depends on how well it aligns with the learning objectives, the quality of the content, and the support provided by educators in integrating technology effectively into the teaching process. (Response to #5 item) Additionally, the responses acknowledged potential challenges and considerations, such as the limitations of technology-based assessments and the varying impact of social media on education. These insights reflected a comprehensive understanding of the exertion dimension within the TPACK framework, highlighting both the opportunities and potential pitfalls of utilizing technology to enhance student engagement and motivation. However, it's crucial to recognize that technology should complement traditional assessment methods and not replace them entirely. Educators must consider the limitations of technology-based assessments and use a balanced approach to evaluate students' success effectively. Technology can enhance assessment practices by providing efficiency, scalability, and additional insights into students' learning, but human judgment and qualitative assessments remain essential components of a comprehensive evaluation process. (Response to #6 item) Overall, the analyses demonstrated an understanding of how technology could play a motivational role in education while also acknowledging the complexities involved. Ethics GPT correctly emphasized that copyright issues are pertinent for teachers when designing and sharing digital resources, underlining the importance of respecting intellectual property rights. Furthermore, the analyses appropriately stressed the significance of addressing student safety and privacy concerns related to technology-based information, reflecting an awareness of ethical responsibilities in the digital age. Copyright is a crucial issue for teachers when designing and sharing digital resources. Just like any other content creators, teachers must be mindful of copyright laws and respect intellectual property rights. When designing and sharing digital resources, teachers need to consider the following copyright-related aspects: (Response to #9th item) The responses also provide accurate insights into the potential impact of technology on copyright compliance. GPT mentioned the Australian Copyright Act, but not the curriculum. Nevertheless, its awareness of legal and ethical implications demonstrated a sound understanding of the need for responsible use of digital resources. The statement is not universally true. Showing television programs to students for educational purposes while employed by a school in Australia may or may not breach copyright, depending on the specific circumstances and how the use aligns with copyright laws and exceptions. While there are provisions for fair dealing for educational purposes under the Australian Copyright Act 1968, fair dealing has limitations and conditions that must be met. (Response to #10) The responses regarding the 11th statement, which concerns the Australian Curriculum's requirement to teach student safety and privacy of technology-based information, demonstrated a clear awareness of the ethical and educational obligations associated with technology integration. As time-bound data was usually considered a limitation of GPT to be used for educational purposes, it appropriately warned about it. As of my last update in September 2021, the Australian Curriculum does emphasize the importance of addressing issues related to student safety(...) It's important to note that curriculum requirements and guidelines can evolve, so I recommend checking with relevant educational authorities or resources for the most up-to-date information. (Response to #11) We found that GPT 3.5's responses showcased a solid understanding of the ethics dimension by addressing key ethical considerations such as copyright, privacy, and responsible technology use, further supporting our initial results. The explanations of GPT demonstrated an awareness of the need for educators to navigate ethical challenges when integrating technology into the educational environment. To conclude this section, the quantitative and qualitative data analyses revealed that GPT 3.5 had remarkable TPACK performance, suggesting that it can be a more knowledgeable other for human preservice teachers to enhance their TPACK. Its responses to TPACK prompts highlighted how NLP could generate explanations in a variety of TPACK questions with a guiding response, which confirmed GPT 3.5 as a potential TPACK peer tutor. LIMITATIONS Unfortunately, there were some limitations. The normative sample consisted of only Australian preservice teachers; data from different countries could enhance the generalizability of the findings. A more advanced version of ChatGPT was available, which was GPT 4.0. It could showcase an even better performance than 3.5. However, it was a paid subscription that could discourage some preservice teachers from testing it as a TPACK peer tutor in future studies. Moreover, even with a huge amount of training data, GenAI models, including GPT, are not without limitations. GenAI models may not interpret or generate language the same way humans do, while exactly how humans process language is still a topic of scholarly discussion. For instance, recent models may fail to process and generate desired language in making social judgments [70, 71], verbalizing abstract things based on prior knowledge [29], critically responding to scientific questions with human reasoning [72], and explaining the reason behind their own behaviors [73]. Additionally, as with every GenAI model, its knowledge is limited to what it was trained on despite its remarkable few-shot performance [74] and advancing zero-shot performance [75]. DISCUSSION & CONCLUSION In the present study, we examined the objective TPACK performance of GPT 3.5 in comparison to human preservice teachers to test the hypothesis that ChatGPT can act as a more knowledgeable other [15, 76-78]. We specifically focused on determining whether ChatGPT could take on the role of a peer tutor for TPACK. Our findings provide compelling evidence to support the hypothesis that ChatGPT can serve as a peer tutor. By revealing the performance of GPT 3.5 as a more knowledgeable other in TPACK, our study suggests that GPT 3.5 can act as a peer tutor for preservice teachers to improve their TPACK. Through careful analyses, we observed that GPT 3.5 demonstrated a level of TPACK surpassing that of human preservice teachers. Its ability to generate coherent and contextually relevant responses across a range of TPACK questions indicated a great knowledge of education's content and technological aspects. ChatGPT could answer learners’ questions and enhance collaboration [79]. This suggests that GPT 3.5 has the potential to play a significant role in education [80] by aiding preservice teachers in their journey of teaching with positive learner perceptions towards GenAI in education [81]. Based on our findings, we may claim that learners can learn by interacting with GPT 3.5, taking on the role of a more knowledgeable other. GPT 3.5's far-out performance in the objective TPACK showcases the advancements in AI and natural language processing [42], where an AI model can simulate the role of a more knowledgeable peer tutor by supporting preservice teachers in TPACK. Moreover, as the present study reveals, it highlights the adaptability and versatility of GPT 3.5, which could be integrated into various educational contexts, including TPACK peer tutoring. Therefore, our findings encourage educators to explore innovative ways to integrate NLP-powered AI tools into their teaching methodologies in Faculties of Education so that preservice teachers can take advantage of a collaborative learning environment that incorporates both human and AI expertise. This study has proposed GPT 3.5 as a more knowledgeable TPACK peer tutor by verifying its TPACK performance. Based on this, several future research suggestions have been made. Firstly, the current study predominantly focused on the objective assessment of TPACK performance. Future qualitative research could scrutinize the subjective experiences of preservice teachers when interacting with GPT 3.5 as a TPACK peer tutor. Investigating several psychological constructs, such as learners' perceptions, preferences, attitudes, and enjoyment toward GenAI peer tutors, could provide insights into such interactions' affective and cognitive dynamics. Secondly, integrating GPT 3.5 into actual educational contexts remains a practical challenge in how and to what extent ChatGPT can be employed. Further research is needed to ascertain the effective methods of incorporating GenAI into pedagogical practices. This could involve designing a GenAI-integrated curriculum and encouraging collaboration between human educators and machine tutors. Additionally, the scalability and generalizability of GPT 3.5's performance as a TPACK peer tutor across different subject domains [79] and age groups need exploration. Adapting the model's responses to cater to diverse learning styles, psychological motives of learners, and academic disciplines could enhance its utility and impact. Furthermore, ethical considerations related to GenAI TPACK peer tutoring should be rigorously examined. Research addressing issues of data privacy, bias mitigation, and the responsible use of AI in learning contexts is required to ensure that the integration of AI in education remains safe. Finally, our primary objective was to assess the objective TPACK performance of GPT 3.5, comparing it with the performance of human preservice teachers. Our research was specifically designed to assess ChatGPT's suitability as a peer tutor for TPACK-related knowledge and abilities for preservice teachers. The findings of our empirical investigation provide strong and convincing support for the claim that ChatGPT may indeed function successfully as a peer tutor for preservice teachers. This information has important implications since it shows that GPT 3.5 is capable of providing a viable way to improve preservice teachers’ TPACK competency and educational preparation. Although our data supports that ChatGPT can be a more knowledgeable other, it must be kept in mind that it is a machine that lacks emotions and feelings that play a significant role in human learning experience. Human learning encompasses not only the acquisition of facts and knowledge but also the complex interplay of emotions, social interactions, and personal experiences. Therefore, while ChatGPT can be used as a TPACK peer-tutor, it cannot replace the vital emotional and human elements that enrich the learning experience. In conclusion, this study contributes to the evolving landscape of GenAI in education by establishing GPT 3.5's competence as a TPACK peer tutor. As GenAI technology progresses, its potential to transform educational practices becomes increasingly evident. By embracing these advancements while addressing associated challenges, educators can pave the way for more personalized and effective smart learning environments. References Consoli T, Désiron J, Cattaneo A. What is technology integration and how is it measured in K-12 education? A systematic review of survey instruments from 2010 to 2021. Comput Educ 2023:104742. Çelik F, Yangın Ersanlı C. The use of augmented reality in a gamified CLIL lesson and students’ achievements and attitudes: a quasi-experimental study. Smart Learn Environ. 2022;9(1):30. Moraes EB, Kipper LM, Hackenhaar Kellermann AC, Austria L, Leivas P, Moraes JAR, Witczak M. Integration of Industry 4.0 technologies with Education 4.0: Advantages for improvements in learning. Interact Technol Smart Educ. 2023;20(2):271–87. Rizvi M. Exploring the landscape of artificial intelligence in education: Challenges and opportunities. 2023 5th International Congress on Human-Computer Interaction, Optimization and Robotic Applications (HORA): 2023: IEEE; 2023: 01–3. Wang S, Sun Z, Chen Y. Effects of higher education institutes’ artificial intelligence capability on students' self-efficacy, creativity and learning performance. Educ Inform Technol. 