When the Algorithm Is Wrong: Defending Against False AI Detection Reports
| | Student Rights
It’s ironic that professors who believe students have misused technology to complete coursework often themselves misuse technology to support their accusations. But it is somewhat understandable. Professors who are pressed for time and nervous about the proliferation of artificial intelligence find it easy to rely on AI detection programs to determine whether students are cheating.
However, those programs frequently produce false results. When that happens, a student’s future is threatened. Accusations that impugn a student’s academic integrity can damage that student’s reputation in ways that can come back to haunt them decades down the line. It is important to take decisive steps to defend against accusations and ensure that the record is set straight.
As attorneys with decades of experience defending students in campus disciplinary proceedings all over the country, we know that taking a reasoned, logical approach is far more effective than reacting with fear and anger. However, we also know that when you’ve been accused of cheating, it’s pretty hard not to get upset. This is one reason it can be so helpful to work with a professional defense advocate when dealing with allegations of academic misconduct.
How AI Detectors Work
Traditional plagiarism checkers compare text submitted for testing with text from an enormous body of academic research, student work, published content, and other sources to identify sections that are similar or identical in wording. This analysis can identify the source of matching text and provide evidence that the text being tested previously existed and is therefore not original.
However, AI detection software cannot provide this definitive evidence. Instead, AI detection software looks for patterns, and when it finds those patterns, the information triggers certain assumptions. Specifically, programs such as GPTZero, Turnitin, and Copyleaks use algorithms to analyze the predictability of phrasing in the text (known as perplexity) and variation in sentence structure (known as burstiness). Patterns commonly used by large language models like ChatGPT often produce a sort of “robotic” consistency that these algorithms recognize as the “fingerprints” of generative AI.
Sometimes Humans Write Like Robots
One problem with AI detection software is that it fails to account for the fact that humans are creatures of habit and mimicry. We mimic the behaviors and patterns of those around us, and when we get in the habit of speaking and writing a certain way, we tend to stick with it until someone suggests we change.
This is particularly true in certain fields. In law, for instance, legal phrases often acquire meaning through court interpretation or by adoption in statutory law. Once a phrase has a legally-accepted meaning, others who want to be understood with certainty will repeat that exact phrasing, even if it is unwieldy and repetitive. These phrases often fit best into stilted speech patterns, so those, too, are copied. The result is that legal writing produced by students, professional academics, attorneys, and judges may sound very formulaic and predictable. It could be flagged as robotic even when it is not.
Similarly, writing in the medical field often follows a predictable template and uses accepted phrases with meanings understood across the field. So nursing and premed students may have their work flagged simply because they have become accustomed to the patterns of their chosen profession and have begun to mimic them. Writing in other technical fields can be flagged for the same reason.
AI Can Be Biased Against Certain Students
Studies conducted by numerous organizations have identified uncomfortable patterns in the AI detection results produced by these software programs. They often “report” that text produced by students who speak English as a second language is AI-generated. These false reports occur at a much higher rate than those of native English speakers. Often, students who learned English after becoming fluent in one or more other languages use formal syntax or a more simplified syntax than students who were raised speaking English, and this is flagged as artificial.
Students with neurodivergent thought processes also frequently have their work falsely reported as AI-generated. Because they think in patterns that differ from the “average,” these students often reason and use words in ways deemed artificial.
There are probably other ways AI detection programs are discriminating against certain student groups that have not yet been revealed.
Protecting Your Rights After an Accusation of AI Misuse
When AI detection software has flagged your work as the product of AI, and your professor has accepted that assessment, it is essential to protect yourself from false accusations. The process begins with gathering information and preserving evidence.
- Find out what the accusations entail. What parts of your text are determined to have been created with the improper use of AI? What detection software was used to produce this finding? Producing evidence of the false-positive rate can often refute the software’s findings.
- Take action to preserve evidence that you produced the work on your own. Save your search histories on your computer as well as notes, outlines, and earlier drafts of the work you turned in
- Examine the professor’s course policies and the school’s code of conduct to determine the restrictions on AI use. You can’t be accused of violating rules that didn’t exist at the time. Many rules and policies are vague about what is prohibited.
- Talk to an attorney about the most effective ways to advocate, depending on the situation. The approach taken in a formal academic disciplinary proceeding might be considerably different from the approach that works best if a professor has simply issued a failing grade without contacting administrators.
Even if the easiest course of action is to accept the result and move on, it is not wise to ignore a false accusation. It is important to put your objections and defense arguments on record so that if the incident is referred to later, the allegations are not left standing unchallenged.
When AI Detection Programs Give False Reports, Find Out How a Student Defense Lawyer Can Help
Professors do not like to be challenged by students or their parents. But when they make critical decisions based on false reports from AI detection software, it is essential to take action to bring the truth to light. There is so much riding on your academic reputation that it is vital to protect it at all costs.
Negative marks on the record can interfere with future educational opportunities and career goals. Accusations of cheating can also impact the way that others view you personally as well as professionally. That is why the team at Nesenoff & Miltenberg is dedicated to protecting students accused of academic misconduct. To learn about the assistance we can provide in informal proceedings, formal hearings, and court, we invite you to schedule a confidential consultation.