Objectives: Traditional evidence-based medicine (EBM) education often presents research as a finished product, which can obscure the dynamic inquiry process and limit students’ understanding of longitudinal decision-making. Because authentic research notes are rarely accessible, generative artificial intelligence (AI) can simulate the thought processes of researchers. This study evaluated the educational potential of a novel AI system that generates simulated research notes based on academic papers. Methods: A cross-sectional survey was conducted with 16 EBM faculty members from Korean medicine colleges. Participants engaged in a hands-on evaluation of a web-based system (utilizing Google Gemini 1.5 Flash API) that generates simulated research notes. A mixed-methods questionnaire comprising 23 Likert-scale items across six domains and six open-ended questions was used for the evaluation. Results: Participants provided moderately positive ratings, with ‘Educational Effectiveness’ (mean 3.97±0.84) and ‘Overall Assessment’ (3.97±0.96) receiving the highest scores. The system’s ability to portray researcher concerns realistically was also well-rated (4.19±0.83), while ‘Practical Applicability’ received the lowest score (3.70±1.01). Qualitative analysis confirmed the system’s educational novelty but highlighted significant concerns regarding AI reliability, including factual inaccuracies, and generated suggestions for practical implementation. Conclusions: This proof-of-concept study demonstrates that AI-generated research note simulations address a gap in process-oriented EBM learning. Faculty acknowledged its conceptual value for teaching longitudinal research thinking; however, substantial technical reliability and implementation challenges must be addressed before it can be utilized by students. While technical limitations may improve with advancements in AI, specific educational challenges require focused pedagogical research. Future efforts should prioritize content accuracy and instructional design for effective classroom integration.
Evidence-based medicine (EBM), which integrates clinical expertise, patient values, and the best available research evidence, has become a foundational paradigm in modern medical decision-making1,2). Consequently, EBM education has been established as an essential component of the medical curriculum. However, the optimal strategies for teaching EBM to undergraduate students are still debated, as prior studies have reported inconsistent outcomes2-4). The predominant pedagogical approach focuses on the critical appraisal of published research articles4). Although this approach is essential for cultivating critical appraisal skills, it presents a major limitation: research is treated as a static final product, obscuring the dynamic and complex processes through which knowledge is generated5-7). For learners, the intellectual journey of a study, from the initial research question and methodological debates to the inevitable setbacks and discoveries, often remains a “black box.”
This focus on the final product, however, can lead to a superficial understanding of scientific inquiry, leaving students without a deep, contextual grasp of the research process itself. A pedagogical shift is therefore necessary, moving beyond the static analysis of a finished paper toward a more holistic, process-oriented longitudinal perspective. This approach does not require extending the curriculum across multiple years; instead, it focuses on reconstructing the complete narrative of a single research project. Following a study’s journey from its initial question to its conclusion allows learners to appreciate the context and complexities of scientific work, a skill essential for navigating the ambiguities of clinical reasoning8). As clinical decision-making becomes more complex, the need for such process-oriented learning grows ever more critical.
The recent emergence of generative artificial intelligence (AI) presents an unprecedented opportunity to realize this educational model. In medical education, generative AI is already being leveraged to create adaptive learning platforms, virtual patient scenarios, and interactive simulations that enhance personalized learning9). However, the application of these powerful tools has largely concentrated on clinical skills and student-teacher interactions9). To date, there has been a notable absence of AI tools designed to simulate the intricate cognitive and decision-making processes of a researcher. If AI could systematically deconstruct a final publication back into its procedural steps, recreating the researcher’s probable thought process, challenges, and decisions, it could offer an innovative solution to the limitations of traditional EBM education. Such a tool would be particularly valuable in fields like Korean medicine (KM), where research often involves complex multi-component interventions, individualized treatments, and the integration of traditional theory with modern scientific methodology10).