2023;28(5):4919–39. Kickmeier-Rust MD, Albert D. A domain model for smart 21st century skills training in game-based virtual worlds. In: 2012 IEEE 12th International Conference on Advanced Learning Technologies: 2012 : IEEE; 2012: 680–681. Budianto L, Arifani Y, Wardhono A, Poedjiastutie D. The TPACK level of in-service EFL teacher online professional training: The perspectives of teachers, students, and trainers. J Digit Learn Teacher Educ. 2023;39(2):87–101. Ekawati E, Prastyo A. Optimization of TPACK-based Project Learning in Micro-teaching Courses in Physics Education Study Programs during the Pandemic. In: Journal of Physics: Conference Series : 2022: IOP Publishing; 2022: 012035. van der Merwe D, Levigne-Lang R. The lessons learnt from emergency remote teaching to strengthen a pre-service teacher education course on lesson design. J Univ Teach Learn Pract. 2023;20(3):03. Max AL, Lukas S, Weitzel H. The pedagogical makerspace: Learning opportunity and challenge for prospective teachers' growth of TPACK. Br J Edu Technol 2023. Stokel-Walker C, Van Noorden R. What ChatGPT and generative AI mean for science. Nature. 2023;614(7947):214–6. Younis HA, Ruhaiyem NIR, Ghaban W, Gazem NA, Nasser M. A Systematic Literature Review on the Applications of Robots and Natural Language Processing in Education. Electronics. 2023;12(13):2864. Vygotsky LS, Cole M. Mind in society: Development of higher psychological processes. Harvard university press; 1978. Vygotsky LS. Thought and language. MIT press; 2012. Stojanov A. Learning with ChatGPT 3.5 as a more knowledgeable other: an autoethnographic study. Int J Educational Technol High Educ. 2023;20(1):35. Patterns 2023, 4(1). Lyonga NAN. Peer Learning amongst Students of Higher Technical Teachers' Training College (HTTTC) of the University of Buea in Kumba, Cameroon. Int J High Educ. 2018;7(2):216–26. Emami A. 12 Natural Language Processing: Current Methods and Challenges. Engineering Mathematics and Artificial Intelligence: Foundations, Methods, and Applications 2023:261. Brown T, Mann B, Ryder N, Subbiah M, Kaplan JD, Dhariwal P, Neelakantan A, Shyam P, Sastry G, Askell A. Language models are few-shot learners. Adv Neural Inf Process Syst. 2020;33:1877–901. Kanwit M, Solon M. Communicative Competence in a Second Language: Theory, Method, and Applications. Taylor & Francis; 2022. Foster-Cohen S. The communicative competence of young children: a modular approach. Routledge; 2014. McDonough S. Applied linguistics in language education. Routledge; 2017. Hutchins J. From first conception to first demonstration: the nascent years of machine translation, 1947–1954. a chronology. Mach Transl. 1997;12:195–252. Turing AM. Computing machinery and intelligence. Springer; 2009. Chomsky N. The Logical Structure of Linguistic Theory Plenum Press. New York 1975. Hewitt J, Thickstun J, Manning CD, Liang P. Backpack Language Models. arXiv preprint arXiv:230516765 2023. Caucheteux C, Gramfort A, King J-R. Evidence of a predictive coding hierarchy in the human brain listening to speech. Nat Hum Behav. 2023;7(3):430–41. Feder A, Keith KA, Manzoor E, Pryzant R, Sridhar D, Wood-Doughty Z, Eisenstein J, Grimmer J, Reichart R, Roberts ME. Causal inference in natural language processing: Estimation, prediction, interpretation and beyond. Trans Association Comput Linguistics. 2022;10:1138–58. Kumar S, Correa CG, Dasgupta I, Marjieh R, Hu MY, Hawkins R, Cohen JD, Narasimhan K, Griffiths T. Using natural language and program abstractions to instill human inductive biases in machines. Adv Neural Inf Process Syst. 2022;35:167–80. Toneva M, Wehbe L. Interpreting and improving natural-language processing (in machines) with natural language-processing (in the brain). Advances in neural information processing systems 2019, 32. Ravimaram S, Sathish A, Vatchala S, Rawat R, TF MR. Robust Transfer Learning Based Modelling for Accelerating the Learning of Ai in the Field of NLP. In: 2023 9th International Conference on Advanced Computing and Communication Systems (ICACCS): 2023 : IEEE; 2023: 1026–1030. Muller B, Gupta D, Fauconnier J-P, Patwardhan S, Vandyke D, Agarwal S. Languages You Know Influence Those You Learn: Impact of Language Characteristics on Multi-Lingual Text-to-Text Transfer. Transfer Learning for Natural Language Processing Workshop: 2023: PMLR; 2023: 88–102. Shen R, Pan W. Research on transfer learning technology in natural language processing. In: Artificial Intelligence in China: Proceedings of the 2nd International Conference on Artificial Intelligence in China: 2021 : Springer; 2021: 483–488. Biggers FB, Mohanty SD, Manda P. A deep semantic matching approach for identifying relevant messages for social media analysis. Sci Rep. 2023;13(1):12005. Galinsky R, Alekseev A, Nikolenko S. Improving Neural Models for Natural Language Processing in Russian with Synonyms. J Math Sci 2023:1–12. Rinaldi AM, Russo C, Tommasino C. Automatic image captioning combining natural language processing and deep neural networks. Results in Engineering. 2023;18:101107. Biswas SS. Role of chat gpt in public health. Ann Biomed Eng. 2023;51(5):868–9. Lund BD, Wang T. Chatting about ChatGPT: how may AI and GPT impact academia and libraries? Libr Hi Tech News. 2023;40(3):26–9. Surameery NMS, Shakor MY. Use chat gpt to solve programming bugs. International Journal of Information Technology & Computer Engineering (IJITC) ISSN: 2455–5290 2023, 3(01):17–22. Jiao W, Wang W, Huang J-t, Wang X, Tu Z. Is ChatGPT a good translator? A preliminary study. arXiv preprint arXiv:230108745 2023. Liu J, Wang C, Liu S. Utility of ChatGPT in clinical practice. J Med Internet Res. 2023;25:e48568. Tlili A, Shehata B, Adarkwah MA, Bozkurt A, Hickey DT, Huang R, Agyemang B. What if the devil is my guardian angel: ChatGPT as a case study of using chatbots in education. Smart Learn Environ. 2023;10(1):15. Whalen J, Mouza C. ChatGPT: Challenges, Opportunities, and Implications for Teacher Education. Contemp Issues Technol Teacher Educ. 2023;23(1):1–23. Pelton T, Pelton LF. Adapting ChatGPT to Support Teacher Education in Mathematics. In: Society for Information Technology & Teacher Education International Conference: 2023 : Association for the Advancement of Computing in Education (AACE); 2023: 1662–1670. Tirado-Olivares S, Navío-Inglés M, O’Connor-Jiménez P, Cózar-Gutiérrez R. From Human to Machine: Investigating the Effectiveness of the Conversational AI ChatGPT in Historical Thinking. Educ Sci. 2023;13(8):803. Deng J, Lin Y. The benefits and challenges of ChatGPT: An overview. Front Comput Intell Syst. 2022;2(2):81–3. Liebrenz M, Schleifer R, Buadze A, Bhugra D, Smith A. Generating scholarly content with ChatGPT: ethical challenges for medical publishing. Lancet Digit Health. 2023;5(3):e105–6. Kalla D, Smith N. Study and Analysis of Chat GPT and its Impact on Different Fields of Study. Int J Innovative Sci Res Technol 2023, 8(3). Zack T, Lehman E, Suzgun M, Rodriguez JA, Celi LA, Gichoya J, Jurafsky D, Szolovits P, Bates DW, Abdulnour R-EE. Coding Inequity: Assessing GPT-4's Potential for Perpetuating Racial and Gender Biases in Healthcare. medRxiv 2023:2023.2007. 2013.23292577.. Koehler MJ, Mishra P, Cain W. What is technological pedagogical content knowledge (TPACK)? J Educ. 2013;193(3):13–9. Gudmundsdottir S, Shulman L. Pedagogical content knowledge in social studies. Scandinavian J Educationl Res. 1987;31(2):59–70. Abubakir H, Alshaboul Y. Unravelling EFL teachers' mastery of TPACK: Technological pedagogical and content knowledge in writing classes. Heliyon; 2023. Harris JB, Phillips M, Koehler MJ, Rosenberg JM. Editorial 33 (3): TPCK/TPACK research and development: Past, present, and future directions. Australasian J Educational Technol 2017, 33(3). Crawford JR, Garthwaite PH, Porter S. Point and interval estimates of effect sizes for the case-controls design in neuropsychology: rationale, methods, implementations, and proposed reporting standards. Cognit Neuropsychol. 2010;27(3):245–60. Crawford JR, Garthwaite PH. Single-case research in neuropsychology: a comparison of five forms of t-test for comparing a case to controls. Cortex. 2012;48(8):1009–16. Ferreira-Correia A, Anderson DG, Cockcroft K, Krause A. The neuropsychological deficits and dissociations in Huntington Disease-Like 2: A series of case-control studies. Neuropsychologia. 2020;136:107238. Crawford JR, Garthwaite PH. Comparison of a single case to a control or normative sample in neuropsychology: Development of a Bayesian approach. Cognit Neuropsychol. 2007;24(4):343–72. Lincoln Y. The SAGE encyclopedia of social science research methods. 2008. Drummond A, Sweeney T. Can an objective measure of technological pedagogical content knowledge (TPACK) supplement existing TPACK measures? Br J Edu Technol. 2017;48(4):928–39. Valtonen T, Sointu E, Kukkonen J, Kontkanen S, Lambert MC, Mäkitalo-Siegl K. TPACK updated to measure pre-service teachers’ twenty-first century skills. Australasian J Educational Technol 2017, 33(3). Kartal T, Kartal B, Uluay G. Technological pedagogical content knowledge self-assessment scale (TPACK-SAS) for pre-service teachers: Development, validity and reliability. Int J Eurasia Social Sci. 2016;7(23):1–36. Hsu C-Y, Tsai M-J, Chang Y-H, Liang J-C. Surveying in-service teachers’ beliefs about game-based learning and perceptions of technological pedagogical and content knowledge of games. J Educational Technol Soc. 2017;20(1):134–43. Yurdakul IK, Odabasi HF, Kilicer K, Coklar AN, Birinci G, Kurt AA. The development, validity and reliability of TPACK-deep: A technological pedagogical content knowledge scale. Comput Educ. 2012;58(3):964–77. Carter N, Bryant-Lukosius D, DiCenso A, Blythe J, Neville AJ. The use of triangulation in qualitative research. In: Oncol Nurs Forum : 2014; 2014: 545–547. Macmillan NA, Creelman CD. Detection theory: A user's guide. Psychology press; 2004. Drummond A. TPACK Objective Dataset. In.; 2023. https://osf.io/3rhxu/ . Gabbiani F, Cox SJ. Mathematics for neuroscientists: Academic Press; 2017. Green DM, Swets JA. Signal detection theory and psychophysics. Volume 1. Wiley New York; 1966. Braun V, Clarke V. Thematic analysis: American Psychological Association; 2012. Liu R, Jia C, Zhang G, Zhuang Z, Liu T, Vosoughi S. Second thoughts are best: Learning to re-align with human values from text edits. Adv Neural Inf Process Syst. 2022;35:181–96. Hendrycks D, Burns C, Basart S, Critch A, Li J, Song D, Steinhardt J. Aligning ai with shared human values. arXiv preprint arXiv:200802275 2020. Lu P, Mishra S, Xia T, Qiu L, Chang K-W, Zhu S-C, Tafjord O, Clark P, Kalyan A. Learn to explain: Multimodal reasoning via thought chains for science question answering. Adv Neural Inf Process Syst. 2022;35:2507–21. Roy NA, Kim J, Rabinowitz N. Explainability Via Causal Self-Talk. Adv Neural Inf Process Syst. 2022;35:7655–70. Wei J, Bosma M, Zhao VY, Guu K, Yu AW, Lester B, Du N, Dai AM, Le QV. Finetuned language models are zero-shot learners. arXiv preprint arXiv:210901652 2021. Katz DM, Bommarito MJ, Gao S, Arredondo P. Gpt-4 passes the bar exam. Available at SSRN 4389233 2023. Li J, Ouyang J, Liu J, Zhang F, Wang Z, Guo X, Liu M, Taylor D. Artificial intelligence-based online platform assists blood cell morphology learning: A mixed-methods sequential explanatory designed research. Med Teach. 2023;45(6):596–603. Philbin CA. Exploring the Potential of Artificial Intelligence Program Generators in Computer Programming Education for Students. ACM Inroads. 2023;14(3):30–8. Guilherme A. Considering AI in education: Erziehung but never Bildung. Artif Intell Incl Education: Specul Futures Emerg Practices 2019:165–78. Gill SS, Xu M, Patros P, Wu H, Kaur R, Kaur K, Fuller S, Singh M, Arora P, Parlikad AK. Transformative effects of ChatGPT on modern education: Emerging Era of AI Chatbots. Internet of Things and Cyber-Physical Systems. 2024;4:19–23. Sobo E. Could ChatGPT Prompt a New Golden Age in Higher Education? Teach Learn Anthropol 2023, 6(1). Chan CKY, Hu W. Students' Voices on Generative AI: Perceptions, Benefits, and Challenges in Higher Education. arXiv preprint arXiv:230500290 2023. Declarations Ethical approval and consent to participate: The authors confirm that all the methods comply with current guidelines and regulations that follow the Declaration of Helsinki. TPACK-objective dataset (a publicly available dataset), which was used in this study as normative data, had been collected with approval from the Social and Behavioural Research Ethics Committee at Flinders University, where participation was voluntary and anonymous, and informed consent had been obtained from all participants. Consent for publication: Not applicable. Competing interests: The authors of this study declare that they have no competing interests. Data Availability: The data collected in this study will be deposited on the Open Science Framework (OSF) and will be publicly accessible. Researchers interested in accessing the GPT dataset can find it on OSF at https://osf.io/xtce4/ upon publication. The TPACK-objective dataset is available at https://osf.io/3rhxu/ Funding: Not applicable. Contributions: Ferdi Çelik, Ceylan Yangın Ersanlı, and Aaron Drummond planned, wrote, edited, reviewed, and agreed on the manuscript. Acknowledgements: Not applicable. Additional Declarations No competing interests reported. 