Benjamin Smart is Director of The Centre for Philosophy of Epidemiology, Medicine, and Public Health, and Associate Professor of Philosophy at the University of Johannesburg. His research focuses primarily on the philosophy of medicine and public health, the metaphysics of science, and the scholarship of teaching and learning.
Catherine Botha is a full professor in Philosophy at the University of Johannesburg in South Africa, her research and teaching is focused mainly on issues in ethics and aesthetics, especially the philosophy of dance, as well as ethics and aesthetics of artificial intelligence.
They recently published an opinion article that first appeared in the Mail & Guardian on 14 March 2023.
Academic concerns over the threat that artificial intelligence (AI) language models such as ChatGPT pose to the integrity of academic assessments have prompted universities to consider abandoning assessments influenced by these systems and returning to exam conditions for all grades.
The risk of undetectable plagiarism is a significant concern, leading many to view this move as the only viable option, but this strategy is misguided.
We propose a practical guide for the ethical use of AI systems such as ChatGPT in academic essay writing. This guide aims to help academics teach students how to write significantly better essays without resorting to cheating, and to assist academics in using these tools to improve their research.
Why encourage the use of ChatGPT-like AI systems?
The authors of this article are academic philosophers. We asked ChatGPT to answer a series of essay questions one might expect a first-year philosophy student to answer. These included:
- “Evaluate Descartes’ Argument from Dreaming.”
- “Is Presentism the most plausible philosophical theory of time?”
- “Is Dispositional Essentialism a viable alternative to Humeanism about causation”
ChatGPT provided a basic answer in each case, with numerous inaccuracies. Were the essay written by a student, one would infer only a superficial understanding of the core issues. Although the essays were well-written, they would receive either a failing grade or a low passing grade because of the lack of references or the inclusion of fabricated ones.
Furthermore, ChatGPT’s inability to evaluate arguments and lack of nuanced understanding of the questions posed made it unsuitable for higher education assessments, especially for subjects such as philosophy that require critical thinking.
Relying on ChatGPT to write essays is not a viable option for obtaining good grades.
But there are ethical ways to use AI systems when writing academic essays. Wealthy students and academics have historically been able to employ copy editors to improve the quality of their work, but this option is exclusionary because of its cost. ChatGPT can serve as a free and accessible copy editor to help disadvantaged students, who often write essays in a second or third language, addressing one of the inequalities they face in higher education.
In addition to copy editing, AI language models can help students and academics plan the structure of their essays and articles, participate in debates to clarify ideas and arguments, provide feedback on the final product and offer suggestions for improvement. Although independent research is necessary for producing high-quality academic work, AI language models can significantly enhance the quality of outputs without resorting to plagiarism or other forms of cheating.
Planning the essay
Students often struggle to properly structure their essays. ChatGPT can provide sample structures of essays targeting well-specified questions. Students should be encouraged to first do independent research and in their instructions to ChatGPT include a list of some of the topics they wish the essay to cover. The following prompt tends to work well:
Prompt 1: “Please provide an essay plan for an academic [insert field] essay entitled [“insert title”]; including, but not restricted to, the following topics: “[Topic 1; Topic 2; Topic 3]”.
This will provide students with a rudimentary but solid structure for their essays. Students can experiment by not including specified topics, which will often result in ideas and arguments they hadn’t previously considered, providing additional fuel for further research.
While some critics might take issue with ChatGPT playing this role, it is little different to students receiving guidance from the lecturers, which they are typically entitled to do.
Further, while these prompts provide fuel for further research, they do not do the research on behalf of the student.
To take advantage of these suggestions, the student must still go to the literature and enhance their understanding of the relevant issues.
Clarifying the student’s Ideas: The Socratic method
AI language models can be surprisingly good at debating with researchers or, alternatively, providing complete Socratic dialogues on one’s question of choice, that help students see both sides of an argument. Students should be encouraged to ask ChatGPT to adopt a particular position, and argue with them in the Socratic style.
Prompt 2: “Please adopt the [insert position opposite to that you’ll argue for] position, and argue with me on the [insert topic] in the Socratic style. I’ll start. [Insert first argument].
For example, “Please adopt the position of the theist, and argue with me on the problem of evil in the Socratic style. I’ll start. The existence of a benevolent, omnipotent, omniscient God is inconsistent with the existence of evil in the world.”
Students should continue to press ChatGPT on its responses, drawing on arguments they have found in the literature, and taking careful note of its responses. This process should take place for all the arguments the writer wishes to make, and should help clarify their ideas, and provide possible responses to their arguments which can later be addressed.
AI systems as copy editors
Lecturers are often frustrated by the quality of many students’ writing. Their poor sentence structure, grammar, spelling and referencing can obfuscate what academics are really interested in — the quality of the student’s understanding and argumentation. Using ChatGPT as a copy editor can be a useful tool to overcome these difficulties, particularly in climates where underprivileged students are not writing in their first language and cannot afford the services of professional human copy editors.
Prompt 3: “Please make the following passage coherent, well-structured, and remove unnecessary words. It should be written in the style of academic [insert field], using the [insert preferred referencing system] referencing system: “[copy and paste passage]”.
Using the prompt above, students can copy and paste their work, a few paragraphs at a time (ChatGPT cannot deal with large quantities of text at a time), making their outputs far more accessible to their lecturers, and allowing them to assess the students’ understanding.
Using AI systems for feedback
Once a researcher or student has a final draft, they can use ChatGPT to provide useful feedback — the kind of feedback a lecturer would provide after assessment. Students can also ask for suggestions for improvement, and then incorporate these suggestions into their work. Lecturers typically have fairly large classes, and feedback can take a while to come back to students when they request it. Feedback from ChatGPT remains a less desirable option to that from the lecturer but it is free and instant.
Prompt 4: “Please provide feedback on the following section from my academic essay, entitled “[insert essay title]”: “[insert section].
Prompt 5: “Please suggest ways in which I can improve the following section from my academic essay, entitled “[insert essay title]: “[insert section]”
ChatGPT will provide feedback on the structure, content and argumentation, in much the same way as a lecturer would. These insights can usefully be addressed and incorporated into the essay prior to final submission.
It is worth issuing a number of warnings to students and academics wishing to make use of AI systems in their academic work. First, ChatGPT requires regular fact-checking. If you ask it about historical matters such as the career of a specified academic, much of what it says is false. Second, if you ask ChatGPT to include references, it makes them up. It will provide references to books and articles and even authors that do not exist.
ChatGPT cannot write good academic essays alone. Of course, submitting an essay entirely generated by ChatGPT would be a clear case of plagiarism (and there are AI systems that can detect the likelihood of a passage being generated by AI), but notably, these essays are not good essays. Whilst using ChatGPT in a responsible way can massively improve the quality of one’s research outputs, left to its own devices, it cannot yet generate high quality academic work.
*The views expressed in this article are that of the author/s and do not necessarily reflect that of the University of Johannesburg.