Therefore, this study presents and evaluates an educational system that applies generative AI to reverse-engineer (i.e., reconstruct) published academic papers into a series of simulated research notes. The primary objective is to assess the educational utility and potential of this system from the perspective of expert educators in EBM and research methodology. Through this evaluation, this study aims to explore the value, limitations, and future directions of a process-oriented, AI-driven approach to EBM education.
The web-based educational tool evaluated in this study was developed on the Google Apps Script platform, utilizing the Google Gemini 1.5 Flash API to automatically generate research notes. The system is designed to take a published academic paper as input and simulate the thought processes and decision-making steps a researcher might have experienced during the actual study. This process results in a series of educational research notes that reconstruct the study’s progression over time (Fig. 1).
This figure illustrates the workflow of the AI system. It analyzes an uploaded research paper using Google Gemini 1.5 Flash API to chronologically reconstruct the study into simulated research notes. It then generates educational outputs, including categorized discussion questions and automated AI feedback, to support a process-oriented approach to EBM education.
The system’s architecture consists of three core components: a paper analysis engine, a research note generation engine, and an educational support module. The analysis engine parses uploaded research papers (in PDF or Word format) to extract key elements such as the research objective, methodology, results, and conclusion. It uses the Google Drive API for temporary file storage and employs optical character recognition and natural language processing to create a structured representation of the paper’s content.
Based on this structured information, the generation engine applies specialized algorithms tailored to different research designs (e.g., literature reviews, clinical intervention studies). For instance, when processing a clinical trial, it simulates the stages of protocol development, institutional review board (IRB) approval, participant recruitment, intervention, and data analysis. Each generated note is designed to reflect the information realistically available to a researcher at that specific point in time, intentionally omitting future results or conclusions to provide an authentic, process-oriented learning experience (Fig. 2).
This figure illustrates the process of generating research notes based on a user’s uploaded research results. It shows how the system takes input parameters (e.g., number of notes, research type) and applies an algorithm to create a series of notes that simulate a realistic, process-oriented research journey. The generated notes are designed to reflect the information available at each stage, mirroring the actual research process.
Finally, the educational support module automatically generates discussion questions for instructors based on the research notes, categorized into three difficulty levels (bachelor’s, master’s, and doctoral). It also includes a feature for providing AI-based feedback on student responses, aiming to reduce the instructor’s workload while offering individualized learning support (Fig. 3).
This figure demonstrates the educational support functionality of the system. It shows how the platform automatically generates discussion questions for instructors at different academic levels (e.g., doctoral) and provides a dedicated interface for students to submit their answers. The system also includes an AI-based feedback module that offers personalized feedback on student responses, streamlining the teaching process and enhancing learning.
This study employed a cross-sectional, descriptive anonymous survey design to evaluate the educational utility of the system. Data were collected via an online survey conducted over one month, from July 31 to August 31, 2025. The target population consisted of full-time faculty members at KM colleges and graduate schools across the nation with experience teaching EBM or research methodology. Inclusion criteria were: (1) current full-time faculty status at a college of KM; (2) experience conducting at least one formal course or informal educational activity (e.g., seminar, workshop) related to EBM or research methodology within the last three years; and (3) voluntary consent to participate. Part-time lecturers and faculty with no relevant teaching experience were excluded.
Due to the specialized nature of the target population and the lack of a formal registry, participants were recruited using a snowball sampling strategy11). Initial contact was made with faculty at the researchers’ university and affiliated institutions. These initial participants were then asked to recommend other eligible colleagues from their professional networks. A target sample size of approximately 20 participants was set, considering the total number of KM colleges in the country. To encourage participation, a mobile gift voucher (equivalent to 10,000 KRW) was provided to each participant upon survey completion.
As no standardized instrument existed for evaluating this type of educational tool, a new questionnaire was developed by the research team based on a review of relevant literature and expert opinion12). The survey consisted of three sections, as follows.
1) Respondent Demographics (6 items): Gender, age group, teaching experience, academic specialty, primary audience (e.g., undergraduate, graduate), and frequency of teaching EBM/research methodology.