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-3388153","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":236915196,"identity":"c5547784-0e3b-4262-9ccf-ab77eeeb50a1","order_by":0,"name":"Ferdi Çelik","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA/ElEQVRIiWNgGAWjYBAC9gYGBokEBgYZNh7mgw8+AEXY2Alo4TkA0cLDxsOWbDgDpIWZGC0gmoGHx0yaByREUAt778MbD2psePh4zphJ2/zaJs/HzMD44WMOHi08x40tEo6l8bDxthVb5/bdNmxjZmCWnLkNtxZ7iTQ2icSGwzxs/Mwbb+f23GYEamFj5sWjhUf+GUjLf6AWBgNpy57b9oS1SLCBtBwAOqzFSJrhx+1Ewlp40piBfkkGBvKxZMPehtvJbcyMzXj9wsN+jPHmjxo7Ofme5IMPfvy5bTu/vfngh494tKACxjYw2UCsehD4Q4riUTAKRsEoGCkAAL1GRr1KqfeBAAAAAElFTkSuQmCC","orcid":"","institution":"Ondokuz Mayıs University","correspondingAuthor":true,"prefix":"","firstName":"Ferdi","middleName":"","lastName":"Çelik","suffix":""},{"id":236915197,"identity":"01c2bb3d-dd9e-4ebe-9187-f27f0b3f5caa","order_by":1,"name":"Ceylan Yangın Ersanlı","email":"","orcid":"","institution":"Ondokuz Mayıs University","correspondingAuthor":false,"prefix":"","firstName":"Ceylan","middleName":"Yangın","lastName":"Ersanlı","suffix":""},{"id":236915198,"identity":"c53c03bf-2619-4987-92a0-19a2045bbc45","order_by":2,"name":"Aaron Drummond","email":"","orcid":"","institution":"University of Tasmania","correspondingAuthor":false,"prefix":"","firstName":"Aaron","middleName":"","lastName":"Drummond","suffix":""}],"badges":[],"createdAt":"2023-09-26 09:59:30","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-3388153/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-3388153/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":43981518,"identity":"44df2439-3b7d-4fb5-b806-8b9ca4c8093d","added_by":"auto","created_at":"2023-10-02 17:42:33","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":54184,"visible":true,"origin":"","legend":"\u003cp\u003ePrior and posterior effect sizes for the difference between ChatGPT and the control sample.\u003c/p\u003e\n\u003cp\u003eNote: effect sizes are negative because the graph indexes the control samples' performance relative to ChatGPT's performance (i.e., GPT's performance is higher than the control sample).\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-3388153/v1/6a5cb1b1172c2e9ee870f08a.png"},{"id":43981519,"identity":"c9f3116c-79a9-44b2-a1e3-2ad0083bd842","added_by":"auto","created_at":"2023-10-02 17:42:33","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":74118,"visible":true,"origin":"","legend":"\u003cp\u003eBayes Factors for the difference between ChatGPT and the normative sample across a range of Cauchy Prior widths.\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-3388153/v1/9b319c1e19000bbb3df193cb.png"},{"id":43982660,"identity":"0fe8c204-732a-4e90-9e32-84430ac26c0a","added_by":"auto","created_at":"2023-10-02 17:50:33","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":66700,"visible":true,"origin":"","legend":"\u003cp\u003eAccumulation of evidence for a difference between ChatGPT and the control sample in a sequential analysis (i.e., the same analysis performed on an increasing subset of the data until all data are included).\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-3388153/v1/6b040bbbbf8b5ab813315b6e.png"},{"id":44461346,"identity":"4058ae30-d121-4ab3-8ee3-79b21939ebb3","added_by":"auto","created_at":"2023-10-11 18:44:07","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":501354,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-3388153/v1/a9c9e268-74b4-4cdb-94ae-1de6de07eecb.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"GPT-PACK: Can GenAI act as TPACK Peer Tutor for Preservice Teachers? A Comparative Study on Objective TPACK of Humans and ChatGPT 3.5","fulltext":[{"header":"INTRODUCTION","content":"\u003cp\u003eThe integration of emerging technologies has become essential for teachers as it enhances student engagement [\u003cspan additionalcitationids=\"CR2\" citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e] facilitates personalized learning [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e], fosters digital literacy [\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e], and prepares students for the demands of the 21st century [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e]. Given the increasing integration of digital technologies into almost every aspect of daily life, teachers need to possess in-depth Technological Pedagogical Content Knowledge (TPACK) [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e, \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e], as it enables them to create meaningful and engaging learning experiences that cater to students' diverse needs and prepare them for success in a technology-driven world.\u003c/p\u003e \u003cp\u003eContinuous support throughout their training process within Faculties of Education may yield better outcomes for preservice teachers by equipping them with the necessary skills and knowledge to teach effectively [\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e]. However, due to the growing student number, limited staffing resources, the demanding workload of academics, and the multi-faceted nature of their responsibilities, it is impractical for academics always to offer availability to the preservice teachers, making peer tutoring a desired option. Technology is evolving so rapidly that many preservice teachers may feel left alone and try to find self-directed ways to improve their TPACK [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e](Max et al., 2023). Indeed, many online resources, such as teaching and technology websites, offer a large amount of data. Nevertheless, they lack a friendly conversation to give a hand whenever the preservice teachers need active guidance.\u003c/p\u003e \u003cp\u003eOn the other hand, generative AI (GenAI) [\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e], empowered by the impressive advancements in natural language processing (NLP) [\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e], has become an accessible and transformative tool for educators. While GenAI holds hope-inspiring potential for shaping how we teach and learn, its specific educational applications remain largely unexplored. This study proposes ChatGPT, an NLP-powered GenAI chatbot, as a possible peer tutor for preservice teachers that can offer them 24/7 support at their own pace by promoting self-directed learning, as preservice teachers can access it at any time using a smartphone, computer, or tablet connected to the internet.\u003c/p\u003e \u003cp\u003eGrounded in the social constructivist school of thought, the Zone of Proximal Development (ZPD) Theory, developed by Lev Vygotsky [\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e], proposes that humans can acquire knowledge and skills through interactions with individuals who possess higher expertise, often referred to as more knowledgeable others [\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]. We believe that GenAI, such as ChatGPT, may act as a supportive partner for preservice teachers, capable of providing constructive feedback [\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e], answering preservice teachers' questions, and having them engage in meaningful discussions [\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e] related to teaching practices and pedagogical strategies due to its advanced NLP capabilities. According to the principles of peer tutoring, the ideal peer tutor is typically more knowledgeable or experienced than the learner. This dynamic allows for the transfer of knowledge from the more knowledgeable peer to the learner within their ZPD. It fosters effective learning through scaffolded support and guidance provided by the more knowledgeable peer [\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e]. Therefore, to consider ChatGPT as a possible peer tutor for preservice teachers to enhance their TPACK, a comprehensive understanding of ChatGPT's objective TPACK performance is crucial compared to human preservice teachers. To this end, the primary aim of this study is to conduct a comprehensive evaluation, juxtaposing ChatGPT against human preservice teachers' Technological Pedagogical Content Knowledge (TPACK).\u003c/p\u003e"},{"header":"LITERATURE REVIEW","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003eNatural Language Processing\u003c/h2\u003e \u003cp\u003eIn recent years, text-based GenAI has experienced considerable development, and NLP is a key contributing factor [\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e, \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e]. It might be tempting to question whether NLP is a technology or a scientific field dealing with how humans naturally process a language. In its essence, it combines advanced computational techniques to replicate not just the surface-level mechanics of human language and use but the very essence of our communicative competence [\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e, \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e] and performance [\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e]. Therefore, NLP, in its current sense, is a technology utilizing advanced techniques to learn from the data for achieving linguistic tasks, and therefore, contributing to the development of GenAI models that are capable of comprehending, interpreting, and generating language in a manner that engages in human-like communication for various objectives. It might be worth noting that the quality of the model is contingent upon the quality of the training data, though.\u003c/p\u003e \u003cp\u003eThe roots of NLP trace back to scholarly discussions in the 1950s, driven by the need for automated translation after World War II [\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e]. Earliest well-known debates include the argument of Turing's \u0026ldquo;Can Machines Think?\u0026rdquo; in 1950 [\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e] (p. 433) and Chomsky's critique in 1955, exemplified by the ambiguous but grammatically valid sentence \u0026ldquo;Colorless green ideas sleep furiously\u0026rdquo; [\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e] (p. 242), highlighting the limitations of early NLP approaches.\u003c/p\u003e \u003cp\u003eNLP has made significant progress through the collaborative effort of scholars in the fields of computer science, linguistics, psychology, and neuroscience [\u003cspan additionalcitationids=\"CR27 CR28 CR29\" citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e]. Through several developmental stages, NLP has currently reached a point where transfer learning [\u003cspan additionalcitationids=\"CR32\" citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e], the ability of AI to transfer the pre-trained model's learned features and adapt them to the new task using a smaller dataset, and deep learning, the ability of AI to automatically learn from the data using deep neural networks [\u003cspan additionalcitationids=\"CR35\" citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e] have emerged as key technologies. This, in turn, equips GenAI models, such as ChatGPT, with increasingly human-like communication capabilities, making them potentially more viable peers for conversational interactions on TPACK.