I generally cringe when students try to cite listicles, but when in Rome I suppose; here are the top 5 reasons why faculty should not rely on AI detectors:
1. AI humanizers are and will continue to be a thing.
There is a bit of an arms race out there, but there are “humanizers” that will take AI written text and revise it so it is undetectable to AI detectors. Take a look at the main ChatGPT custom GPT page for writing to get a sense of where this is going:
Because of the presence of humanizers and because AI detectors are fallible even without humanizers being ubiquitous, common wisdom is that faculty can never know for sure if a piece of student work was written by an AI. This is why many institutions have turned off the AI detection function of Turnitin.
2. It is fine that these detectors are being turned off because AI detectors are often unnecessary anyway.
In the range of student work, there is work that is clearly not AI generated and work that clearly is. In most cases, and especially in freshman composition, teachers don’t need an AI detector to know the difference. The detector may help with the feeling of certainty, but in fact, for cases where students use AI without finesse, it’s pretty easy to be certain when a piece of text has been generated by AI. Take the following segment from a real student paper:
The impact of TikTok stretches far beyond mere entertainment. To me, TikTok has
become a virtual agora, a meeting place where cultures from around the world converge. It’s a breeding ground for new forms of digital expression, where creativity is unleashed and shared in ways that were unimaginable just a few years ago. As a user and observer, I’m continually amazed by the endless possibilities it presents, and I’m excited to see how it will continue to shape our culture and communication in the years to come.
My favorite part of this is the way the voice personalizes the concept of the virtual agora – “To me.” It is completely unnecessary to confirm with a detector whether the above paragraph was written by AI. Beyond that, the certainty that comes with the detector can be counterproductive anyway.
3. We need to be having conversations, not certainty.
This question of certainty is critical for how we approach students who are suspected of plagiarizing using AI. The main reason detectors have been turned off at many institutions is that faculty will use the perceived certainty of a positive result from an AI detector to fail students without recourse.
In my own teaching environment at a rural community college serving many first-generation non-traditional students, students need to be brought into the conversation when they make a mistake like using AI and should have the opportunity to correct and improve their behavior. The goal is learning, and if a student can appropriately accept responsibility for their mistakes and do the work they tried to use AI to avoid, they should be able to continue on their path towards their degree with minimal disruption.
Nationally, community college graduation rates are lower than we would like and as an institution we are focused on supporting our students on their pathways to graduation, doing as much we can to improve our student success and graduation rates. Many of my students live with significant chaos in their lives and are in or on the edge of poverty. It often only takes one significant disruption to be the difference between success and failure. An automatic 0 or failing grade in a class can be the thing that makes the student decide to stop showing up to class. I want to help students move from where they are forward, and having the judgment of an AI detector, regardless of the degree of certainty, is not helpful in this regard.
4. If the only way to determine whether or not a student is using AI is by using an AI detector, teachers probably have many students using AI without their knowledge.
There are so many ways to humanize and AI generated text, from the integrated humanizers pictured above, to dedicated apps, to the old-fashioned rewriting of the AI language in one’s own voice. (I call that old-fashioned because for years I have seen students try to plagiarize by using a summary of Wikipedia with slight vocabulary and syntax changes. An AI detector for sure won’t catch that, and it’s impossible to use a quick Google search to catch it as was the case with the Wikipedia revisions.)
And beyond these strategies, what about students who use AI to generate just one sentence in an assignment, or who use it to generate an outline and write the text themselves from the outline, or when students learn to train AI on their own voices and the AI writes in a faithful approximation of the student’s voice? There are myriad undetectable ways students can, are, and will be using AI inappropriately in the classroom. The presence of AI detectors masks this challenging reality.
5. AI detectors prevent teachers from addressing the real issue.
The false sense of security AI detectors provide allows us to live in a fantasy world where we won’t need to change our assessment methods to ensure students are not using AI to avoid the work they need to do to learn. We teachers didn’t ask for this new reality where cheating would become ubiquitous and undetectable, but since it is here, the only way we can know what our students are learning is through proctored exams. I would love to hear other ideas that are not hackable by all this new technology, but if we want to know what our students know and can do, we need an environment where the use of AI is completely off the table and in that environment, an AI detector is unnecessary and irrelevant.
Leave a Reply