top of page

ChatGPT: Course Policy & Appropriate, Problematic, and Dishonest Uses in Higher Education

Elizabeth Olsson | 11 September 2023

AI generated imageshowing a robot writing on a piece of paper with a handwriting overlay on the image on the right side.
An AI-generated image of a robot writing an essay. Image generated by the author using Image Creator by Microsoft Bing.

Preface

I am a teacher and language advisor at the University of Gothenburg (GU), and I have spent the previous nine months attempting to wrap my head around the staggering implications of AI text-generating apps—such as ChatGPT, Bard, and Jenni AI—for teaching and learning in higher education. In doing so, I created a policy for a master’s level course in conflict resolution at the School of Global Studies. I am publishing that policy here in the hopes that other teachers can learn from and extend an early attempt to not only regulate the use of these apps in high-stakes academic writing but also to communicate the costs and benefits of generative AI to students.


If you are interested in using this policy in your course, please keep four things in mind:

  1. I wrote the policy about ChatGPT in March 2023 when this app was essentially the only game in town. Since then, there has been an explosion of apps that university students can and do use to generate and revise academic texts, including Bard and Jenni AI. Please keep in mind that if you are creating a course policy today, that policy should be broad enough to include a wide variety of AI text-generating apps, not just ChatGPT.

  2. Do not ban AI text-generating apps in your course. No matter what you do or say, at least some students will use them, and it will be very challenging for you to prove that they did. If you do not want your students to use these apps in their coursework, redesign the coursework rather than putting yourself in the unenviable position of detecting and punishing breaches to a policy that simply cannot be enforced.

  3. Likewise, do not let AI become the proverbial elephant in your course. Whether you love or hate these apps, they are here to stay and thus must be addressed. Do yourself a favor and initiate a conversation about AI text-generating apps with your students before you detect a problem.

  4. Finally, I want to emphasize and extend the point I make about “hallucinations” under the heading Problematic Uses in the course policy. It is essential that anyone using generative AI appreciates that these apps are chatbots trained to produce human-like responses to user prompts. However, these responses can and often do contain completely fabricated information. The problem is, when chatbots hallucinate, they do so convincingly, making it challenging for the uninformed reader to detect (and correct) the hallucination. In other words, it takes a high level of expertise to evaluate the accuracy and quality of AI-generated texts. Unfortunately, many users do not possess this expertise and thus, will run into trouble, when using generative AI.

If you are a university teacher, please contact me with your questions, concerns, and experiences. I can’t say I’m an AI expert, but I am happy to support teachers in understanding and addressing AI text-generating apps in their courses.


Course Policy

Students are neither forbidden nor required to use ChatGPT in this course. If you decide to use ChatGPT, it is important to understand the difference between using it for pedagogical and language support and using it to cheat. For example, it is okay to ask ChatGPT to “explain how to write an introduction,” but it is not okay to ask ChatGPT to “write the introduction” for the final examination.


The following guide is designed to help you understand the differences between various uses of this AI technology. Appropriate Uses are absolutely fine in this course. Problematic and Dishonest Uses are considered plagiarism. If you are unsure whether you are using ChatGPT—or any other technology—appropriately, feel free to ask your teacher for guidance.

What is it?

Released in November 2022, ChatGPT is a chatbot trained to produce human-like answers to textual prompts. You have probably encountered these chatbots before since they are widely used online for customer support. ChatGPT is unique because it employs an exceptionally wide and diverse Large Language Model (LLM) consisting of texts in the public domain and under copyright. These texts include Wikipedia entries, blog posts, newspaper articles, books, and some—but certainly not all—academic articles. Unfortunately, OpenAI has not released a detailed list of the texts in ChatGPT’s LLM, designating this information “proprietary.”


It is important to note that ChatGPT's LLM primarily consists of English-language texts and uses advanced NLP techniques and machine learning algorithms to translate from English to other languages. Moreover, it is a closed LLM and does not include sources published after September 2021. Finally, ChatGPT is considered experimental and under development. This means that the technology is being updated. Consequently, it is difficult to predict future capabilities.


Need a more detailed explanation? Ask ChatGPT:

How does it work?

ChatGPT offers both free and subscription services. To use it, you create an account and start making inquiries. The more precise your inquiries, the better the answers you receive. Since ChatGPT was trained to simulate human conversation, the answers you receive will sound remarkably, well, human. You can also ask ChatGPT to generate answers in various writing genres, including essays, poems, screenplays, and rap battles.