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec4\" class=\"Section2\"\u003e \u003ch2\u003eChatGPT\u003c/h2\u003e \u003cp\u003eChatGPT is a text-based GenAI developed by OpenAI. While OpenAI offers a subscription plan for users to engage with GPT 4.0 (highlighted by its capability) for a subscription fee, GPT 3.5 (highlighted by its processing speed) is more commonly used worldwide because a free research preview is available online. New Android and IOS applications have been released in less than a year, making GPT more accessible. At least a hundred million users are interacting with ChatGPT, and OpenAI reached a \u003cspan\u003e$\u003c/span\u003e30\u0026nbsp;billion valuation after the release of ChatGPT. This is because it may be regarded as one of the first open-access GenAI models to offer highly advanced NLP capabilities that provide human-like communication, backed by its massive knowledge base acquired during its training process.\u003c/p\u003e \u003cp\u003eChatGPT has been proposed as a tool in several fields, including but not limited to public health [\u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e], improving library services [\u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e](Lund \u0026amp; Wang, 2023), fixing software bugs [\u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e], translation [\u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e40\u003c/span\u003e], clinical practice [\u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e41\u003c/span\u003e]. Specifically, in the field of learning and teaching [\u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e42\u003c/span\u003e] (Tlili et al., 2023), ChatGPT has been interacted with for self-directed learning for its role as a more knowledgeable other [\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e], teacher training[\u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e43\u003c/span\u003e, \u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e44\u003c/span\u003e], and improving preservice teachers' historical thinking skills [\u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e45\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eDespite its great potential, it has limitations such as overreliance on training data [\u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e46\u003c/span\u003e], which is time-bound (currently up to 2021), and inability to access external resources. Additionally, several issues have been raised related to ethics [\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e, \u003cspan citationid=\"CR47\" class=\"CitationRef\"\u003e47\u003c/span\u003e], safety [\u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e42\u003c/span\u003e], misleading information generation [\u003cspan citationid=\"CR48\" class=\"CitationRef\"\u003e48\u003c/span\u003e], and the potential for racial and gender biases inherent in the models and the content they produce [\u003cspan citationid=\"CR49\" class=\"CitationRef\"\u003e49\u003c/span\u003e].\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec5\" class=\"Section2\"\u003e \u003ch2\u003eTechnological Pedagogical Content Knowledge\u003c/h2\u003e \u003cp\u003eTechnological Pedagogical and Content Knowledge, abbreviated as TPACK, is a comprehensive framework to indicate the knowledge of teachers in using technology in education [\u003cspan citationid=\"CR50\" class=\"CitationRef\"\u003e50\u003c/span\u003e], which was built after the pedagogical content knowledge framework [\u003cspan citationid=\"CR51\" class=\"CitationRef\"\u003e51\u003c/span\u003e] with the domination of technology in our everyday life. In the past, pedagogical and content knowledge (PCK) has been seen as the two main domains that make up the educational environment. It encompasses the particular area of professional expertise that educators possess, enabling them to successfully teach students content knowledge through a variety of pedagogical techniques. TPACK is a concept that has emerged as a result of Schulman's framework's evolution in the context of modern education to include a new technology dimension. This paradigm shift emphasizes how important technology-supported educational approaches are becoming in contemporary pedagogical practices. TPACK is a model that underscores the relationship between the content, pedagogical, and technological knowledge of teachers and is required for the effective integration of technology into teaching [\u003cspan citationid=\"CR52\" class=\"CitationRef\"\u003e52\u003c/span\u003e]. According to Harris [\u003cspan citationid=\"CR53\" class=\"CitationRef\"\u003e53\u003c/span\u003e], TPACK represents the intertwined and context-specific knowledge essential for proficiently incorporating digital tools and resources into curriculum-driven instruction.\u003c/p\u003e \u003cp\u003eThe framework involves three fundamental, interrelated constructs;\u003c/p\u003e \u003cp\u003e \u003cem\u003ePedagogical Content Knowledge\u003c/em\u003e refers to the integration of subject matter proficiency and the mastery of pedagogical skills. It involves the effective synthesis of content expertise and instructional methods.\u003c/p\u003e \u003cp\u003e \u003cem\u003eTechnological Content Knowledge\u003c/em\u003e focuses on how the utilization of technology influences the ways content is presented. This involves recognizing how technology alters content representation and learning dynamics.\u003c/p\u003e \u003cp\u003e \u003cem\u003eTechnological Pedagogical Knowledge\u003c/em\u003e encompasses a teacher's awareness of the array of technological tools available and their capacity to effectively select and employ these technologies to enhance the desired learning outcomes. It entails the strategic integration of technology within pedagogical contexts to optimize educational experiences.\u003c/p\u003e \u003cp\u003eThe TPACK theoretical framework is frequently employed for the analysis of educators' integrated knowledge encompassing technology, pedagogy, and content. This approach makes it easier to examine how teachers use technology to present, organize, and change subject information in a way that improves students' understanding of it.\u003c/p\u003e \u003c/div\u003e"},{"header":"METHOD","content":"\u003cdiv id=\"Sec7\" class=\"Section2\"\u003e \u003ch2\u003eResearch Design\u003c/h2\u003e \u003cp\u003eA single case-control design was chosen for this study to conduct a robust comparison of the objective TPACK performance of ChatGPT to that of human preservice teachers [\u003cspan citationid=\"CR54\" class=\"CitationRef\"\u003e54\u003c/span\u003e]. A single case-control design is a research approach that involves the comparison of a single case (in the present study, GPT 3.5) with a control group (human preservice teachers) to draw meaningful conclusions about the case's performance [\u003cspan citationid=\"CR55\" class=\"CitationRef\"\u003e55\u003c/span\u003e]. Thus, it was a good fit for the present study. As this design provided a structured framework for examining the performance of a single case compared to a normative or control sample through Bayesian statistics [\u003cspan citationid=\"CR54\" class=\"CitationRef\"\u003e54\u003c/span\u003e, \u003cspan citationid=\"CR56\" class=\"CitationRef\"\u003e56\u003c/span\u003e, \u003cspan citationid=\"CR57\" class=\"CitationRef\"\u003e57\u003c/span\u003e], it contributed to a well-rounded understanding of GPT 3.5's potential as a \u003cem\u003emore knowledgeable other\u003c/em\u003e.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003eSampling\u003c/h2\u003e \u003cp\u003eThis study employed a purposive sampling method [\u003cspan citationid=\"CR58\" class=\"CitationRef\"\u003e58\u003c/span\u003e]. We chose GPT 3.5 as a possible GenAI peer tutor due to its advanced NLP capabilities, accessibility, language processing speed, user-friendly interface, sophisticated knowledge base, and free-to-use feature. We also utilized the objective TPACK performance of 93 Australian preservice teachers (32 males) [\u003cspan citationid=\"CR59\" class=\"CitationRef\"\u003e59\u003c/span\u003e] as normative data considering the aim of the present study.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec9\" class=\"Section2\"\u003e \u003ch2\u003eInstruments\u003c/h2\u003e \u003cp\u003eThere were a variety of scales measuring the TPACK levels of the preservice teachers, such as TPACK-deep, TPACK-21 scale [\u003cspan citationid=\"CR60\" class=\"CitationRef\"\u003e60\u003c/span\u003e], TPACK SAS [\u003cspan citationid=\"CR61\" class=\"CitationRef\"\u003e61\u003c/span\u003e], and TPACK-G [\u003cspan citationid=\"CR62\" class=\"CitationRef\"\u003e62\u003c/span\u003e]. However, these measures heavily relied on self-reports provided by preservice teachers. Using the data collected via these scales could introduce issues related to social desirability bias, in which preservice teachers might have provided responses they found socially acceptable instead of reflecting their true understanding or skills. Additionally, asking GPT to answer self-report items might not be desirably reliable.\u003c/p\u003e \u003cp\u003eAfter a rigorous literature review and based on our evaluation as scholars, we decided to use the objective TPACK test. This was also because the correlation between how preservice teachers subjectively view their proficiency in TPACK (TPACK-deep) [\u003cspan citationid=\"CR63\" class=\"CitationRef\"\u003e63\u003c/span\u003e] and the concrete retention of that knowledge (objective-TPACK) was relatively weak [\u003cspan citationid=\"CR59\" class=\"CitationRef\"\u003e59\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eIn addition to collecting GPT 3.5's true and false responses to the objective-TPACK test, we had GPT 3.5 provide explanations for each response to comprehensively understand its TPACK performance by analyzing this qualitative data. This would also enhance the credibility and validity of findings by triangulating the data [\u003cspan citationid=\"CR64\" class=\"CitationRef\"\u003e64\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eTo sum up, in the present study, we collected both quantitative and qualitative data. We collected the quantitative data through the objective-TPACK scale and qualitative data by triggering GPT's explanations for each item in the scale.\u003c/p\u003e \u003cp\u003e[\u003cspan citationid=\"CR65\" class=\"CitationRef\"\u003e65\u003c/span\u003e], which reveals whether the correct responses are given by chance or based on knowledge.\u003c/p\u003e \u003cp\u003e As the case that would take the objective-TPACK test was a GenAI, we needed to investigate all the items in the scale carefully and adapt them as necessary. A few changes had been made: string \u0026ldquo;(following you)\u0026rdquo; was added after the word \u0026ldquo;students\u0026rdquo; in the 8th item, and the string \u0026ldquo;to be taught to students\u0026rdquo; was added at the end of the 11th item for clarification after consulting with the developers of the objective-TPACK test. As the original scale was developed in 2013 and published in 2016, we had to scrutinize whether the items were still valid regarding the TPACK performance of participants. We concluded that the items were valid and that there was no need for further changes.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec10\" class=\"Section2\"\u003e \u003ch2\u003eTPACK Objective Dataset\u003c/h2\u003e \u003cp\u003eAs a base performance of human preservice teachers, we analyzed the TPACK Objective dataset [\u003cspan citationid=\"CR66\" class=\"CitationRef\"\u003e66\u003c/span\u003e], a publicly available dataset containing responses of 93 Australian preservice teachers (mean age 22.7) to the objective-TPACK scale. The original dataset included dedicated columns for \u003cem\u003ed'\u003c/em\u003e prime and \u003cem\u003ec\u003c/em\u003e index scores. We recalculated the scores in the dataset to verify the scores and ensure the computing steps to apply the consistent analysis to analyze the TPACK performance of GPT 3.5 and compare it.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003eProcedure\u003c/h2\u003e \u003cp\u003eAfter having the necessary permissions to use the scale and TPACK Objective dataset, an account was created on the official OpenAI website to access ChatGPT (version 3.5). We used \u0026ldquo;Can you be an English language teacher?\u0026rdquo; as the first prompt. Then, we used the \u0026ldquo;Say true or false: The design of interactive teaching materials supports different learning theories\u0026rdquo; prompt, which contained the first item in the Objective TPACK scale. As the first prompt triggered GPT to generate explanations, we did not need to ask again. As for the following prompts, we entered the test items respectively. The outputs of GPT 3.5 were exported and saved in a text document for analysis.