2) Quantitative Evaluation (23 items): A 5-point Likert scale (1=strongly disagree, 5=strongly agree) was used to assess six domains, including Content Quality (5 items), Educational Effectiveness (4 items), System Usability (4 items), AI Content Reliability (4 items), Practical Applicability (4 items), and Overall Assessment (2 items).
3) Qualitative Feedback (6 open-ended questions): Questions explored perceived educational contributions, potential limitations, ethical considerations, feedback on specific features, and suggestions for improvement (Supplementary File 1).
Participants were first instructed to interact directly with the educational system. They were asked to upload a research paper of their choosing, select the study type, and set the number of research notes to be generated. This hands-on approach allowed them to evaluate the system’s performance on a topic relevant to their own teaching interests. After experiencing the system, participants completed the anonymous online survey to provide their feedback.
This study was approved by the Institutional Review Board of Dong-eui University Korean Medicine Hospital on July 25, 2025 (Approval No: DH-2025-09). All participants were provided with a detailed explanation of the study’s purpose, procedures, potential risks and benefits, and data privacy measures before providing informed consent. All collected data were anonymized using research codes. Personally identifiable information was limited to mobile phone numbers for gift voucher distribution and was deleted immediately after disbursement.
Given the small sample size, the quantitative data were analyzed using descriptive statistics. Continuous variables from the Likert scale items were presented as means and standard deviations (SD). Categorical demographic data were summarized as frequencies and percentages. For the qualitative data from the open-ended questions, a thematic content analysis was conducted to identify and categorize key themes and opinions13). All statistical analyses were performed using Microsoft Excel 365 (Microsoft Corp., Redmond, WA, USA).
A total of 20 potential participants were asked to access the system and participate in this survey, of which 16 members participated. A half of the participants were female (n=8, 50%) with the largest group being in their 40s (n=10, 62.5%). Regarding teaching experience, the majority had 5 years or less of experience (n=8, 50.0%), followed by 10∼15 years (n=3, 18.75%), 5∼10 years (n=3, 18.75%), and 15 years or more (n=2, 12.5%). In terms of academic specialization, basic medicine had the highest representation (n=8, 50%), followed by clinical medicine (n=6, 37.5%) and medical humanities (n=2, 12.5%). Most participants taught undergraduate students, with 15 (93.75%) teaching clinical students and 7 (43.75%) teaching pre-clinical students. Ten participants (62.5%) also taught graduate students, while none reported teaching residents/trainees. Regarding EBM/research methodology teaching frequency over the past three years, 12 participants (75%) conducted regular coursework at least once annually, while 2 (12.5%) conducted regular coursework less than once annually, and 2 (12.5%) had no regular educational activities in the past three years. For informal educational activities, 9 participants (56.25%) engaged in activities such as seminars or workshops at least once yearly, 6 (37.5%) participated less than once annually, and 1 participant (6.25%) had no informal activities (Table 1).
The quantitative evaluation results across six domains are presented in Table 2. Overall, participants provided moderately positive ratings across most evaluation domains.
AI: artificial intelligence, EBM: evidence-based medicine, SD: standard deviation.
The highest-rated domain was Educational Effectiveness (3.97±0.84) and Overall Assessment (3.97±0.96), followed by System Usability (3.95±0.67). Content Quality received a mean score of 3.93±0.81, while AI Content Reliability scored 3.75±0.75. The Practical Applicability domain received the lowest mean score (3.70±1.01), with “immediate applicability to current courses” (Q24) receiving the lowest individual item score (3.44±1.09). The highest individual item scores were for “realistic portrayal of researcher concerns” (Q10: 4.19±0.83) and “intuitive and convenient interface” (Q16: 4.19±0.66). Other highly rated items included “improvement in research methodology understanding” (Q12: 4.06±1.06), “suitability of instructor discussion questions” (Q17: 4.06±0.68), and “high educational value system overall” (Q28: 4.06±0.93). Notably, there was considerable variability in responses, with one participant providing consistently low scores (1∼2 points) across most domains, while others provided predominantly high scores (4∼5 points).
