10 Best AI Detection Tools for Teachers & Professors
October 13, 2023
As a writing coach and university teacher, I’ve been puzzling through the pickle that is AI-generated writing and AI detection for teachers. To be honest, I’ve dragged my feet in getting familiar with ChatGPT. I love writing. I like editing, too. Further, I believe there’s something inherently inimitable, even mystical, about expressing the human condition through words. So why would I want to get involved with a chat box programmed to do the exact opposite? To imitate human language, without any experience in what it means to be alive?
Well, as a lifelong bookworm, I’m well aware of the risks of living like a misanthrope. (“Misanthrope,” according to Merriam-Webster, is “a person who hates or distrusts humankind.”) Novels like The Catcher in the Rye and Notes from the Underground and Crime and Punishment have warned me about the futility of attempting to live as a recluse for the sake of principles. Humankind has invented, and in many cases, endorsed AI. Therefore I, choosing to partake in society, must learn to live with it. (Though this doesn’t mean I must trust or embrace it.)
I mention this personal perspective on purpose, to prove that someone (me!) is writing this article. As far as I’ve seen, ChatGPT has yet to develop deeply convincing fictional personages and pass them off as real. Nor does it develop a nuanced, complex position 100% of the time. You can find cracks in its argumentation when you look for proof of critical thinking. And yet, AI-generated writing will only continue to improve. Meanwhile, students who dislike or struggle with writing are turning to AI to generate everything from emails to academic papers. What should educators do?
Humans, the First AI Detection Tool
The first step towards AI detection involves putting thoughtful parameters in place in the classroom and around assignments. Including in-class writing assignments allow teachers to get an initial glimpse of their students’ writing levels. Teachers may want to note down specific strengths and weaknesses. Then, if a student turns in a typed assignment that doesn’t align with their handwritten work, their teacher will notice.
Another technique for detecting AI-generated writing involves selecting contemporary, lesser-known course materials. If the course involves Shakespeare, you can ask your students to compare the canonical text with a lesser-known one. (ChatGPT will likely have trouble with this.) You can also create extra hurdles that a chat box will have trouble jumping over. For example, include an aspect of personal reflection.
When I sat down to write this article, I decided it was time to know my enemy. I asked ChatGPT to write a poem about AI detection in the style of Christina Rosetti. And, I asked, please use the word “pickle.” The poem began admirably, I admit. Yet by the second stanza, it stopped making sense. The “author” began encouraging students to use AI detection in their “perplexed quest.” This I found perplexing. Next, the bot randomly inserted a briny cucumber into “a sea of text.” A student, on the other hand, might think of using this word more relevantly. (A “pickle,” according to Merriam-Webster, can also be “a difficult situation.”) Finally, when I asked ChatGPT to write this poem in the style of Anne Carson, the bot seemed flummoxed. It continued parodying Rosetti’s style.
AI Detection Tools, At Your Service
At some point, however, human detection will malfunction. Many students, even at the college level, have trouble articulating their ideas, using textual evidence, and building transitions. These writing flaws can resemble those of AI-generated writing. Furthermore, students writing in a second language may use appropriate online translation tools (neural machine translations and online dictionaries). These tools can suggest vocabulary and syntax that in turn work to produce writing which appears to be AI-generated.
When in doubt, teachers should consider putting their students’ writing through an AI detection tool. Below, I’ve put together a list of the best AI detection tools for teachers and professors. This list attempts to speak in layman’s terms, and considers the tools’ purposes, prices, and appeal.
1) Copyleaks AI Content Detector
In a recent study published on Cornell University’s open-access archive arXiv, four researchers named the Copyleaks AI Content Detector the “most accurate” in detecting text generated by Large Language Models (LLM). This AI detection tool boasts a 99.1% accuracy in detecting AI-generated source code from GPT-4, Bard, GitHub Copilot, and other bots. It can detect AI content in over 15 languages, and shows users which sections were likely AI-generated, through highlighting. It can even detect content that has been paraphrased from AI. While the basic model is free and user-friendly, it is also limited. A monthly fee of $9.16 includes access to the Copyleaks Plagiarism Detector as well.
Crossplag conceives of itself as a holistic tool that can “democratize plagiarism checking” through affordable prices. Indeed, the service offered to educators promises “usage-based pricing starting from $1 per user annually.” Texts are evaluated and scored with a confidence percentage, estimating the probability that the text was generated by AI.
Famously developed by a Princeton University computer science undergrad, GPTZero boasts being the “best and most reliable AI detector.” This claim comes from testing results from multiple independent sources, including TechCrunch. What makes GPTZero so reliable? Not only does the AI detection tool detect ChatGPT and other bots, it also looks for AI influence on the sentence, paragraph, and document level. It specializes in English texts, and is free for users who make an account. Lastly, it promises to interpret results for you through a description rather than a simple number.