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003eData Analysis\u003c/h2\u003e \u003cp\u003eIn order to assess GPT 3.5's TPACK performance compared to human preservice teachers, we analyzed the quantitative data acquired through objective TPACK scale. The quantitative data was initially analyzed based on the signal detection theory, a mathematical theory commonly used by neuroscientists [\u003cspan citationid=\"CR67\" class=\"CitationRef\"\u003e67\u003c/span\u003e, \u003cspan citationid=\"CR68\" class=\"CitationRef\"\u003e68\u003c/span\u003e]. After entering the GPT 3.5's answers to an MS Excel file, we summed its correct answers to true items (hits) and incorrect answers to false items (false alarms). A flattening constant was applied to prevent failing the calculation of \u003cem\u003ed'\u003c/em\u003e and \u003cem\u003ec\u003c/em\u003e scores by subtracting 0.5 from the perfect scores and adding 0.5 to the scores of 0 while they were still the lowest/highest possible score. Next, we computed the hit rate by dividing the number of corrected hits by the total count of true items, and we calculated the false alarm rate by dividing the number of corrected false alarms by the total number of false items. Then, both the false alarm rate and hit rate needed to be transformed to a normalized \u003cem\u003eZ\u003c/em\u003e value, and \u003cem\u003ed'\u003c/em\u003e score is calculated by subtracting the normalized hit rate from the normalized false alarm rate.\u003c/p\u003e \u003cp\u003eAfter acquiring the TPACK \u003cem\u003ed'\u003c/em\u003e and \u003cem\u003ec\u003c/em\u003e scores of GPT 3.5, we compared it to that of human preservice teachers by using the Bayesian analysis method of [\u003cspan citationid=\"CR57\" class=\"CitationRef\"\u003e57\u003c/span\u003e], which was developed to compare a single case's performance with that of a normative or control sample. This was because Bayesian analysis would provide a more intuitive way to handle small sample sizes, which was often the case when assessing AI systems. Common statistical computing software used in social sciences (e.g. IBM's SPSS v26) did not have a built-in feature for this analysis. Therefore, we used \u003cem\u003eSingleBayes_ES\u003c/em\u003e (a dedicated PC software built based on the original manuscript [\u003cspan citationid=\"CR54\" class=\"CitationRef\"\u003e54\u003c/span\u003e, \u003cspan citationid=\"CR57\" class=\"CitationRef\"\u003e57\u003c/span\u003e]. Using Bayesian methods produced probability distributions, which enabled us to quantify uncertainty and variability in our conclusions rather than solely relying on point estimates. This ensured that our findings were not only statistically robust but also reflective of the inherent uncertainties in our data. Lastly, for the qualitative data, a thematic analysis was conducted [\u003cspan citationid=\"CR69\" class=\"CitationRef\"\u003e69\u003c/span\u003e].\u003c/p\u003e \u003c/div\u003e"},{"header":"RESULTS","content":"\u003cp\u003eThe present study aimed to evaluate GPT 3.5\u0026apos;s TPACK performance compared to 93 human preservice teachers to propose GPT as a peer tutor. A meticulous analysis compared human preservice teachers and the possible AI-driven peer tutor, ChatGPT.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eQuantitative Data\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe normative sample, comprising 93 human preservice teachers, exhibited a mean TPACK score of 0.85 and a standard deviation of 0.57. These parameters provided a contextual foundation for the subsequent comparisons. Upon calculating ChatGPT\u0026apos;s \u003cem\u003ed\u0026apos;\u0026nbsp;\u003c/em\u003escore, we have,\u003cem\u003e\u0026nbsp;\u003c/em\u003eintriguingly, seen that ChatGPT achieved a remarkable TPACK performance score of 2.68 out of a possible 3.07. To determine whether this result was higher than the average performance of the normative sample, we used Bayesian one-sample t-tests employing default priors to compare GPT 3.5\u0026apos;s scores with the control sample. This analysis yielded definitive evidence that ChatGPT\u0026apos;s score (2.68) was higher than the average performance of the normative sample (\u003cem\u003eM\u003c/em\u003e = 0.85, \u003cem\u003eSD\u003c/em\u003e = 0.57), BF\u003csub\u003e10\u003c/sub\u003e = 4.41 x 10\u003csup\u003e46\u003c/sup\u003e, \u0026delta; = 3.11 95% CI [2.747, 3.525]. The Bayes factor suggests that it is 4.41 x 10\u003csup\u003e46\u003c/sup\u003e times more likely that there is a difference between ChatGPT\u0026apos;s performance and the performance of the normative sample with a considerable effect size. Figure 1 shows the prior and posterior effect sizes for the difference between ChatGPT and the control sample. We utilized the Bayesian method of Crawford-Garthwaite [57] to scrutinize the proportion of control sample scores that would fall below ChatGPT 3.5\u0026apos;s performance and control sample scores using SingleBayes ES software. This analysis suggested that approximately 99.9037% of the control population would obtain a score lower than ChatGPT.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eFigure 2 shows robustness checks for the difference between ChatGPT and the control sample based on different Cauchy prior widths. The priors adopted within a Bayesian analysis can impact whether the evidence for an effect is considered strong enough to increase our confidence in our priors \u0026ndash; for instance, if we expect a large effect, default priors might overestimate the amount of evidence a given analysis yields because the default priors are smaller than what was expected to be observed. Our analysis used uninformed default priors with a Cauchy prior width of 0.707. As shown in Figure 2, the analysis is robust to a wide range of priors, with evidence remaining definitive that ChatGPT outperformed the normative sample irrespective of the priors chosen for this analysis. \u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cbr\u003e\u003c/p\u003e\n\u003cp\u003eFigure 3 shows a graph of how evidence accumulates across a sequential analysis of the data. Although Bayesian analyses are more robust to outliers and small \u003cem\u003en\u003c/em\u003e samples, outliers, and extreme scores can sometimes impact the results as they do in a frequentist analysis. Figure 3 shows that as more cases are included in the analysis, a steady accumulation of evidence occurs, suggesting that no particular score or scores greatly influence the results. This analysis increases our confidence that the findings accurately represent the data.\u003c/p\u003e\n\u003cp\u003e\u003cbr\u003e\u003c/p\u003e\n\u003cp\u003eIdeally, a perfect responder will have a \u003cem\u003ec\u003c/em\u003e score of 0, indicating no bias in their response. Positive \u003cem\u003ec\u0026nbsp;\u003c/em\u003escores indicate a tendency to report false answers to the questions irrespective of the correct response. In contrast, negative \u003cem\u003ec\u003c/em\u003e scores indicate a tendency to report true to answers irrespective of the correct response. We analyzed GPT 3.5\u0026apos;s responses using the \u003cem\u003ec\u0026nbsp;\u003c/em\u003eindex; the analysis indicated that the model provided relatively well-calibrated answers based on its existing knowledge base rather than random guessing when answering questions (c = -0.19). Although ChatGPT appeared to be slightly more likely to report true than the control sample (which did not differ from 0; \u003cem\u003eM\u003c/em\u003e = -0.037, \u003cem\u003eSD\u003c/em\u003e = 0.399), the effect size was relatively small, BF\u003csub\u003e10\u003c/sub\u003e = 55.2, \u0026delta; = 0.369 95% CI [0.160, 0.578], and the absolute deviation from zero was slight. This suggests that GPT 3.5 is well calibrated to the objective TPACK task.\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eOverall, these results shed light on the model\u0026apos;s ability to draw from its training to generate responses, emphasizing its capacity to provide contextually relevant information regarding TPACK. However, it was essential to consider other evaluation measures to gain a comprehensive understanding of the model\u0026apos;s performance and knowledge utilization. Therefore, we analyzed the qualitative data from GPT 3.5 in the following subsection.\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eQualitative Data\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cem\u003e\u0026nbsp;\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eDesign\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eThe design is a crucial aspect of TPACK within education to create effective learning experiences for the learners. Findings indicated that, in the provided GPT explanations, the design aspect was comprehensively explored. The responses highlighted that interactive teaching materials could align with learning theories and cater to diverse styles. The responses indicated that GPT critically assessed that using the latest technology would not guarantee improved outcomes; it was about thoughtful integration and alignment with content and pedagogy. Additionally, the importance of identifying students\u0026apos; needs was underscored by GPT, which it thought was for enabling tailored technology use. The analysis demonstrated GPT\u0026apos;s grasp of how design involves aligning technological tools with pedagogical approaches and content to enhance learning:\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eFor example, behaviorist approaches may use interactive materials that offer immediate feedback and reinforcement, while constructivist approaches may use materials that encourage exploration and critical thinking. Ultimately, the goal is to engage learners and enhance their understanding and retention of the subject matter. (Response to #1 item)\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003e\u003cem\u003e\u0026nbsp;\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eGPT 3.5 acknowledged the potential benefits of technology while cautioning that improved learning outcomes were not guaranteed. GPT suitably emphasized the importance of effective integration, quality content, and consideration of students\u0026apos; readiness and access. It depicted technology as a tool for enhancing learning rather than a guaranteed solution.\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eTechnology should be seen as a tool to enhance and support learning rather than a magic solution that automatically leads to improved learning. It\u0026apos;s essential to combine technology with effective pedagogical practices and tailor it to meet the specific needs and goals of the students and the learning objectives. (Response to #2 item)\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003e\u003cem\u003e\u0026nbsp;\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eWhen we asked GPT the item four, \u0026ldquo;A needs analysis of technologies to be used in the teaching and learning process involves the identification of technological skills that students could benefit from\u0026rdquo;, GPT 3.5 mistakenly asserted that a needs analysis involved identifying the technological skills that students could benefit from. However, this claim, as written in the initial survey, appeared to be more accurate in the absence of specific aims or requirements necessitating particular technological competencies. While the response correctly described the systematic nature of a needs analysis and its role in aligning technology with student requirements, it overlooked the crucial point that technological skill identification was not a default component of \u003cem\u003eevery\u003c/em\u003e needs analysis. When we checked the raw data, we saw that it was GPT\u0026apos;s only incorrect answer in the objective TPACK test:\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eYes, you are absolutely right. A needs analysis of technologies to be used in the teaching and learning process involves identifying the technological skills that students could benefit from. (...) In the context of educational technology, a needs analysis may include the following steps: (...) By conducting a comprehensive needs analysis, educators can make informed decisions about integrating appropriate technologies into the teaching and learning process, leading to a more effective and engaging educational experience for students.