Whether you love or hate artificial intelligence, it is making its way into the heart of academic life. Some professors embrace it, while others have no-tolerance policies. One thing is certain: The lack of standardized institutional AI protocols at UC San Diego have left faculty and students to navigate a new technological space with little regulation.

So, what actually constitutes AI use in academia? It’s hard to avoid — a simple Google search yields an AI summary, and most people would rather read a few incorrect bullet points than click through three different links.

UCSD does not seem to have a concrete answer; rather, it leaves it up to the instructor to decide. The Academic Integrity Office and the Library advise students to ask their professors about the boundaries of AI in their classes and to consider whether the use case undermines the point of the assignment.

Where UCSD lacks a standard process for dealing with suspected AI use, the AIO fails to bridge the gap. Its limited guidance to instructors is even less clear on whether professors should use AI detection software. 

It is impossible for a professor to be 100 percent certain that a student has used AI; conversely, it is impossible to be 100 percent certain that a student didn’t. So, where do we go from here?

If an instructor decides a student has inappropriately used AI in their course, they can choose to meet with the student to discuss the alleged cheating. The instructor will then report that student to the AIO through an allegation form.

In the case of a false accusation, the student can choose to not accept responsibility for the alleged cheating, but is then subject to the Academic Integrity Review process. If found guilty, possible consequences include a warning, grade penalty, or suspension, depending on the student’s degree level and previous academic integrity violations.

While AI may make our lives easier in some ways, it complicates the relationship between student, professor, and work product. Without guidelines, it’s more difficult than ever to make a conscientious decision on prosecuting students for using AI. 

Hear from two members of the UCSD community who received accusations of AI use in their schoolwork and endured the punishments — even though they were innocent.

AI detectors do not hold a gavel. Professors should not treat them as such.

Written by Gabriel Lozano, Contributing Writer

As AI technology increasingly allows students to generate their work with little effort, academia has had to find a new way to flag issues of academic integrity: AI detectors. 

Seven weeks into a U.S. history course, my assignment was falsely flagged for AI use by Turnitin, a popular AI and plagiarism detector that disclaims that its detection is subject to false positives. Throughout my discussion with the professor regarding the accusation, a false positive was never in consideration — my professor assumed my guilt without further interrogation. 

AI detectors are not sufficient judges of academic integrity, and we must acknowledge their shortcomings — especially for already disadvantaged students. We must understand where and how they can go wrong in order to utilize AI detectors to their full extent. Until then, we cannot use this technology as a judge over human conversations. 

When it happened to me, I felt as though the effort I put into the assignment wasn’t worth the professor’s time. Worse, they treated the time I spent trying to prove that it was my work as a waste of their time. Misplaced trust in AI detectors props them up not as a tool, but as a replacement for conversations between students and professors.

AI technology is very new, and AI detection software is even newer. As of now, the software is not a reliable judge of generative AI use, but rather is meant to be grounds for meeting with students to discuss the possibility of AI use. GPTZero says itself that “No AI detector is 100% accurate, and AI itself is changing constantly. Results should not be used to punish or as the final verdict.” Despite these warnings, many professors continue to wield AI detectors with finality. 

False positives are a known flaw in AI detection models. When detecting generative AI, the most commonly used algorithm judges writing using two variables: “perplexity” — a measurement of how common a sequence of words is — and “burstiness,” based on the variation of sentence structure. The more similar each sentence’s structure is, the more likely it is for the detector to mark it as AI. 

The problem with this method of generative AI detection comes from the way it is trained. AI detectors measure perplexity using previously written papers in their database. Generative AI is trained in the same way, pulling from online text to generate its writing. What is common structuring and word sequence for AI-generated papers, then, is also common among writing generally. 

AI detectors especially have difficulty assessing writing by English language learners, flagging over half of the essays in a Stanford study as AI. This should be of particular concern in higher education, where non-native English speakers have historically been underrepresented. However useful AI detectors may be, they represent just another punitive barrier these students must cross in order to achieve academic success.

In the midst of a national literacy crisis, students who are struggling on written assignments are most at risk of having their work flagged as AI. Perplexity and burstiness are learned in formal education, meaning that students who already find reading and writing difficult will bear the brunt of false flagging. 

As much as generative AI shouldn’t replace students’ work, AI detection cannot be the replacement for conversations around academic integrity.

Stop the AI Witch Hunt

Written by Aldan Creo, a graduate student in Data Science at HDSI. His work focuses on the technical side of AI safety. He also writes about technology, education, and the human side of computer science.

I’m not perfect; in fact, I’m a perfectionist. That’s not good in a number of ways, but perhaps one I truly didn’t expect is the trouble it would bring me this quarter.

I got accused of “cheating” by doing my homework with AI.

Let me be clear: I don’t think there’s anything wrong with using LLMs. In fact, I’m a big supporter of integrating them into education. But what offended me the most was that I hadn’t actually used one. Academic integrity matters, and I take it seriously, which is why being misjudged felt so painful.

As a Spaniard, I’m familiar with something called the Spanish Inquisition. You see, we pioneered the art of accusing people of things that are impossible to prove. And while I was happy to think those days are over, what I see is that we’re all in that state of collective paranoia again. The AI inquisition is here.

The problem is, when there’s no way to prove whether someone used AI, people go with their gut feelings. And there’s nothing you can rationally do to defend yourself. So people start to avoid writing properly — goodbye em dashes — since that’s how ChatGPT writes now.

But I didn’t even use an em dash! So why did they accuse me? My homework seemed fine and the AI detectors didn’t flag anything. What hurt me most was what I heard in office hours: “Next time, write more like a human.” So what am I?

Then I realized: chatbots and I have more in common than I’d like… we’re both perfectionists. When writing, I’ve been taught to explain every single step in detail, and I like to include extra contextual notes and explanations. Unfortunately, when training LLMs, the answers we optimize for are those that are correct, detailed, and well-explained. Back home, that was the expectation; I needed to be hyper-detailed. Plus, I’ve spent years trying to improve my English, so I’d feel like I’m wasting that effort if I didn’t try to use rich vocabulary and complex sentences.

Yes, I may be a perfectionist and meticulous, but that doesn’t mean I’m not human! So, I went online to read about how AI (Academic Integrity) deals with AI (the one you’re thinking about) at UCSD. Turns out, as a grad student, such an accusation can get you suspended.

Wait… so if I get suspended, I will lose my scholarship, my visa, and I won’t be able to come back. And what after? My whole life goes out the window. Goodbye to the cute seals at La Jolla Cove, adios to the froyo shop, and sayonara to my dreams.

…then I gathered all these thoughts and submitted a regrade request: “please, stop the AI witch hunt.”