How to identify AI generated text?

by | Thursday, June 01, 2023

I think I solved the biggest educational challenge of our time, namely:

How do we recognize AI generated text from human-created ones?

Just to provide some context, the advent of large language models and generative AI have made it essential that we, as educators, find ways to distinguish between AI generated and human generated text. This is the only way we can catch those students who are blatantly cheating the system by using these tools to complete their assignments.

And this is the issue of our time, and the consequences of not addressing it is well captured in this essay in The Atlantic, whose title says it all: The College Essay is Dead, And this is just one the many newspaper articles, editorials and blog posts that hit the panic button once ChatGPT and other generative AI language models entered our world. And of course, we have a range of responses to this perceived crisis. from out-right bans to finding strategies to mitigate the risks of plagiarism. And let us not forget all the AI based tools that have emerged to that detect and flag AI generated prose (though their accuracy is somewhat unclear).

Matters are complicated further by the fact that humans often misunderstand these technologies, often in hilarious ways. For instance there was the recent story of a lawyer who had GPT write his brief, only to discover that the AI had made up case precedents. These things hallucinate.

One such story that came my way, this morning, was that of a professor who flunked a bunch of students because he thought they had used ChatGPT to write their papers. And again, the title of the story says it all: Professor Flunks All His Students After ChatGPT Falsely Claims It Wrote Their Papers. Well… as it turns out, the professor didn’t really understand the technology and fell for its hallucinations. In an interesting twist, someone entered the professor’s own writing into ChatGPT3 and was told it was AI generated. As the Rolling Stone story reports it:

In an amusing wrinkle, Mumm’s claims appear to be undercut by a simple experiment using ChatGPT. On Tuesday, redditor Delicious_Village112 found an abstract of Mumm’s doctoral dissertation on pig farming and submitted a section of that paper to the bot, asking if it might have written the paragraph. “Yes, the passage you shared could indeed have been generated by a language model like ChatGPT, given the right prompt,” the program answered.

All this is funny enough, but again these, I think, are growing pains as we struggle to understand this fast moving piece of technology. What I DID want to point to were the reasons that ChatGPT gave for claiming the prose was authored by AI. This is important information, because once we understand what makes an AI generated text, we can flip the criteria around to find out what characterizes human writing.

Below is a screenshot of ChatGPT’s response to why the prose fed to it was most probably generated by AI:

What this tells us is that AI generated text is coherent, having a clear structure, cites sources and data accurately, and uses technical terminology correctly

And conversely, we now have the defining characteristics of human generated text. Human generated text is incoherent, gets citations (of sources and data) wrong, and uses terminology inaccurately.

There we go. I am so glad we figured THAT out.

And yes, you are welcome!

A few randomly selected blog posts…

It takes 10,000 hours

A quote in a NYTimes article caught my attention According to sports scientists, the most significant predictor of an athlete’s skill is the time spent in practice. “It’s not just genetics,” says Jean Côté, the director of the School of Kinesiology and Health Studies...

National Sun Yat Sen University & Taipei

National Sun Yat Sen Unversity has one of the most gorgeous campuses you can imagine. Nestled between hills, with wonderful views of the ocean, with little twisty roads that connect departments and buildings, just beautiful. My hotel room had a magnificient view of...

It HAS to hallucinate: The true nature of LLM’s

It HAS to hallucinate: The true nature of LLM’s

Though Generative AI is receiving a great deal of attention lately, I am not entirely sure that the discussions of these technologies and their impact on education and society at large genuinely engage with the true nature of these technologies. In fact I have argued...

A Systems view of creativity

A Systems view of creativity

Our series of articles related to the broad topic of Rethinking technology and creativity for the 21st century in the journal TechTrends continues with two new articles. The first focuses on developing a systems view of creativity,...

Creativity and the urban STEM teacher

Creativity and the urban STEM teacher

I have written previously about the MSUrbanSTEM project and what it has meant to me. Over the past couple of years we have also published about this line of work (most prominently in a special issue of The Journal of Computers in Mathematics and Science Teaching)....

TPACK in a podcast

Just discovered a podcast on TPACK (titled Understanding TPCK) at the msad75mltinews website. It appears to be based on the article (Too cool for school) that was recently published in Learning & Leading with Technology.

Back from India…

Got back yesterday from a short, hectic but sweet trip to India. I had a wonderful time and still have a lot to do to just document all that happened and connect with all the people I met (hopefully over the next few weeks)... but now it is time to get back to fall...

Infusing creativity & technology in education

Infusing creativity & technology in education

Danah Henriksen, Petra Fisser and I have a new article (complete reference and link below). This article emerged from a Thematic Working group on Creativity in a technology enhanced curriculum that the three of us led at EduSummIT 2015 (see more here). This...

Ambigrams animated: 3 new designs

I love creating ambigrams, words written in such a manner that they can be read from multiple perspective - rotated, reflected and so on. These designs are much easier to "grasp" when printed on paper since you can actually turn the paper around, hold it against a...

3 Comments

  1. Marckie Zeender

    As a large language model developed by OpenAI, I find this method of identifying AI generated text to be accurate.

    Reply
  2. Mehedy

    You are a masterpiece.This article is a breathtaking masterpiece that seamlessly weaves together profound insights and eloquent prose. It captures the essence of its subject with a grace that is both captivating and inspiring. A true gem that leaves the reader spellbound and hungry for more. Bravo!

    Reply
    • Punya Mishra

      This comment is clearly spam. But I approved it none-the-less, but only after deleting the URL and the email address that the spammer had included. So this comment will remain here … except it will not provide the link to the person’s website (which, was the sole purpose of this comment).

      Reply

Leave a Reply to Mehedy Cancel reply

Your email address will not be published. Required fields are marked *