ChatGPT’s emergence and subsequent evolution as a generative artificial intelligence tool introduces new ways of assisting students with research design. Fostering research skills with undergraduate students presents opportunities and challenges for faculty to aid with drafting research plans, questions for investigation, and methods for conducting the research. While some educators rightfully voice concerns over the ethical aspects of such a tool, this article will draw on my own experiences using ChatGPT 4.0 as a tool in research project supervision. I demonstrate how to prompt ChatGPT to give useful suggestions that can be used as actionable feedback. I also discuss how to instruct students to include ChatGPT in their research methodology when using the tool to refine research questions.
Large language models1
Large language models (LLMs) like ChatGPT are complex algorithms developed through a type of machine learning called deep learning. In such a process, the algorithm is trained on large datasets of text. By learning the patterns, language, structures, and relationships between words, ChatGPT “remembers” the content in a manner that permits it to engage in human-like conversations, answer questions, create content, or write essays.
LLMs are part of the natural language processing models of artificial intelligence that began in the 1950s with attempts to process and understand human language based on hand-coded rules and logic to interpret text. With the advent of machine learning in the 1980s, statistical models could be trained to learn patterns in text, produce decision trees, or recognize speech and classify text. Neural networks and deep learning allowed more sophisticated understandings of language. Word embedding led to better understanding of context. In the late 2010s, OpenAI’s Generative Pre-trained Transformer (GPT) and Google’s Bidirectional Encoder Representations from Transfers were developed through pre-training on massive datasets that could later be fine-tuned for specific tasks. The 2020s have seen larger, more sophisticated models like Open AI’s GPT-3.5 and GPT-4 that can perform a wide range of language tasks with little to no task-specific training.
AI tools necessitate quality indicators
Predictably, the flurry around ChatGPT and other AI apps has led to the creation of a number of platforms offering subscriptions for AI-based tools for research, including a browser extension that acts as a research assistant and manuscript writing tools, to give just two examples. The growth in AI-powered products resembles the emergence of Web 1.0 and 2.0 tools that followed the advent of the World Wide Web. Those tools were not just used to develop and deliver instruction, they became useful to students for their own course projects, assignments, and assessments.
Educational organizations and universities developed quality indicators to guide in the adoption of web tools. It remains to be seen what indicators will be adopted to ensure the quality of GPT-based tools used in higher education classrooms. At minimum such indicators should consider matters such as bias in the datasets, accessibility, and whether it requires a paid subscription. The challenge will be the scale of the undertaking because the plethora of AI tools that have arisen in just one year’s time has far outstripped that of the web tools. New generation Web 3.0 tools already integrate AI, so the quality indicators we currently use will need updating and upgrading to address the AI aspects of the technology (AI and Web3: How Are They Related? | Rather Labs Careers, n.d.). Nevertheless, educators are already employing ChatGPT and other AI-assisted tools for skills enhancement in teaching and research (Crompton and Burke, 2023).
ChatGPT as a co-creator for research
Knowledge co-creation benefits students, enabling them to participate more easily and with greater motivation in the education process (Pocol et al., 2022). In their research projects, my students co-create at three levels: student-client, student-supervisor, and student/supervisor-AI assistant. This three-tier creation produces a dynamic opportunity to develop knowledge that has immediate practical use for the client, as well as skill-building benefits for the student. As the supervisor, I gain insights into teaching and learning with AI-assisted tools.
Refining research questions with ChatGPT: Supervisor use case
Students in their last year of bachelor studies acquire an applied research project from the workforce. As their supervisor, I guide them in the development of a research question, a research plan to conduct the research and to adopt the methods, and relevant resources to undertake the research. ChatGPT can help at any point in the research process; however, I want to share an example of how I used ChatGPT’s response to give more targeted feedback to my supervisee.
I designed a prompt that contextualized the task by explaining I was supervising a research project. It also instructed the tool to revise the question in a way that identified a specific legal issue.
My prompt:
My student is completing a research project. As supervisor, I have reviewed the main research question and see that it does not adequately identify a legal problem. Evaluate the question and give three suggestions for revising the question in a way that identifies a legal issue. Here is the research question: [paste of the question.]