The qualitative analysis of open-ended responses revealed three major themes regarding participants’ perspectives on the educational system.
Participants acknowledged the system’s innovative approach to providing a process-oriented learning experience. One clinical medicine faculty member noted, “I was surprised by how well it incorporated the concerns researchers would likely have. The sequence and slight differences are similar to the actual concerns I had” (Participant A). A medical humanities faculty member emphasized the novelty of this approach, stating, “Research process education generally requires direct research experience, but experiencing it in advance through research notes is quite novel” (Participant C). The system was seen as a valuable tool for research methodology education, offering students a chance to “understand what kind of process research would involve, like a simulation in advance” (Participant E). Another participant commented that the tool could guide “students who have not yet conducted research... step-by-step while reducing fear” (Participant L) (Supplementary File 2).
Despite the positive reception of its concept, participants reported some issues with the system’s output and performance. The most critical concern was the accuracy of the AI-generated content. One basic medicine faculty member reported significant content distortions: “The research notes generated by the current system do not properly understand the paper and write content different from the paper, distorting results (recording opposite results or stating that a systematic review paper marked as IRB-exempt needed IRB preparation)” (Participant J). This participant further noted that when a general LLM was given the same task, “there are fewer result distortions or interpretation errors”. Technical limitations were also a recurring theme. Participants mentioned that “research note generation took longer than expected (over 10 minutes)” (Participant F) and that “the system is unstable. It generated completely different content from the uploaded paper” (Participant I). Another participant reinforced these concerns, noting that the generation speed was “on the slow side” and that there was “a slight tendency for the AI to arbitrarily make up some of the content” (Participant P). Minor inconsistencies, such as incorrect numbers of subjects or submitted journals, were also noted, though one participant found them reasonably acceptable: “As with any AI, the details are somewhat different from the paper (the number of subjects, the journal to which they are submitted, etc.) but I think it can be sufficiently considered” (Participant A) (Supplementary File 2).
Beyond identifying limitations, the participants offered extensive constructive feedback aimed at bridging the gap between the system’s innovative concept and its current practical application. The suggestions ranged from immediate technical fixes to fundamental questions about the system’s pedagogical framework. On a practical level, participants highlighted several ways to improve the user experience and the simulation’s authenticity. A common point of friction was the system’s text-heavy interface, which a participant noted made “readability difficult” (Participant H). To make the tool more versatile for both instructors and students, users requested basic functionalities like the ability to have “research note generation results... downloaded as files” (Participant A) and ensuring that “tablets work smoothly” (Participant M). Suggestions also aimed to enrich the educational content itself. To move beyond plain text, participants envisioned “adding functions for generating tables, organizational charts, and task execution stages” (Participant C), with another suggesting that if “charts or tables were utilized, it would become more intuitive and increase understanding” (Participant P). To make the simulation more immersive, another suggested allowing users to input the “names of first authors, corresponding authors, and research assistants separately so they can be automatically used in research notes” (Participant L).
The feedback on the educational features revealed a desire for more refined and targeted tools. The instructor question-generator, while appreciated in concept, was seen as needing improvement. One clinical medicine faculty member suggested that a structure based on the research process, “‘research questions,’ ‘major research methodology,’ ‘interpretation,’ and ‘conclusion drawing’”, would be more effective than the current academic levels (Participant E). Another faculty member recommended providing “example answers along with questions” (Participant G) to better guide instructors. The AI-powered feedback module received specific criticism, with one participant noting that its scoring system required “much improvement,” particularly because “a high score is measured even when there is no answer content” (Participant P).
