Skip to Main Content

Generative Artificial Intelligence

AI Tools and Resources

AI Pedagogy Project

AI Detection Tools [from Rutgers.edu]

The following two lists are used with permission of Rutgers.edu.

While tools for detecting AI-generated content exist (Turnitin, GPTZero, Copyleaks), such tools are not 100% effective and have been found to exhibit several problems. For instance, AI detectors:

  • Can be prone to produce false positives or negatives

  • Offer no explanation or supporting evidence for their scores

  • Can easily be evaded through the manipulation of prompts or rewriting outputs

  • Tend to be biased against non-native English writers

  • Do not guarantee privacy of any data submitted

Instead of relying on AI detection tools, it is recommended that instructors:

  • Implement clear policies and guidelines governing the use of AI in the classroom

  • Teach students how to use AI ethically and effectively and critically evaluate outputs

  • Create scaffolded assignments that encourage intrinsic motivation and focus more on process than product

  • Develop rubrics that emphasize critical thinking, problem solving, and applied knowledge rather than the memorization and summarization of content

Some AI detection tools, such as Winston AI Detector and Copyleaks AI Content Detector, boast having "over 99% accuracy" in detecting AI-generated text. It is important to recognize that this accuracy score can be misleading (aside from the fact that, with private data, these claims have no way to be verified).

For example, take the concept of a flu detector. Say that we have a group of people where 5% of them currently have the flu. If we had a "flu detector" that, no matter what, said that a person is healthy, then it would have 95% accuracy. Of course, this is not a useful "detector," but it does illustrate that accuracy alone is not a sufficient metric to show the reliability of a detection tool. To find more examples, look into accuracy scores and the costs of wrong classification in data science.

In the context of AI, we must be careful to not wrongly accuse students of utilizing GenAI. Here are some strategies to use in your classroom outside of relying on AI detecting tools:

  • Consider using in-class sessions for writing that your students turn in. This will show their habits, strengths, and weaknesses in their writing. This will help to illustrate whether their work seems to be their own in a longer writing assignment. This also helps in situations where, instead of GenAI, another person ghost writes the student's work.
  • Select lesser-known materials for your classroom, or ask your students to compare well-known works to them. GenAI tools will be much more reliable in writing about well-known titles than on a contemporary author, and comparison utilizes critical thinking skills which is where GenAI struggles.
  • Ask your students to turn in drafts of their work as they complete it. Not only will this make them better writers, but it also safeguards against known situations where utilizing proofreading software will make one's work be flagged as AI-detected. 
    • Alternatively, suggest for your students to write their documents in Google Docs, so that the full editing history is shown. This will ensure that if a student is mistakenly flagged, then they can have evidence that their work is legitimate.
  • To teach students the inaccuracies of AI, perhaps create an assignment where a student will generate an essay using GenAI and then fact check it. This will make it much less likely for students to use GenAI in their personal work.

Resources for Instructors

List includes recommended tools from rutgers.edu, harvard.edu, and ucla.edu.