Detecting Plagiarism: Safeguarding the Code

Jan 17, 2025

AI is an uninvited guest in many classrooms. Sometimes, I look longingly back on pre-AI reality when, I feel, my teaching instincts were more pure, certainly more reflexive. I’ve depended on instincts all along—until now.

What’s changed since Thanksgiving 2022 and this turkey named Sam Altman—he’s joining Musk, Bezos, and Zuckerberg on the dais to witness Trump’s ignauguration (sic)—debuted ChatGPT3.5? (Anyone planning to watch it?)

LLMs today mediate the space between teachers and learners even when they don’t. There is continuous doubt about the purity of student reading and writing in this rewired reality. In the competition for credit between the named and the nameless, the cited and the uncited, in this match it is advantage nameless and uncited. But the game has just begun.

A Brief History of Detection

Before the rise of surveillance tools like Turnitin, plagiarism detection was handled through manual methods. Teachers relied on their own expertise to identify copied content, a practice which was time-consuming and stressful.

In the late 1990s, Turnitin appeared, one of the first online plagiarism detection tools, using text-matching algorithms to compare classroom submissions against databases of academic papers and web content. Other early competitors, such as PlagScan and Copyscape, also provided basic digital plagiarism detection services. Eventually, TurnItIn became a feature in Learning Management Systems and a requirement before teachers considered a submission to be credit-worthy.

Simultaneously, there were services that recycled and resold essays. Essay banks offered pre-written essays for purchase or hire, often marketed as study aids but misused by bad actors for dishonest purposes. Rumor has it that fraternities and sororities maintained archives of past exams, essays, and assignments to help members prepare for courses.

Additionally, publishers created collections of classroom-tested essays1 intended for teaching purposes. As difficult as it is to believe, though these resources were provided as mentor texts, they occasionally circulated informally among students, and, slightly mangled in predictably imperfect ways, submitted.

A Makeover for TurnItIn

AI has turned the tables on the TurnItIn monopoly and on the essay resale market. Logically, it makes sense that TurnItIn would morph from its role as an academic chimney sweep into a one-stop full-service literacy mechanic2.

AI-based plagiarism detection tools in competition with TurnItIn are increasingly used because they also can identify not only exact copying but also more nuanced forms of plagiarism, such as paraphrasing or cross-language duplication. However, their reliability is not much good, either. False positives, difficulties in detecting hybrid or manipulated texts, and inconsistencies across different tools are real challenges. Some institutions caution against relying solely on AI results3.

Additionally, ethical concerns arise regarding privacy and the use of students' original work in training datasets, an ironic motive for increased activity in the resale essay market for AI training purposes. AI tools offer efficiency over traditional methods, perhaps of some small use, they must still be used in conjunction with good old fashioned teacher oversight to ensure fairness and accuracy in credit allocation.

How Bad Is It?

How bad can it get? Texts like the U.S. Constitution have been incorrectly flagged as AI-generated by some tools4, 5 . Additionally, intentional manipulation, such as paraphrasing with AI tools like QuillBot, can evade detection altogether6.

Moreover, inconsistencies across different detection tools highlight the lack of standardization in this technology. Studies have shown varying results between tools when analyzing the same text, raising questions about their reliability in practical scenarios7.

The use of AI detectors raises ethical concerns about student privacy and the potential misuse of students' work in training datasets for these tools. I’ve already pointed out the slippery slope: AI companies buy the best pure student essays and use them to calibrate AI detection algorithms.

AI tools have revolutionized plagiarism detection by offering faster and more nuanced analysis than traditional methods. It makes me chuckle to think of artificial plagiarism detection—there is something oxymoronic about it like jumbo shrimp.

Waiting for the Right Answer? Stop Right Here.

For now, from my perspective, these tools are best used as part of a broader strategy to uphold academic integrity rather than as standalone solutions. For me, the absolute failure of plagiarism detection brought on by artificial intelligence is a symptom of a flawed assumption in traditional education, an assumption about what makes people tick.

Which is worse: Students plagiarize? Or students have such weak motivation to do their own work that plagiarism is an attractive option? Or something else entirely?

The real crisis may not be that AI has broken our detection systems. AI exposes the fragility of an American institution that has served us, however imperfectly, longer than we have been alive. When artificial intelligence can both generate and detect the writing our children do, we're forced to confront uncomfortable truths about our educational paradigm.

Perhaps the solution isn't better detection technology, but a revision of how we engage students in their own learning. The question isn't whether AI belongs in our classrooms; it’s here, uninvited but undeniable. The question is how we transform our teaching to make real, engaged learning the norm and plagiarism irrelevant, unthinkable, rather than just harder to detect. A challenge for sure, but not impossible, not beyond our reach.

Learning to Read, Reading to Learn

Discussion about this post