Ultimately, these specific suggestions pointed to a larger, more fundamental issue: the need for a clearer pedagogical purpose. Even with technical improvements, some faculty were unsure how to integrate the system into their teaching. This challenge was best captured by one participant who asked, “I don’t know how to use it in class... if the system proceeds in its own way and doesn’t match the direction of my class, it would be difficult to use” (Participant B). For the system to be successfully adopted, he concluded, “specific (detailed) purposes are needed, and learning objectives for specific topics should come first” (Participant B) (Supplementary File 2).
To our knowledge, this study is the first formal evaluation of a generative AI-based educational system designed to simulate the research process through automated research note generation. While recent scoping reviews have identified various AI applications in medical education including intelligent tutoring systems, virtual patients, and adaptive learning platforms, none have specifically addressed the simulation of longitudinal research processes14,15). As a proof-of-concept evaluation, the findings provide initial insights into the feasibility and potential of using AI to address a fundamental limitation in current EBM education, the inability to observe the longitudinal thought processes of researchers during actual studies5-7).
The generally favorable ratings across most domains suggest that the underlying educational concept—using simulated research notes to provide process-oriented learning experiences—was well received by medical educators. The highest individual item score for “realistic portrayal of researcher concerns” (4.19±0.83) indicates that participants recognized the educational value of exposing students to the temporal progression of research decision-making. This approach addresses a long-standing pedagogical challenge in EBM education. Traditional methods rely heavily on retrospective analysis of published papers, which inevitably obscure the uncertainties, setbacks, and iterative decisions that characterize actual research processes5-7). Since authentic research notes from published studies are rarely available to educators, AI-generated simulation represents a potentially viable alternative for providing students with longitudinal exposure to research thinking. The qualitative findings support this conceptual framework, with participants describing the approach as “quite novel” and potentially valuable for students lacking research experience. One participant noted that “research process education generally requires direct research experience, but experiencing it in advance through research notes is quite novel,” highlighting the unique educational niche this approach could fill.
The content reliability issues identified in this study must be understood within the context of the specific AI model employed (Gemini 1.5 Flash API) and the current state of generative AI technology. The AI Content Reliability domain scored moderately (3.75±0.75), but qualitative feedback revealed significant problems including content distortions and factual inaccuracies. Multiple participants reported instances where the system generated opposite research results or inappropriate procedural recommendations. These technical limitations likely reflect the current capabilities of the specific AI model used rather than fundamental flaws in the educational concept itself. Given the rapid pace of large language model development, many of these accuracy issues may be addressed through future technological improvements rather than changes in research design16). The reported superiority of general-purpose language models in avoiding content distortions (“When I use a commonly available LLM to upload the same paper and ask it to write 10 days’ worth of research notes, there are not many result distortions or interpretation errors”) may indicate that newer, more sophisticated AI systems could perform substantially better.
However, certain challenges will require targeted educational research solutions regardless of technological improvements. Key challenges include establishing pedagogical frameworks that integrate AI-generated content with existing EBM curricula and developing instructor training protocols to support process-oriented discussions. Equally important is the creation of assessment methods that evaluate students’ understanding of research processes rather than solely their final outcomes. Additionally, educational researchers must address content validation challenges specific to academic contexts, determining how to verify that simulated research notes accurately reflect realistic decision-making patterns, establishing quality control standards for educational AI applications, and developing methods to customize content for different learning levels and institutional contexts. These education-specific challenges require domain expertise in medical pedagogy and cannot be resolved simply through general advances in AI capabilities, as they involve fundamental questions about how students best learn complex, longitudinal thinking processes that are distinct from the technical problems of content generation accuracy17).
