Yes, false positives can happen with AI detectors. Not only can they happen, but we know—based on convincing evidence—they do happen. When false positives arise during the evaluation of university students’ work, the consequences of those false positives can be dire. Typically, the falsely accused student bears the brunt of the technology’s error.
The first step is recognizing that AI detectors raise false positives. The second step is protecting a student from being victimized by such false positives. The LLF National Law Firm Student Defense Team is well known for preventing unjust sanctions against students who are wrongly accused of AI-related misconduct.
By the way, we also fight for students who use AI in a prohibited manner. Such mistakes should not define you. Without an effective advocate, though, harsh discipline can define the course of your life. We want to make sure that doesn’t happen.
Call the LLF National Law Firm Student Defense Team today at 888-535-3686 or contact us online to discuss how we can help you or your student overcome any (fair or foul) accusation of AI-related misconduct.
What About AI Detectors Cause Them to Make Incorrect Judgments?
Some AI detectors evaluate metadata within text. However, much of the way these detectors function relies on pattern recognition. Pattern-spotting is an imperfect science, and false positives might arise from:
-
A human’s writing style coincidentally mirroring AI’s writing patterns: AI detectors look at the repeated use of certain words, phrases, and punctuation as potential signs of AI-generated content. However, some of AI’s tendencies (like saying “However” or using dashes) can also be tendencies of human writers.
-
A human writer using AI-powered tools in their editing process: Many editing tools permitted in collegiate settings, like Grammarly, are implementing AI in their products. As students accept the recommendations of these AI proofreaders, the work may be increasingly likely to trigger the AI-detecting software.
-
A small sample size: The smaller the sample size an AI detector evaluates, the less reliable the detector’s determinations tend to be.
-
Humans’ writing becoming more and more like AI as time passes: As students work closely with AI tools over time, they cannot help but absorb some of AI’s writing tendencies by osmosis. Conversely, as AI continues to improve and write more like humans, the ability to differentiate human writing from AI-generated content should become increasingly difficult.
While those who sell AI detectors imply (or state outright) that their products are highly technical, precise investigative machines, close examination reveals a healthy dose of guesswork.
How Common Are False Positives with AI Detectors?
Different AI detectors admit to different false-positive rates. While Turnitin claims a false-positive rate below 1%, that rate rises to 4% when evaluating a single sentence.
We can expect such statistics to be skewed in the manufacturer’s favor. When we evaluate independent studies (which are likely to be more reliable), we find:
-
About 9% of human-written text is falsely classified as AI-generated by AI detectors
-
Industry leaders like OpenAI claiming that AI detectors are not “reliable enough, given that educators could be making judgments about students with potentially lasting consequences”
-
A plethora of other stats and expert opinions suggesting the typical AI detector is not reliable, and certainly not reliable enough to base disciplinary decisions on
The data leaves no doubt: AI detectors put students at risk of false allegations of misconduct and the harm that too often follows such allegations.
Even These Products’ Manufacturers and Marketers Admit Their Imperfection
When you read marketing materials from those who sell AI detectors, like Turnitin, you encounter seemingly contradictory assertions. On the very same page, you will read that:
-
The platform (in this case, Turnitin) has a false-positive rate of less than 1%
-
If the Turnitin AI detector says that the writing has been AI-generated, the user can be “very sure” that the writing has been AI-generated
-
At the same time, “there is still a small risk of false positives” when using the Turnitin AI detector
We cannot be 100% certain that Turnitin’s promise of a false-positive rate of less than 1% is accurate. It is undoubtedly in the company’s interest to make the product appear as precise as possible.
The Turnitin website concedes that false positives by its platform can result in students being “falsely accused” of misconduct. Because of the admitted potential for false positives, false accusations are inevitable.
Can Someone, Like a Professor, Know That an AI Detector Has Raised a False Positive?
The typical AI detector shows the user which sentences are being flagged as potentially AI-generated. It may disclose other information as well, but professors and administrators have limited capacity to evaluate the accuracy of the AI detector’s findings.
In certain instances, a professor may disregard the findings of an AI detector due to suspicions that the results are not accurate. However, it is just as likely that a professor or university administrator trusts the flawed results of an AI detector, unaware of its high inaccuracy rate (or how to interpret the software’s findings).
Does This Mean That a Student Can Be Unfairly Disciplined Based on an AI Detector’s False Positive?
Yes, a student can be sanctioned based on inaccurate findings from an AI detector. While many schools discourage professors from using AI detectors (due to their known flaws), others permit AI detector results as evidence of misconduct.
Some schools, like Wake Forest University, employ a sort of hybrid policy. Wake Forest’s approach is that “detector results may be included as evidence” in academic misconduct proceedings, but the detector results “are insufficient on their own.”
AI detectors spell trouble for university students, whether or not they engage in AI-related misconduct. Whether you’ve been unfairly flagged or an AI detector exposed a lapse in judgment (or a mistake you didn’t know you were committing), it’s time to mount your defense.
Sanctions against you now can resonate for years to come. Give everything you have to your defense by calling the LLF National Law Firm Student Defense Team today at 888-535-3686 or contacting us online as soon as possible.