AI Content Detectors to (Potentially) Assist Identifying Whether Text is Human or AI-generated

The face of a female-appearing android overlaid with microchip pattern over random, color-coded source code in the background

Looking for help to detect texts or parts of a text generated by GPT-3 or GPT-2 or another artificial intelligence model?

[Note that it’s extremely difficult to definitively determine whether a language model was used to generate a piece of text, and no tool is guaranteed to be 100% effective. However, they could be potentially useful to help you identify potential instances of language model-generated text.]

That said, here’s a short and ever-growing list:

DetectGPT
 
GPTZeroX
 
CatchGPT
 
Writer AI Content Detector
 
AI Text Classifier
 
OpenAI’s GPT2 Output Detector
 

How do these tools detect AI-generated content?

These tools generally:

  • Look for common linguistic features or patterns in machine-generated text. AI-generated text, for example, may be more repetitive or have a reduced degree of complexity and variability compared to text written by a human.
  • Check for specific formatting or structural features that are common in machine-generated text. AI-generated text might have a more uniform structure or lack the variety of formatting styles that is typical of human-written text.
  • Check for certain keywords or phrases that are commonly used in AI-generated text. That’s a lot harder already. In general, there are statistically significant patterns known and detectable of which word combinations a model like GPT3.5 picks.
  • Compare the content with known examples of machine-generated text to determine the likelihood that it was generated by a machine.

There’s still time to sign up for tomorrow’s Technology and Tacos lunch-and-learn session: 

Gray human brain situated against a background with mathematical concepts on the left (logic) and a tangle of colorful swirling lines on the right (creativity) Tuesday, Jan 31 @ 12 PM | Leyburm 119. A good defense is the best offense and grounding your teaching in good pedagogical practice can help to ensure that ChatGPT's disruption of our students' learning is minimal. Join Dr. Paul Hanstedt, Director of the Harte Center for Teaching and Learning, for a discussion of effective practices in the age of AI. Sign up at go.wlu.edu/techandtacos.