The study design appropriately focused on expert educator evaluation before student deployment, providing essential formative feedback for system development. The Practical Applicability domain’s lower scores (3.70±1.01) and qualitative concerns about classroom integration reflect the expected limitations of a prototype system rather than insurmountable barriers to the educational approach. Faculty requests for enhanced functionality (e.g., downloadable outputs, tablet compatibility, visual elements) represent straightforward development priorities that could significantly improve usability. More substantive pedagogical concerns, such as the need for clearer learning objectives and structured implementation guides, highlight the importance of comprehensive educational design beyond the technological components. The considerable variability in participant responses, including one consistently negative evaluation, suggests that successful implementation will require addressing diverse faculty perspectives and comfort levels with educational technology. This heterogeneity is typical in educational innovation adoption and emphasizes the need for flexible implementation approaches18). Research on educational technology adoption among faculty has demonstrated that early adopters and late majority groups differ significantly in their approaches and needs, suggesting that effective implementation strategies must account for these individual differences rather than adopting a one-size-fits-all approach19).
This study’s implications extend beyond individual classroom applications to the broader medical education ecosystem. The demonstration that AI can simulate authentic research processes suggests potential applications in continuing medical education, where practicing physicians need to stay current with evolving research methodologies. Healthcare institutions investing in evidence-based practice initiatives may find process-oriented AI tools valuable for training clinical staff in research literacy. Additionally, the concept of reverse-engineering published studies into learning experiences could inform medical journal publishers and professional societies in developing educational supplements to research publications. Moreover, this research raises important questions about standardization and quality control in AI-assisted medical education. Traditional assessment methods in medical education may be insufficient for measuring the complex thinking processes that this type of process-oriented simulation aims to develop. Furthermore, as AI tools become more prevalent in medical education14,15,17), regulatory bodies may need to establish guidelines for validating educational AI applications, particularly those that simulate professional thinking processes.
Several limitations constrain the interpretation of these findings. The small sample size (n=16) and convenience sampling approach limit generalizability, particularly given recruitment through professional networks. The evaluation represents initial impressions of a prototype system rather than sustained use or measured learning outcomes. The absence of student perspectives represents a significant limitation, as educator perceptions may not predict actual learning effectiveness. The focus on KM faculty, while appropriate for initial evaluation, raises questions about applicability across different medical disciplines. More critically, this study evaluated a specific technical implementation rather than the broader educational concept. The technology-dependent limitations identified here may not reflect the ultimate feasibility of AI-assisted process-oriented learning as AI capabilities continue advancing. Furthermore, the survey instrument combined questions assessing the educational concept with those evaluating the system’s technical performance. This design means that participants’ opinions on one aspect could have influenced their evaluation of another. For instance, technical frustrations with the prototype, such as slow generation speed or instability, may have led to a more critical assessment of the underlying educational methodology. Conversely, enthusiasm for the novel teaching approach might have resulted in a more lenient evaluation of the system’s technical flaws. Finally, it is worth noting a limitation in the survey’s design itself, where the conceptual overlap between domains like ‘Content Quality’ and ‘AI Content Reliability’ could have led to ambiguity in how participants responded. This lack of clear distinction may have influenced the consistency of our findings.
Future research should distinguish between technology development and educational research priorities. Immediate technical improvements (e.g., enhanced accuracy, better content validation, improved user interfaces) may emerge naturally from general AI advancement16). Educational research should focus on areas requiring domain-specific investigation: optimal pedagogical integration strategies, instructor training approaches, student learning outcome measurement, and comparative effectiveness studies. Longitudinal evaluation of actual classroom implementation, preferably using more advanced AI systems, represents a critical next step. Such studies should include student learning outcomes, comparative analyses with traditional teaching methods, and assessment of long-term impact on research methodology understanding. The development of standardized evaluation frameworks for educational AI systems would facilitate more rigorous assessment of future iterations. Additionally, research into optimal content curation and validation processes for AI-generated educational materials could inform broader applications beyond research methodology training.
