Policy

Why prohibit?

LLMs can improve education if used correctly, but using them in certain situations can lead to adverse outcomes. In an article written on March 13, 2023, titled “Chatting and cheating: Ensuring academic integrity in the era of ChatGPT,” Cotton et al. outline several negative aspects of ChatGPT ( Citation: Cotton, 2023 Cotton, D. (2023). Chatting and cheating: Ensuring academic integrity in the era of ChatGPT. ) .

• The goal of education is the challenge students so that they learn. Using AI to generate answers removes the challenge and leads to less learning.
• It is unfair to students who work hard to write an essay independently when their peers are using ChatGPT to generate the same material.
• Assessing how well a student understands the material in the form of an essay is hard to do when an AI-generated text can obtain a passing grade. This leads to a devaluation of degrees ( Citation: Cotton, 2023 Cotton, D. (2023). Chatting and cheating: Ensuring academic integrity in the era of ChatGPT. ) .

However, a blanket ban on GPT in education could cause more problems than it solves because it is hard to enforce. One problem an unenforceable ban on ChatGPT would cause is that the people who use ChatGPT would attempt to hide its use. This is a problem because it would be unfair to honest students in competitive situations. The other problem with banning ChatGPT is that students must learn to use the tools ethically. Therefore, a school needs access to AI tools to teach students about them.

Detection

OpenAI has released an AI Text Classifier ( Citation: OpenAI, OpenAI, O. (s.d.). Retrieved from https://platform.openai.com/ai-text-classifier ) . However, this technology has several limitations. In addition to this tool emitting both false positives and false negatives, there are several ways to avoid detection, including using ChatGPT ( Citation: Goggin & Tolentino, 2023 Goggin, B. & Tolentino, D. (2023). ChatGPT can help you fool OpenAI’s anti-cheating tool. ) . In general, detecting AI-generated text is related to knowing the limitations and weaknesses of the technologies available. Here are several strategies for detecting AI-generated text.

• Plagiarism Detection. The ( Citation: OpenAI, OpenAI, O. (s.d.). Retrieved from https://platform.openai.com/ai-text-classifier ) provided by OpenAI and ( Citation: Tian, Tian, E. (s.d.). Retrieved from https://gptzero.me/ ) are two free tools that can be used to detect AI-generated text.
• Check References. According to the Clare ( Citation: Williams, 2023 Williams, C. (2023). Hype, or the future of learning and teaching? 3 Limits to AI’s ability to write student essays. ) , the references ChatGPT often creates “collapses into a haphazard Harvard style.” Clare ( Citation: Williams, 2023 Williams, C. (2023). Hype, or the future of learning and teaching? 3 Limits to AI’s ability to write student essays. ) also mentioned a student that fabricated an entire essay combined with a bibliography. However, the references did not check out, and the student later confessed to using AI. The current versions of GPT fail to create sophisticated references to reputable sources, so requiring writing that involves research can help prevent the use of ChatGPT.
• Check for Hallucinations. An AI hallucination is when the AI is confident about a fact because it “seems about right” rather than verifying it.

Alternative assignments

Halaweh ( Citation: Halaweh, Halaweh, M. (n.d.). ChatGPT in education: Strategies for responsible implementation. ) , Atlas ( Citation: Atlas, 2023 Atlas, S. (2023). ChatGPT for Higher Education and Professional Development: A Guide to Conversational AI. ) and Clare ( Citation: Williams, 2023 Williams, C. (2023). Hype, or the future of learning and teaching? 3 Limits to AI’s ability to write student essays. ) all provide advice on how to create assignments that limit the possibility of cheating when GPT is banned.

• Use Current Events. ChatGPT 3.5 and 4 are limited because their data set is cut off at a particular date. This means that it becomes harder for a student to say, “Write me an essay about X,” where X is an event that happened in the past couple of years {{ cite “hype-or-future” }}. This is not a foolproof way to prevent students from using GPT in situations it is banned, but it can help.
• Require Presentations. Presenting research is a good way to ensure students learn the material rather than copy the output of an AI. While students can still use ChatGPT to generate a presentation, they at least have to copy the material into a PowerPoint and memorize the slides.

None of these ways prevent entirely students from abusing AI in situations it is banned, but they can help. It is important to note that these strategies are workarounds that target the current language models’ weaknesses. However, as these models improve, the strategies for preventing their use should also change.

References