\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003e\u003cem\u003e\u0026nbsp;\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eOverall, the analysis revealed that GPT effectively addressed the design dimension of TPACK. While one response contained a misleading assertion, the overall insights offered a sound understanding of the design dimension, underscored by the importance of informed technology selection based on a holistic understanding of educational needs.\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eExertion\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eRegarding the exertion theme, we saw that the provided explanations offered insights into how technology could enhance student participation and success. While acknowledging the motivational potential of technology, GPT\u0026apos;s explanations also highlighted the importance of balanced and thoughtful integration. It recognized that technology alone was not a guaranteed motivator and emphasized the need for alignment with learning objectives and instructional strategies.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eWhile technology can be a powerful motivational tool, it\u0026apos;s essential to use it thoughtfully and in conjunction with effective pedagogical practices. The success of technology in motivating students depends on how well it aligns with the learning objectives, the quality of the content, and the support provided by educators in integrating technology effectively into the teaching process. (Response to #5 item)\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eAdditionally, the responses acknowledged potential challenges and considerations, such as the limitations of technology-based assessments and the varying impact of social media on education. These insights reflected a comprehensive understanding of the exertion dimension within the TPACK framework, highlighting both the opportunities and potential pitfalls of utilizing technology to enhance student engagement and motivation.\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eHowever, it\u0026apos;s crucial to recognize that technology should complement traditional assessment methods and not replace them entirely. Educators must consider the limitations of technology-based assessments and use a balanced approach to evaluate students\u0026apos; success effectively. Technology can enhance assessment practices by providing efficiency, scalability, and additional insights into students\u0026apos; learning, but human judgment and qualitative assessments remain essential components of a comprehensive evaluation process. (Response to #6 item)\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003e\u003cem\u003e\u0026nbsp;\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eOverall, the analyses demonstrated an understanding of how technology could play a motivational role in education while also acknowledging the complexities involved.\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eEthics\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003e\u003cem\u003e\u0026nbsp;\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eGPT correctly emphasized that copyright issues are pertinent for teachers when designing and sharing digital resources, underlining the importance of respecting intellectual property rights. Furthermore, the analyses appropriately stressed the significance of addressing student safety and privacy concerns related to technology-based information, reflecting an awareness of ethical responsibilities in the digital age.\u003c/p\u003e\n\u003cp\u003e\u003cem\u003e\u0026nbsp;\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eCopyright is a crucial issue for teachers when designing and sharing digital resources. Just like any other content creators, teachers must be mindful of copyright laws and respect intellectual property rights. When designing and sharing digital resources, teachers need to consider the following copyright-related aspects: (Response to #9th item)\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003e\u003cem\u003e\u0026nbsp;\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eThe responses also provide accurate insights into the potential impact of technology on copyright compliance. GPT mentioned the Australian Copyright Act, but not the curriculum. Nevertheless, its awareness of legal and ethical implications demonstrated a sound understanding of the need for responsible use of digital resources.\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eThe statement is not universally true. Showing television programs to students for educational purposes while employed by a school in Australia may or may not breach copyright, depending on the specific circumstances and how the use aligns with copyright laws and exceptions. While there are provisions for fair dealing for educational purposes under the Australian Copyright Act 1968, fair dealing has limitations and conditions that must be met. (Response to #10)\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003e\u003cem\u003e\u0026nbsp;\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eThe responses regarding the 11th statement, which concerns the Australian Curriculum\u0026apos;s requirement to teach student safety and privacy of technology-based information, demonstrated a clear awareness of the ethical and educational obligations associated with technology integration. As time-bound data was usually considered a limitation of GPT to be used for educational purposes, it appropriately warned about it.\u003c/p\u003e\n\u003cp\u003e\u003cem\u003e\u0026nbsp;\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eAs of my last update in September 2021, the Australian Curriculum does emphasize the importance of addressing issues related to student safety(...) It\u0026apos;s important to note that curriculum requirements and guidelines can evolve, so I recommend checking with relevant educational authorities or resources for the most up-to-date information. (Response to #11)\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003e\u003cem\u003e\u0026nbsp;\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eWe found that GPT 3.5\u0026apos;s responses showcased a solid understanding of the ethics dimension by addressing key ethical considerations such as copyright, privacy, and responsible technology use, further supporting our initial results. The explanations of GPT demonstrated an awareness of the need for educators to navigate ethical challenges when integrating technology into the educational environment.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eTo conclude this section, the quantitative and qualitative data analyses revealed that GPT 3.5 had remarkable TPACK performance, suggesting that it can be \u003cem\u003ea more knowledgeable other\u003c/em\u003e for human preservice teachers to enhance their TPACK. Its responses to TPACK prompts highlighted how NLP could generate explanations in a variety of TPACK questions with a guiding response, which confirmed GPT 3.5 as a potential TPACK peer tutor.\u003c/p\u003e"},{"header":"LIMITATIONS","content":"\u003cp\u003eUnfortunately, there were some limitations. The normative sample consisted of only Australian preservice teachers; data from different countries could enhance the generalizability of the findings. A more advanced version of ChatGPT was available, which was GPT 4.0. It could showcase an even better performance than 3.5. However, it was a paid subscription that could discourage some preservice teachers from testing it as a TPACK peer tutor in future studies.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eMoreover, even with a huge amount of training data, GenAI models, including GPT, are not without limitations. GenAI models may not interpret or generate language the same way humans do, while exactly how humans process language is still a topic of scholarly discussion. For instance, recent models may fail to process and generate desired language in making social judgments [70, 71], verbalizing abstract things based on prior knowledge [29], critically responding to scientific questions with human reasoning [72], and explaining the reason behind their own behaviors [73]. Additionally, as with every GenAI model, its knowledge is limited to what it was trained on despite its remarkable few-shot performance [74] and advancing zero-shot performance [75].\u0026nbsp;\u003c/p\u003e"},{"header":"DISCUSSION \u0026 CONCLUSION","content":"\u003cp\u003eIn the present study, we examined the objective TPACK performance of GPT 3.5 in comparison to human preservice teachers to test the hypothesis that ChatGPT can act as a\u003cem\u003e\u0026nbsp;more knowledgeable\u003c/em\u003e \u003cem\u003eother\u0026nbsp;\u003c/em\u003e[15, 76-78]. We specifically focused on determining whether ChatGPT could take on the role of a peer tutor for TPACK. Our findings provide compelling evidence to support the hypothesis that ChatGPT can serve as a peer tutor. By revealing the performance of GPT 3.5 as a \u003cem\u003emore knowledgeable other\u003c/em\u003e in TPACK, our study suggests that GPT 3.5 can act as a peer tutor for preservice teachers to improve their TPACK.\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThrough careful analyses, we observed that GPT 3.5 demonstrated a level of TPACK surpassing that of human preservice teachers. Its ability to generate coherent and contextually relevant responses across a range of TPACK questions indicated a great knowledge of education\u0026apos;s content and technological aspects. ChatGPT could answer learners\u0026rsquo; questions and enhance collaboration\u0026nbsp;[79]. This suggests that GPT 3.5 has the potential to play a significant role in education\u0026nbsp;[80]\u0026nbsp;by aiding preservice teachers in their journey of teaching with positive learner perceptions towards GenAI in education\u0026nbsp;[81]. Based on our findings, we may claim that learners can learn by interacting with GPT 3.5, taking on the role of a more knowledgeable other.\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eGPT 3.5\u0026apos;s far-out performance in the objective TPACK showcases the advancements in AI and natural language processing\u0026nbsp;[42], where an AI model can simulate the role of a more knowledgeable peer tutor by supporting preservice teachers in TPACK. Moreover, as the present study reveals, it highlights the adaptability and versatility of GPT 3.5, which could be integrated into various educational contexts, including TPACK peer tutoring. Therefore, our findings encourage educators to explore innovative ways to integrate NLP-powered AI tools into their teaching methodologies in Faculties of Education so that preservice teachers can take advantage of a collaborative learning environment that incorporates both human and AI expertise.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThis study has proposed GPT 3.5 as a more knowledgeable TPACK peer tutor by verifying its TPACK performance. Based on this, several future research suggestions have been made. Firstly, the current study predominantly focused on the objective assessment of TPACK performance. Future qualitative research could scrutinize the subjective experiences of preservice teachers when interacting with GPT 3.5 as a TPACK peer tutor. Investigating several psychological constructs, such as learners\u0026apos; perceptions, preferences, attitudes, and enjoyment toward GenAI peer tutors, could provide insights into such interactions\u0026apos; affective and cognitive dynamics. Secondly, integrating GPT 3.5 into actual educational contexts remains a practical challenge in how and to what extent ChatGPT can be employed. Further research is needed to ascertain the effective methods of incorporating GenAI into pedagogical practices. This could involve designing a GenAI-integrated curriculum and encouraging collaboration between human educators and machine tutors.\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eAdditionally, the scalability and generalizability of GPT 3.5\u0026apos;s performance as a TPACK peer tutor across different subject domains\u0026nbsp;[79]\u0026nbsp;and age groups need exploration. Adapting the model\u0026apos;s responses to cater to diverse learning styles, psychological motives of learners, and academic disciplines could enhance its utility and impact. Furthermore, ethical considerations related to GenAI TPACK peer tutoring should be rigorously examined. Research addressing issues of data privacy, bias mitigation, and the responsible use of AI in learning contexts is required to ensure that the integration of AI in education remains safe.