This study provides a proof-of-concept evaluation of a novel AI-based system designed to simulate the research process for EBM education. Faculty educators recognized the conceptual value of the system in addressing a key gap in traditional pedagogy: the lack of exposure to the longitudinal, process-oriented nature of scientific inquiry. The potential to provide students with a simulated experience of a researcher’s journey was seen as a significant educational innovation. However, the evaluation also revealed critical barriers to practical implementation, most notably the significant concerns regarding the reliability and accuracy of the AI-generated content. While participants acknowledged the educational potential, the system’s current technical instability and the need for a more defined pedagogical framework for classroom integration prevent its immediate adoption. Ultimately, these findings suggest that while AI-generated research note simulation is a promising approach to enrich EBM education, its successful deployment is contingent upon overcoming both technology-dependent limitations, which may improve with AI advancements, and education-specific challenges that require dedicated pedagogical research and instructional design.
(Items Q1-Q6 covered respondent's gender, age group, teaching experience, academic specialty, primary teaching audience, and frequency of teaching EBM/research methodology.)
(5-point Likert scale: 1=Strongly Disagree, 5=Strongly Agree)
A. Content Quality
• Q7. The generated research notes effectively capture the core content of the paper.
• Q8. The progression of the research process is naturally expressed in each entry.
• Q9. The academic expressions used are appropriate yet easy to understand.
• Q10. The researcher's concerns and decision-making processes are portrayed realistically.
• Q11. The option to choose between a 'diary style' and a 'formal note style' is educationally useful.
B. Educational Effectiveness
• Q12. It will be helpful for improving students' understanding of research methodology.
• Q13. It can be an effective tool for achieving EBM educational goals.
• Q14. It can contribute to the development of students' critical thinking skills.
• Q15. It explains the complex research process well, tailored to the students' level.
C. System Usability
• Q16. The system interface is intuitive and convenient to use.
• Q17. The discussion questions for instructors are suitable for use in class.
• Q18. The feature for collecting student responses is useful for educational evaluation.
• Q19. The overall user experience is satisfactory.
D. AI Content Reliability
• Q20. The generated content is logical and consistent.
• Q21. The quality is sufficient for use as educational material.
• Q22. The automated evaluation of student answers is of sufficient quality to be useful.
• Q23. It seems applicable to various academic fields.
E. Practical Applicability
• Q24. It can be immediately used in the courses I am currently teaching.
• Q25. It can efficiently reduce class preparation time.
• Q26. It can be effectively used in conjunction with existing teaching methods.
• Q27. I am willing to recommend it to my colleagues.
F. Overall Assessment
• Q30. Educational Contribution: How do you think this system can contribute to EBM/research methodology education? Please describe specific applications or expected effects.
• Q31. Suggestions for Improvement & Limitations: Please freely describe any limitations or areas for improvement you foresee when using the system.
• Q32. Ethical Considerations: Please share your opinions on ethical aspects or reliability issues related to using generative AI in education.
• Q33. Evaluation of Instructor Question Feature: Please evaluate the discussion questions for instructors generated by the system (e.g., usability, difficulty, improvement directions).
• Q34. Suggestions for Additional Features: Please suggest any features or improvement ideas you would like to see added to the system.
• Q35. Additional Comments (Optional)
Note. Q30. Educational Contribution: How do you think this system can contribute to EBM/research methodology education? Please describe specific applications or expected effects.; Q31. Suggestions for Improvement & Limitations: Please freely describe any limitations or areas for improvement you foresee when using the system.; Q32. Ethical Considerations: Please share your opinions on ethical aspects or reliability issues related to using generative AI in education.; Q33. Evaluation of Instructor Question Feature: Please evaluate the discussion questions for instructors generated by the system (e.g., usability, difficulty, improvement directions).; Q34. Suggestions for Additional Features: Please suggest any features or improvement ideas you would like to see added to the system.; Q35. Additional Comments (Optional)
SUPPLEMENTARY MATERIALS Supplementary materials can be found via https://doi.org/10.7231/jon.2025.36.3.107
This work was partly supported by the Institute of Information & Communications Technology Planning & Evaluation (IITP)-Innovative Human Resource Development for Local Intellectualization program grant funded by the Korea government (MSIT) (IITP-2025-RS-2020-II201791, 50%) and Dong-eui University Grant (202500980001, 50%).