Best AI Detection for Teachers and Professors (Continued)
4) Hugging Face OpenAI Detector
Despite its dubious name, the Hugging Face OpenAI Detector is a legitimate open access AI detection software. It’s free and accessible to all. Though it cannot currently detect ChatGPT-3 or GPT-3.5, it does use the GPT-2 output detection model, and will likely evolve. Because of this shortcoming, it serves as a reinforcement tool alongside other AI detectors.
5) Giant Language Model Test Room
Like the last detector, Giant Language Model Test Room (GLTR) can currently detect GPT-2. GLTR was designed by the MIT-IBM Watson AI Lab and Harvard Natural Language Processing Group. This language model attempts to predict the next work in a sequence, and thereby determine if a sentence “too likely.” If so, it may be AI-generated.
Pasting the last two sentences I just wrote into the GLTR demo tells me that 48% of my words were in the top 10 most-likely predicted words to follow the word behind it. Meanwhile, 15% of my words were in the top 100 most-likely predicted words to follow the word behind it. 19% of my words were in the top 1000 most-likely predicted—but I think you know how this sentence ends. Though the demo is not especially user-friendly, it is free and educational for those trying to understand the mechanics behind AI detection.
6) OpenAI’s AI Text Classifier
The maker of ChatGPT, OpenAI, put out their own AI Text Classifier back in January 2023. It was free, and offered five different responses to a potentially AI-generated text (very unlikely, unlikely, unclear, possibly, and likely). However, in July, OpenAI withdrew the tool due to “its low rate of accuracy.” Why is it still on my list? It will likely reappear on the scene. Its makers are “working to incorporate feedback and are currently researching more effective provenance techniques for text.” Furthermore, it’s important to point out that many, if not all of these AI detection tools should come with a warning clause. None are fully reliable, nor should any be used as the sole decision-making tool. Detection, analysis, intervention, and discipline must begin and end in the classroom, with equipped educators.
Best AI Detection for Teachers and Professors (Continued)
7) Passed.AI Detection & Plagiarism
This AI detection tool is geared towards educators. It was adapted from Originality.AI’s detector, which was built for content marketers. Passed.AI’s AI Detection & Plagiarism model has been trained on GPT-J, GPT-NEO and GPT-3. It doesn’t require installing software, and once the text is pasted into the scanning box, a percentage is obtained. This score indicates the likelihood that the text was generated by AI. A Google Chrome extension provides additional checks to further verify the score. Signing up gives users a 5-day free trial, followed by a monthly subscription for $9.99.
8) Turnitin AI Detection
Turnitin recently came out with an AI writing detector, an AI detection tool accessible for institutions. For this reason, the tool is not easily accessible, requiring consultation with a sales representative. Prices remain undisclosed, and individual licenses are not available to purchase. The product works in tandem with other Turnitin programs, such as Turnitin Originality. The software will break down a text into segments and rate the sections between 0 (human-produced) and 1 (AI-generated).
9) Winston AI Detection
Winston AI Detection calls themselves the “most trusted” among industry-leading AI content detection tools. It was designed for writers (including students), educators, and web publishers, and boasts a 99.6% accuracy rate. For $14 a month, users can try to ascertain if texts were written by ChatGPT-4. The tool offers a percentage of the likelihood that the text was human-produced, and also checks for plagiarism. While the tool currently only checks English and French texts, AI detection in more languages is on its way.
Best AI Detection for Teachers and Professors (Continued)
10) Writer AI Content Detection
Writer offers things like AI-generated writing templates and guides to an AI future. Yet it also provides an AI content detector. The free detector will check anything under 1,500 characters and can detect ChatGPT, as well as its offshoots (GPT-2, GPT-3, and GPT3.5). For educators, it may be helpful when a small portion of text, say, a paragraph, looks suspicious.
Following Up After AI Detection
As I suggested previously, all of these AI detection tools should be taken with a full spoon of salt. Every AI detector comes with limitations. The irony of relying on AI to detect AI should not be overlooked. Given the advanced knowledge these tools require to understand their operations, users can’t expect to fully comprehend their results.
After discovering the likelihood that your student has used AI, I recommend inviting the student for a chat. Encourage honesty. You might tell them that the work “appears” to have been produced by AI. Or, you can simply say it doesn’t reflect their previous work. They may confess to having used AI then and there. If not, you might ask them to describe their writing process and demonstrate their knowledge on the subject. Did they produce various drafts they can show you? Did they receive peer edits?
Finally, emphasize the fact that school is an environment which aims to foster academic and personal growth. Using AI may help a student understand how the technology works. Yet the student will miss out on all the mental exercises that lead to growth. I’m talking about the way a brain collects, processes, and articulates information. Nor does AI foster individuality. Personally, I’d rather have my students turn in papers that show evidence of human error than a perfect and perfectly bland paper synthesizing average ideas. In this age of AI, I think educational institutions will place a new emphasis on the value of the creative process.
If you are a student interested in studying please see our list of the Best Colleges for AI.