A year ago, I believed I’d be the victim of AI. Asking my students to produce essays that indicated their learning progress seemed to be ruled out. They just prompt ChatGPT or whichever tool they prefer to generate text. A year back, the common opinion was that we – as teachers -can’t prove the provided essay was their own or AI-generated. Their word against my word.

Now, only one year later, I changed my mind. And I changed my teaching methods. First, I encourage my students to use AI (not to generate text, but to get a better handle on the problem I’m asking them to solve). Second, I still expect them to write their own opinion pieces, for example, on discussion boards.

Do they learn better? Yes, I think so. It’s like in (my) old days. Back then, we learned methods and approaches to better understand the problem and ideate some solutions, not mandatorily memorize specific formulas or stuff.

This semester, I’m using an AI TEXT DETECTION tool to check if the posts on the discussion board are human-generated. Here are my findings (and no, this is not a full-scale study):

20% of the posted texts were flagged as AI-generated. I used GPTZero as the AI Text Detection tool. Now, when you consider the single data points, a completely different picture appears: while 20 blogs were flagged as ‘100% AI-generated’, 98 blogs were categorized as ‘100% human-generated’ (N=118).

AI detects AI. This is what I learned over the last year. My students know they must get more creative to prevent an AI Detection Flag.

If you want to learn more about AI Text Detection Tools, please read on.

These software applications analyze any text and calculate the probability of whether it was written by a human or generated by an AI model. Like their AI text generation brothers and sisters, they use linguistic, statistical, and machine learning techniques.

During my ‘study,’ I used GPTZero. It was developed for teachers to detect AI-generated essays and assignments. GPTZero evaluates text based on how predictable the words are and how varied the sentence structure is. The result is a Confidence Score, a probability indicating that the text was Human or AI-generated.

I don’t want to promote GPTZero here. Search for “AI Text Detection Tools,” and you get a number of hits, including OpenAI’s AI Text Classifier, Turnitin AI Detection, Winston AI, and many more.

The catch: all tools want to make money, and typically, they offer a free tier to experiment with. But to analyze hundreds of texts from students, you might need a paid version.

What are the challenges of AI Detection Tools?

The tools are not 100% accurate. I don’t use the results to penalize students rather than initiate conversations. What can you do if they insist they wrote the blog themselves, but the tools calculate an 85% probability of AI-generated text?

What developed over the last year clearly shows that AI models are getting better; AI models are used in Generative AI and AI Text Detection tools.

And to give you a feeling for AI-generated text vs. Human-generated text, here comes an AI-generated “conclusion” that, for sure, lives in a generic la-la-land.

“Conclusion. AI text detection tools like GPTZero play a crucial role in distinguishing AI-generated content from human-written text. While they offer significant benefits in academia, publishing, and corporate settings, they also face challenges such as false positives and the rapid evolution of AI models. The future of AI detection will likely involve continuous advancements in machine learning, improved accuracy, and ethical considerations to balance innovation and authenticity.

As AI technology progresses, the importance of responsible AI usage and transparent content verification will only grow. Whether you are an educator, writer, or business professional, understanding AI detection tools can help you navigate the evolving digital landscape effectively.”

P.S. This blog (minus the conclusion) received a 3% AI-generated probability from GPTZero. GPTZero classified the Conclusion paragraph at the end with 95% confidence as AI-generated. In fact, it is 100% AI-generated by ChatGPT4o.