\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eFinally, our primary objective was to assess the objective TPACK performance of GPT 3.5, comparing it with the performance of human preservice teachers. Our research was specifically designed to assess ChatGPT\u0026apos;s suitability as a peer tutor for TPACK-related knowledge and abilities for preservice teachers. The findings of our empirical investigation provide strong and convincing support for the claim that ChatGPT may indeed function successfully as a peer tutor for preservice teachers. This information has important implications since it shows that GPT 3.5 is capable of providing a viable way to improve preservice teachers\u0026rsquo; TPACK competency and educational preparation. Although our data supports that ChatGPT can be a more knowledgeable other, it must be kept in mind that it is a machine that lacks emotions and feelings that play a significant role in human learning experience. Human learning encompasses not only the acquisition of facts and knowledge but also the complex interplay of emotions, social interactions, and personal experiences. Therefore, while ChatGPT can be used as a TPACK peer-tutor, it cannot replace the vital emotional and human elements that enrich the learning experience.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eIn conclusion, this study contributes to the evolving landscape of GenAI in education by establishing GPT 3.5\u0026apos;s competence as a TPACK peer tutor. As GenAI technology progresses, its potential to transform educational practices becomes increasingly evident. By embracing these advancements while addressing associated challenges, educators can pave the way for more personalized and effective smart learning environments.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eConsoli T, D\u0026eacute;siron J, Cattaneo A. What is technology integration and how is it measured in K-12 education? A systematic review of survey instruments from 2010 to 2021. Comput Educ 2023:104742.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e\u0026Ccedil;elik F, Yangın Ersanlı C. The use of augmented reality in a gamified CLIL lesson and students\u0026rsquo; achievements and attitudes: a quasi-experimental study. Smart Learn Environ. 2022;9(1):30.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMoraes EB, Kipper LM, Hackenhaar Kellermann AC, Austria L, Leivas P, Moraes JAR, Witczak M. Integration of Industry 4.0 technologies with Education 4.0: Advantages for improvements in learning. Interact Technol Smart Educ. 2023;20(2):271\u0026ndash;87.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRizvi M. Exploring the landscape of artificial intelligence in education: Challenges and opportunities. 2023 5th International Congress on Human-Computer Interaction, Optimization and Robotic Applications (HORA): 2023: IEEE; 2023: 01\u0026ndash;3.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWang S, Sun Z, Chen Y. Effects of higher education institutes\u0026rsquo; artificial intelligence capability on students' self-efficacy, creativity and learning performance. Educ Inform Technol. 2023;28(5):4919\u0026ndash;39.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKickmeier-Rust MD, Albert D. A domain model for smart 21st century skills training in game-based virtual worlds. In: \u003cem\u003e2012 IEEE 12th International Conference on Advanced Learning Technologies: 2012\u003c/em\u003e: IEEE; 2012: 680\u0026ndash;681.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBudianto L, Arifani Y, Wardhono A, Poedjiastutie D. The TPACK level of in-service EFL teacher online professional training: The perspectives of teachers, students, and trainers. J Digit Learn Teacher Educ. 2023;39(2):87\u0026ndash;101.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eEkawati E, Prastyo A. Optimization of TPACK-based Project Learning in Micro-teaching Courses in Physics Education Study Programs during the Pandemic. In: \u003cem\u003eJournal of Physics: Conference Series\u003c/em\u003e: 2022: IOP Publishing; 2022: 012035.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003evan der Merwe D, Levigne-Lang R. The lessons learnt from emergency remote teaching to strengthen a pre-service teacher education course on lesson design. J Univ Teach Learn Pract. 2023;20(3):03.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMax AL, Lukas S, Weitzel H. The pedagogical makerspace: Learning opportunity and challenge for prospective teachers' growth of TPACK. Br J Edu Technol 2023.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eStokel-Walker C, Van Noorden R. What ChatGPT and generative AI mean for science. Nature. 2023;614(7947):214\u0026ndash;6.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYounis HA, Ruhaiyem NIR, Ghaban W, Gazem NA, Nasser M. A Systematic Literature Review on the Applications of Robots and Natural Language Processing in Education. Electronics. 2023;12(13):2864.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eVygotsky LS, Cole M. Mind in society: Development of higher psychological processes. Harvard university press; 1978.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eVygotsky LS. Thought and language. MIT press; 2012.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eStojanov A. Learning with ChatGPT 3.5 as a more knowledgeable other: an autoethnographic study. Int J Educational Technol High Educ. 2023;20(1):35.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e\u003cem\u003ePatterns\u003c/em\u003e 2023, 4(1).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLyonga NAN. Peer Learning amongst Students of Higher Technical Teachers' Training College (HTTTC) of the University of Buea in Kumba, Cameroon. Int J High Educ. 2018;7(2):216\u0026ndash;26.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eEmami A. 12 Natural Language Processing: Current Methods and Challenges. \u003cem\u003eEngineering Mathematics and Artificial Intelligence: Foundations, Methods, and Applications\u003c/em\u003e 2023:261.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBrown T, Mann B, Ryder N, Subbiah M, Kaplan JD, Dhariwal P, Neelakantan A, Shyam P, Sastry G, Askell A. Language models are few-shot learners. Adv Neural Inf Process Syst. 2020;33:1877\u0026ndash;901.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKanwit M, Solon M. Communicative Competence in a Second Language: Theory, Method, and Applications. Taylor \u0026amp; Francis; 2022.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFoster-Cohen S. The communicative competence of young children: a modular approach. Routledge; 2014.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMcDonough S. Applied linguistics in language education. Routledge; 2017.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHutchins J. From first conception to first demonstration: the nascent years of machine translation, 1947\u0026ndash;1954. a chronology. Mach Transl. 1997;12:195\u0026ndash;252.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTuring AM. Computing machinery and intelligence. Springer; 2009.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChomsky N. The Logical Structure of Linguistic Theory Plenum Press. \u003cem\u003eNew York\u003c/em\u003e 1975.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHewitt J, Thickstun J, Manning CD, Liang P. Backpack Language Models. \u003cem\u003earXiv preprint arXiv:230516765\u003c/em\u003e 2023.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCaucheteux C, Gramfort A, King J-R. Evidence of a predictive coding hierarchy in the human brain listening to speech. Nat Hum Behav. 2023;7(3):430\u0026ndash;41.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFeder A, Keith KA, Manzoor E, Pryzant R, Sridhar D, Wood-Doughty Z, Eisenstein J, Grimmer J, Reichart R, Roberts ME. Causal inference in natural language processing: Estimation, prediction, interpretation and beyond. Trans Association Comput Linguistics. 2022;10:1138\u0026ndash;58.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKumar S, Correa CG, Dasgupta I, Marjieh R, Hu MY, Hawkins R, Cohen JD, Narasimhan K, Griffiths T. Using natural language and program abstractions to instill human inductive biases in machines. Adv Neural Inf Process Syst. 2022;35:167\u0026ndash;80.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eToneva M, Wehbe L. Interpreting and improving natural-language processing (in machines) with natural language-processing (in the brain). \u003cem\u003eAdvances in neural information processing systems\u003c/em\u003e 2019, 32.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRavimaram S, Sathish A, Vatchala S, Rawat R, TF MR. Robust Transfer Learning Based Modelling for Accelerating the Learning of Ai in the Field of NLP. In: 2023 \u003cem\u003e9th International Conference on Advanced Computing and Communication Systems (ICACCS): 2023\u003c/em\u003e: IEEE; 2023: 1026\u0026ndash;1030.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMuller B, Gupta D, Fauconnier J-P, Patwardhan S, Vandyke D, Agarwal S. Languages You Know Influence Those You Learn: Impact of Language Characteristics on Multi-Lingual Text-to-Text Transfer. Transfer Learning for Natural Language Processing Workshop: 2023: PMLR; 2023: 88\u0026ndash;102.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eShen R, Pan W. Research on transfer learning technology in natural language processing. In: \u003cem\u003eArtificial Intelligence in China: Proceedings of the 2nd International Conference on Artificial Intelligence in China: 2021\u003c/em\u003e: Springer; 2021: 483\u0026ndash;488.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBiggers FB, Mohanty SD, Manda P. A deep semantic matching approach for identifying relevant messages for social media analysis. Sci Rep. 2023;13(1):12005.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGalinsky R, Alekseev A, Nikolenko S. Improving Neural Models for Natural Language Processing in Russian with Synonyms. J Math Sci 2023:1\u0026ndash;12.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRinaldi AM, Russo C, Tommasino C. Automatic image captioning combining natural language processing and deep neural networks. Results in Engineering. 2023;18:101107.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBiswas SS. Role of chat gpt in public health. Ann Biomed Eng. 2023;51(5):868\u0026ndash;9.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLund BD, Wang T. Chatting about ChatGPT: how may AI and GPT impact academia and libraries? Libr Hi Tech News. 2023;40(3):26\u0026ndash;9.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSurameery NMS, Shakor MY. Use chat gpt to solve programming bugs. \u003cem\u003eInternational Journal of Information Technology \u0026amp; Computer Engineering (IJITC) ISSN: 2455\u0026ndash;5290\u003c/em\u003e 2023, 3(01):17\u0026ndash;22.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJiao W, Wang W, Huang J-t, Wang X, Tu Z. Is ChatGPT a good translator? A preliminary study. arXiv preprint arXiv:230108745 2023.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLiu J, Wang C, Liu S. Utility of ChatGPT in clinical practice. J Med Internet Res. 2023;25:e48568.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTlili A, Shehata B, Adarkwah MA, Bozkurt A, Hickey DT, Huang R, Agyemang B. What if the devil is my guardian angel: ChatGPT as a case study of using chatbots in education. Smart Learn Environ. 2023;10(1):15.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWhalen J, Mouza C. ChatGPT: Challenges, Opportunities, and Implications for Teacher Education. Contemp Issues Technol Teacher Educ. 2023;23(1):1\u0026ndash;23.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePelton T, Pelton LF. Adapting ChatGPT to Support Teacher Education in Mathematics. In: \u003cem\u003eSociety for Information Technology \u0026amp; Teacher Education International Conference: 2023\u003c/em\u003e: Association for the Advancement of Computing in Education (AACE); 2023: 1662\u0026ndash;1670.