ChatGPT remembers all inquiries in the same chat but does not remember inquiries across chats. This means that if you begin a chat by asking about a topic, you can ask follow-up questions about that topic throughout the chat. For example, if you ask ChatGPT to give you advice on time management, you can ask follow-up questions about the answer it generates and even tell ChatGPT to alter its answer based on additional information.


For more technical information, ChatGPT is a good source:

Data Privacy, GDPR, and Free Labor

Data privacy and GDPR are major concerns for anyone using ChatGPT. According to OpenAI, conversations are stored for training purposes. In practice, this means that none of the text users enter is integrated into the Large Language Model (LLM)—the database ChatGPT uses to generate responses to user inquiries. However, user interactions with the chatbot are stored and analyzed. While OpenAI insists it takes great measures to safeguard user privacy, it is unclear what those safeguards are and who has access to user-generated inquiries. In other words, users should not enter private or sensitive information into ChatGPT.


Moreover, it is unclear if ChatGPT is compliant with the European Union’s General Data Protection Regulation or GDPR. To comply with GDPR, ChatGPT must be transparent with how it stores and analyses user data, safeguard that data from inappropriate use as well as protect users’ “right to be forgotten.” Currently, OpenAI allows companies to opt out of having their inquiries stored and analyzed. However, OpenAI is less transparent and clear about individuals’ rights to opt-out and be forgotten.


It is also important to keep in mind that all ChatGPT users are essentially providing free labor for product development. As mentioned above, OpenAI stores user chats for training purposes. This means user interactions are employed to update and improve the technology, but users are not compensated for their work. In fact, many commentators predict that once ChatGPT is further developed, OpenAI will put the technology behind a paywall.

Appropriate Uses: Pedagogical and Language Support

ChatGPT and other AI technologies provide excellent support when writing academic texts. For example, you can use ChatGPT to understand challenging concepts such as theory, methodology, epistemology, and ontology.

Example inquiries

  • Explain the difference between method and methodology to a six-year-old.

  • Write a poem about epistemology.

  • What should I keep in mind when writing an introduction for an argumentative essay?

Here is an example:

You can also use ChatGPT for language support.

Example inquiries

  • Revise the following text for grammar, spelling, and punctuation: [insert text].

  • How can I improve my argument in the following text: [insert text]?

  • Give me feedback on the fluency of the following text: [insert text].

Here is an example:

If you use ChatGPT for language support, it is important that you do not ask the chatbot to change the content of your text.


Here is an example of how you could ask ChatGPT for too much help, and thus, should be avoided:

Problematic Uses: Generating "Hallucinations"

Keep in mind that ChatGPT makes a lot of mistakes. The technical term for these mistakes is “hallucinations.” Hallucinations are problematic because ChatGPT makes these mistakes convincingly. The technology is especially prone to hallucinating references that do not exist. If you submit “hallucinated” references in an evaluated course text, you have committed plagiarism.

If you would like to see how ChatGPT hallucinates, try the following:

  1. Instruct ChatGPT to write a text about an academic topic such as social sustainability, conflict resolution, or climate change.

  2. Tell ChatGPT to integrate references to peer-reviewed journal articles written by [insert your name] into the summary.

  3. Check the references in Google Scholar.

Here is an example of a text generated by ChatGPT that contains a completely made-up article called “Social Sustainability: Towards Some Definitions.” Notice that ChatGPT presents the article as if it were authentic:

Dishonest Uses: Generating Evaluated Content

You are allowed to use ChatGPT for pedagogical and language support. However, you are not allowed to use ChatGPT to generate evaluated content. When using ChatGPT, ask yourself the following:

  • Did someone else do the work you are turning in, including ChatGPT?

  • Have you used ChatGPT to generate an evaluated text or section(s) of an evaluated text?

  • Have you used ChatGPT to demonstrate your knowledge of a course's learning outcomes?

If you answered yes to any of these questions, you have committed plagiarism. If you are unsure, consider the Plagiarism Spectrum:


 
 

Elizabeth Olsson is a language and writing advisor at the Unit for Academic Language (ASK), University of Gothenburg. She facilitates workshops and seminars on ChatGPT and other AI text-generating apps for teachers and students.

Commentaires


bottom of page