ChatGPT gave a number of suggestions based on three possible approaches: a specific legal framework, a comparative approach, or a focus on a legal dispute. This was useful for framing the project and gave the student alternate frameworks for the research problem.
One suggestion involving a specific legal framework suggested this rephrasing: “To what extent did the COVID-19 measures taken in Germany align or conflict with the constitutional right to freedom of travel under German Basic Law and international human rights treaties?” (Personal communication with ChatGPT 4.0 on December 28, 2023.)
I used this framing to advise the student to choose one human rights covenant to use as a legal framework thereby narrowing the scope of the research and allowing it to meet the minimum word and content requirements.
Overall, ChatGPT was instrumental in providing the type of actionable feedback that “offers a chance of closing a gap between current performance and the performance expected…” (Mamoon-Al-Bashir, Kabir, and Rahman, 2016). I could suggest the three approaches and share one of the revised research questions with the specific legal framework in my feedback to the student.
Ethical considerations
When suggesting the use of ChatGPT, I ask students to acknowledge its use in a footnote and to include the prompts as an annex to the document or to link them in the footnote. I also explain that using ChatGPT is part of their research method and I ask them to include a paragraph explaining how they prompted ChatGPT and how they used the results to modify their original research questions.
The inclusion in the methodology section goes a step further than what has been suggested in literature for citing ChatGPT or acknowledging its uses (Castellanos-Gomez, 2023). My approach recognizes that ChatGPT is part of the research process as a whole and can be employed to define questions, devise research plans or even find appropriate research methodologies. Therefore, the process of how it was used should be described in the methodology section.
Sample methodology language
To refine and further specify the legal framework for my research, I used chatGPT4.0. I drafted a research question, then asked ChatGPT to suggest ways to make it a concrete legal inquiry. Using the suggestions given, I revised and refined my original questions. ChatGPT’s prompts and responses are attached in Annex One.
Conclusion
ChatGPT has proven itself valuable in formulating research questions. Supervisors can utilize responses to provide actionable feedback, allowing students to meet learning outcomes for the research project. ChatGPT not only enhances the research process but also introduces a novel approach to knowledge co-creation, where students, supervisors, and AI assistants collaborate effectively. However, this integration comes with the responsibility of acknowledging the use of ChatGPT in research methodologies, ensuring ethical compliance and transparency in academic practices.
Tamara N. Lewis Arredondo serves as a senior lecturer, teacher trainer for The Hague Centre for Teaching and Learning, and researcher for the Global and Inclusive Learning Centre of Expertise at The Hague University of Applied Sciences in The Netherlands.
References
‘AI and Web3: How Are They Related? | Rather Labs Careers’. n.d. Accessed 27 December 2023. https://www.ratherlabs.com/blog/ai-and-web3-how-are-they-related.
Castellanos-Gomez, Andres. 2023. ‘Good Practices for Scientific Article Writing with ChatGPT and Other Artificial Intelligence Language Models’. Nanomanufacturing 3 (2): 135–38. https://doi.org/10.3390/nanomanufacturing3020009.
Crompton, Helen, and Diane Burke. 2023. ‘Artificial Intelligence in Higher Education: The State of the Field’. International Journal of Educational Technology in Higher Education 20 (1): 22. https://doi.org/10.1186/s41239-023-00392-8.
Mamoon-Al-Bashir, Md, Md Rezaul Kabir, and Ismat Rahman. 2016. ‘The Value and Effectiveness of Feedback in Improving Students’ Learning and Professionalizing Teaching in Higher Education’. Journal of Education and Practice 7 (16): 38.
Pocol, Cristina Bianca, Liana Stanca, Dan-Cristian Dabija, Ioana Delia Pop, and Sergiu Mișcoiu. 2022. ‘Knowledge Co-Creation and Sustainable Education in the Labor Market-Driven University–Business Environment’. Frontiers in Environmental Science 10. https://www.frontiersin.org/articles/10.3389/fenvs.2022.781075.