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTirado-Olivares S, Nav\u0026iacute;o-Ingl\u0026eacute;s M, O\u0026rsquo;Connor-Jim\u0026eacute;nez P, C\u0026oacute;zar-Guti\u0026eacute;rrez R. From Human to Machine: Investigating the Effectiveness of the Conversational AI ChatGPT in Historical Thinking. Educ Sci. 2023;13(8):803.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDeng J, Lin Y. The benefits and challenges of ChatGPT: An overview. Front Comput Intell Syst. 2022;2(2):81\u0026ndash;3.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLiebrenz M, Schleifer R, Buadze A, Bhugra D, Smith A. Generating scholarly content with ChatGPT: ethical challenges for medical publishing. Lancet Digit Health. 2023;5(3):e105\u0026ndash;6.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKalla D, Smith N. Study and Analysis of Chat GPT and its Impact on Different Fields of Study. Int J Innovative Sci Res Technol 2023, 8(3).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZack T, Lehman E, Suzgun M, Rodriguez JA, Celi LA, Gichoya J, Jurafsky D, Szolovits P, Bates DW, Abdulnour R-EE. Coding Inequity: Assessing GPT-4's Potential for Perpetuating Racial and Gender Biases in Healthcare. \u003cem\u003emedRxiv\u003c/em\u003e 2023:2023.2007. 2013.23292577..\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKoehler MJ, Mishra P, Cain W. What is technological pedagogical content knowledge (TPACK)? J Educ. 2013;193(3):13\u0026ndash;9.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGudmundsdottir S, Shulman L. Pedagogical content knowledge in social studies. Scandinavian J Educationl Res. 1987;31(2):59\u0026ndash;70.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAbubakir H, Alshaboul Y. Unravelling EFL teachers' mastery of TPACK: Technological pedagogical and content knowledge in writing classes. Heliyon; 2023.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHarris JB, Phillips M, Koehler MJ, Rosenberg JM. Editorial 33 (3): TPCK/TPACK research and development: Past, present, and future directions. Australasian J Educational Technol 2017, 33(3).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCrawford JR, Garthwaite PH, Porter S. Point and interval estimates of effect sizes for the case-controls design in neuropsychology: rationale, methods, implementations, and proposed reporting standards. Cognit Neuropsychol. 2010;27(3):245\u0026ndash;60.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCrawford JR, Garthwaite PH. Single-case research in neuropsychology: a comparison of five forms of t-test for comparing a case to controls. Cortex. 2012;48(8):1009\u0026ndash;16.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFerreira-Correia A, Anderson DG, Cockcroft K, Krause A. The neuropsychological deficits and dissociations in Huntington Disease-Like 2: A series of case-control studies. Neuropsychologia. 2020;136:107238.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCrawford JR, Garthwaite PH. Comparison of a single case to a control or normative sample in neuropsychology: Development of a Bayesian approach. Cognit Neuropsychol. 2007;24(4):343\u0026ndash;72.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLincoln Y. The SAGE encyclopedia of social science research methods. 2008.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDrummond A, Sweeney T. Can an objective measure of technological pedagogical content knowledge (TPACK) supplement existing TPACK measures? Br J Edu Technol. 2017;48(4):928\u0026ndash;39.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eValtonen T, Sointu E, Kukkonen J, Kontkanen S, Lambert MC, M\u0026auml;kitalo-Siegl K. TPACK updated to measure pre-service teachers\u0026rsquo; twenty-first century skills. Australasian J Educational Technol 2017, 33(3).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKartal T, Kartal B, Uluay G. Technological pedagogical content knowledge self-assessment scale (TPACK-SAS) for pre-service teachers: Development, validity and reliability. Int J Eurasia Social Sci. 2016;7(23):1\u0026ndash;36.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHsu C-Y, Tsai M-J, Chang Y-H, Liang J-C. Surveying in-service teachers\u0026rsquo; beliefs about game-based learning and perceptions of technological pedagogical and content knowledge of games. J Educational Technol Soc. 2017;20(1):134\u0026ndash;43.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYurdakul IK, Odabasi HF, Kilicer K, Coklar AN, Birinci G, Kurt AA. The development, validity and reliability of TPACK-deep: A technological pedagogical content knowledge scale. Comput Educ. 2012;58(3):964\u0026ndash;77.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCarter N, Bryant-Lukosius D, DiCenso A, Blythe J, Neville AJ. The use of triangulation in qualitative research. In: \u003cem\u003eOncol Nurs Forum\u003c/em\u003e: 2014; 2014: 545\u0026ndash;547.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMacmillan NA, Creelman CD. Detection theory: A user's guide. Psychology press; 2004.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDrummond A. TPACK Objective Dataset. In.; 2023. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://osf.io/3rhxu/\u003c/span\u003e\u003cspan address=\"https://osf.io/3rhxu/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGabbiani F, Cox SJ. Mathematics for neuroscientists: Academic Press; 2017.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGreen DM, Swets JA. Signal detection theory and psychophysics. Volume 1. Wiley New York; 1966.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBraun V, Clarke V. Thematic analysis: American Psychological Association; 2012.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLiu R, Jia C, Zhang G, Zhuang Z, Liu T, Vosoughi S. Second thoughts are best: Learning to re-align with human values from text edits. Adv Neural Inf Process Syst. 2022;35:181\u0026ndash;96.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHendrycks D, Burns C, Basart S, Critch A, Li J, Song D, Steinhardt J. Aligning ai with shared human values. arXiv preprint arXiv:200802275 2020.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLu P, Mishra S, Xia T, Qiu L, Chang K-W, Zhu S-C, Tafjord O, Clark P, Kalyan A. Learn to explain: Multimodal reasoning via thought chains for science question answering. Adv Neural Inf Process Syst. 2022;35:2507\u0026ndash;21.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRoy NA, Kim J, Rabinowitz N. Explainability Via Causal Self-Talk. Adv Neural Inf Process Syst. 2022;35:7655\u0026ndash;70.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWei J, Bosma M, Zhao VY, Guu K, Yu AW, Lester B, Du N, Dai AM, Le QV. Finetuned language models are zero-shot learners. arXiv preprint arXiv:210901652 2021.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKatz DM, Bommarito MJ, Gao S, Arredondo P. Gpt-4 passes the bar exam. Available at SSRN 4389233 2023.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLi J, Ouyang J, Liu J, Zhang F, Wang Z, Guo X, Liu M, Taylor D. Artificial intelligence-based online platform assists blood cell morphology learning: A mixed-methods sequential explanatory designed research. Med Teach. 2023;45(6):596\u0026ndash;603.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePhilbin CA. Exploring the Potential of Artificial Intelligence Program Generators in Computer Programming Education for Students. ACM Inroads. 2023;14(3):30\u0026ndash;8.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGuilherme A. Considering AI in education: Erziehung but never Bildung. Artif Intell Incl Education: Specul Futures Emerg Practices 2019:165\u0026ndash;78.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGill SS, Xu M, Patros P, Wu H, Kaur R, Kaur K, Fuller S, Singh M, Arora P, Parlikad AK. Transformative effects of ChatGPT on modern education: Emerging Era of AI Chatbots. Internet of Things and Cyber-Physical Systems. 2024;4:19\u0026ndash;23.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSobo E. Could ChatGPT Prompt a New Golden Age in Higher Education? Teach Learn Anthropol 2023, 6(1).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChan CKY, Hu W. Students' Voices on Generative AI: Perceptions, Benefits, and Challenges in Higher Education. arXiv preprint arXiv:230500290 2023.\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eEthical approval and consent to participate:\u0026nbsp;\u003c/strong\u003eThe authors confirm that all the methods comply with current guidelines and regulations that follow the Declaration of Helsinki.\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003eTPACK-objective dataset (a publicly available dataset), which was used in this study as normative data, had been collected with approval from the Social and Behavioural Research Ethics Committee at Flinders University, where participation was voluntary and anonymous, and informed consent had been obtained from all participants.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConsent for publication:\u0026nbsp;\u003c/strong\u003eNot applicable.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting interests:\u003c/strong\u003e The authors of this study declare that they have no competing interests.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData Availability:\u003c/strong\u003e The data collected in this study will be deposited on the Open Science Framework (OSF) and will be publicly accessible. Researchers interested in accessing the GPT dataset can find it on OSF at\u0026nbsp;\u003ca href=\"https://osf.io/xtce4/\"\u003ehttps://osf.io/xtce4/\u003c/a\u003e upon publication. The TPACK-objective dataset is available at\u0026nbsp;\u003ca href=\"https://osf.io/3rhxu/\"\u003ehttps://osf.io/3rhxu/\u003c/a\u003e\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding:\u003c/strong\u003e Not applicable.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eContributions:\u0026nbsp;\u003c/strong\u003eFerdi \u0026Ccedil;elik, Ceylan Yangın Ersanlı, and Aaron Drummond planned, wrote, edited, reviewed, and agreed on the manuscript.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAcknowledgements:\u0026nbsp;\u003c/strong\u003eNot applicable.\u003c/p\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"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":"","lastPublishedDoi":"10.21203/rs.3.rs-3388153/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-3388153/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eThis study employs a single case-control design to evaluate the objective TPACK performance of ChatGPT by comparing it to human preservice teachers. A purposive sampling method selected ChatGPT 3.5 as a potential GenAI peer tutor and 93 Australian preservice teachers as the normative sample. Objective TPACK performance was measured using an adapted TPACK scale. Bayesian analysis revealed that ChatGPT significantly outperformed the human control group with a substantial effect size. Qualitative data analysis of ChatGPT\u0026rsquo;s explanations further demonstrated its comprehensive understanding of TPACK dimensions. While certain limitations were identified, including the narrow focus on Australian preservice teachers and the availability of more advanced AI models, the study emphasizes ChatGPT 3.5\u0026rsquo;s potential as a more knowledgeable other for preservice teachers to enhance their TPACK. This research encourages further investigation into the integration of AI in education and its implications for pedagogical practices. ChatGPT 3.5\u0026rsquo;s remarkable TPACK performance suggests its role as a valuable peer tutor in TPACK-related domains, complementing human educators while recognizing the irreplaceable human aspects of the learning experience.\u003c/p\u003e","manuscriptTitle":"GPT-PACK: Can GenAI act as TPACK Peer Tutor for Preservice Teachers? A Comparative Study on Objective TPACK of Humans and ChatGPT 3.5","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2023-10-02 17:42:28","doi":"10.21203/rs.3.rs-3388153/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":"d04394d9-71e7-49d0-b212-d001b2196de8","owner":[],"postedDate":"October 2nd, 2023","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2023-10-11T18:36:00+00:00","versionOfRecord":[],"versionCreatedAt":"2023-10-02 17:42:28","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-3388153","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-3388153","identity":"rs-3388153","version":["v1"]},"buildId":"_2-kVJe1T_tPrBINL-cwx","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}