Pros and cons of AI detection

Should I use AI detection tools in my courses?

When we suspect that students are misusing generative AI to short-circuit their learning, our feelings of frustration can lead us to focus our energy and attention on catching them in the act. Many plagiarism checkers like Copyleaks have AI detection features, and those tools present themselves as a quick and easy solution to the problem. However, as is often the case, turning to technology to navigate our teaching can have unintended consequences. Before making any decision to use AI detection, it’s important to think carefully about the potential for harm. 

This guide will provide an overview of how AI detection tools work, how using them can negatively affect students and your relationship with them, and alternative ways to reduce misuse of AI by students. The final section of this article is for instructors who believe that the potential benefits of using AI detection outweigh the potential harms, providing suggestions for using these tools cautiously, transparently, and with students’ learning in mind.

How AI detection tools work

As you decide whether to use AI detection tools in your courses, it’s important first to know how they work. While there are differences in the specifics methods each tool uses, at their core they work in similar ways. First, they rely on an AI model that has been trained on a large body of text and a set of rules to help them analyze the properties of a text and determine whether text is more likely to have been AI-generated or human-generated. A recent Ars Technica article (Edwards, 2023) uses GPTZero as an example. That detector looks at a text’s “perplexity” (i.e., how commonly used the phrases in that text are) and “burstiness” (i.e., the amount of variation in sentence length and construction). If many of the sentences in a text are highly predictable (or have low perplexity) and are consistent or uniform in style, GPTZero will say that the text is very likely to have been generated by AI. 

As you read the description above, you may have noticed that the characteristics of a text that an AI detector will likely flag as AI-generated are similar to the writing of many novices in a field: full of commonly-used phrases and low in sentence style variation. This means that these tools are not likely to be very accurate in predicting whether undergraduate student writing was generated by a human or by AI: false positives may result because our developing students write in ways that are, in fact, similar to the way AI writes! In fact, many higher education institutions have turned off AI detection features in their plagiarism detection tools because of the high risk of false positive results and unfounded accusations of plagiarism. Moreover, while false positives are a potential problem across the board, they are particularly likely to affect English language learners.

The problem with focusing on “detection”

The reliability of AI detection tools is questionable at best, but that isn’t the only reason we need to be cautious about using them: we also need to be mindful of the message we send to students when we choose to use these tools. Choosing to surveil our students can create an environment of distrust, putting us in the role of police person and our students in the role of potential criminals. When students see us as their adversaries, they are unlikely to believe that we are there to support their learning. 

Alternatives to AI detection

Rather than putting our focus on catching students so they can be punished when they use AI inappropriately, it’s more productive to think of the ways that we can create the conditions in our course that will reduce students’ perception that they need to rely on AI to complete their work. This means giving thought to the way we design our assignments, the way we help students prepare for those assignments, and the ways that students recognize and reflect on their preparation. Please visit our resource How can you design assignments to that students use AI responsibly—or don’t use it at all? for a concrete set of steps to help you plan assignments that will reduce students’ perceived need to rely on AI to complete their work. 

Using AI detection to support student learning

If, after careful consideration of the drawbacks and alternatives listed above, you still believe that the potential benefits of using AI detection outweigh the potential harms, you should proceed with great care. This means being transparent with students, being critical of the results AI detection tools produce, and helping students reflect on their work and their learning. These three approaches are described below, along with some examples of how you might implement them.

Be transparent about your decision

Students have a right to know that you will be using AI detection tools. Your decision should be clearly laid out in any assignments where you will be submitting students’ work to any database and should be part of a conversation with students as well. Be clear about why you have made the decision to use AI detection and how you think it will support students’ learning. 

When you explain why you are using AI detection, emphasize that you’ve made this choice because you think it will support students’ learning in the course. Explain how they will be using what they learn from the reports they receive to help them make good decisions in their work. You may also explain that you see AI detection as a tool to help them resist the temptation to rely on AI in ways that will harm their learning. Remind them, also that you want to ensure their grades have meaning: you want to help protect the meaning and value of their program and their college experience. Finally, explain that learning to use AI effectively also means learning NOT to use AI when it will stand in the way of them developing important cognitive skills. Making good decisions about using AI is something they will be doing throughout their academic and professional lives, and you believe that using this tool will help them do that. 

Below are some examples of language you can use to communicate your decision to students.