CWE Generative AI Detector Statement

August 2024
While AI detector software might seem useful as a way to check for or prevent unpermitted students’ AI-use, they are problematic, in many of the same ways plagiarism detector software is problematic, especially in an educational setting.

Why are AI Detectors problematic?

  • Detection software is inherently flawed and can be easily fooled. There are many websites that claim they can “humanize AI writing” so they can’t be detected.
  • Even detection systems that claim to have been tested for accuracy can be flawed and evaded through other software designed to mediate the detection of AI language and phrasing.
  • Using detectors may violate students’ rights and sense of autonomy in classrooms.
  • AI detectors have accuracy issues:
    • They generate false positives and false negatives: they have high-error rates in detecting AI-generated text or identifying non-AI text
    • They give an illusion of accuracy: you cannot rely on lists of phrases or words to identify text generated by AI
    • It’s unclear how they are built: we don’t know the mathematical formulas that AI detectors use, so they shouldn’t be considered reliable
  • AI detectors can create social harm and bias:
    • False accusations: the potential for unfair punishments due to erroneous detections; there have been several high profile cases where students were wrongly accused of using AI.
    • Demographic bias: text from non-native speakers of English and minority groups can be disproportionately flagged as AI generated.

If AI Detectors are flawed, what is the alternative?

Assessment practices that rely solely on evaluation of a discrete finished product are particularly vulnerable to abuses of generative AI; therefore, working with students through the various stages of the writing process, and scaffolding longer projects, may reduce the perceived need to use detectors. In other words, the same strategies for discouraging plagiarism also work when we want to discourage the misuse of AI: design engaging classrooms, activities, and assignments; make it clear why you are asking students to do things, what they will learn from them, and how they will benefit; use best practices in writing assignment design, including building in attention to the writing process; and scaffold longer assignments into smaller chunks that are discussed and used in class. Asking students to do peer-response and revise their writing, meet with their instructors to discuss their writing, give oral presentations on their topic, and engage in written reflection about their writing are also not only best practices in writing instruction which help students learn content and become more effective writers, they also help discourage the misuse of AI.

Works Cited

Akben, Mustafa. ”Recent Advancements, AI Tools, and AI Detectors.” August 2, 2024. AI Assessment Workshop.

MLA-CCCC Joint Task Force on Writing and AI, “Generative AI and Policy Development: Guidance from the MLA-CCCC Task Force,” April 2024, https://aiandwriting.hcommons.org/working-paper-2/

Steere, Elizabeth, “The Trouble With AI Writing Detection,” Inside Higher Ed, October 18, 2023, https://www.insidehighered.com/opinion/career-advice/teaching/2023/10/18/faculty-should-know-tools-students-use-beat-ai-detection

Teo, Kai Xiang, “Even OpenAI’s Own Detection Service Can’t Tell AI-Generated Work Apart — The Company Quietly Took It Down Over Accuracy Concerns,” Yahoo! Tech, July 26, 2023. https://www.yahoo.com/tech/even-openais-own-detection-cant-093459778.html

Thompson, Stuart A., and Tiffany Hsu. “How Easy Is It to Fool A.I.-Detection Tools?” The New York Times, 28 June 2023, https://www.nytimes.com/interactive/2023/06/28/technology/ai-detection-midjourney-stable-diffusion-dalle.html