Generative AI Detectors
Generative AI: Detection
As faculty we are grappling with the disruptions of generative AI, and many of us wish there were a reliable way to ensure our students are not overusing AI. However, automated AI detection in its current form comes with significant flaws Links to an external site.:
- Inequitable results: Detectors are unreliable, Links to an external site. as TurnItIn acknowledges. Experts in the field suggest that AI plagiarism detection predominantly disadvantages minoritized students; as well, there is documented bias against non-native English writers Links to an external site..
- Problematic Triggers: Detectors are triggered by commonly and widely used apps like Grammarly Links to an external site. and Google translate; they can also be triggered simply by using formulaic writing styles.
- Easily Circumvented: They are easily circumvented by text spinners and rewriting apps Links to an external site., so will only “catch” the students who aren’t privy to or able to afford secondary apps.
- Automated Ethics: The price of attempting to automate ethics Links to an external site. is high.
For these reasons, many colleges and universities have decided not to endorse Links to an external site. the use of AI detection services (SDSU included).
We all want to ensure our students’ work is authentic, and some of the strategies for doing so involve implementing the good practices we’re likely familiar with: assignment redesigns Links to an external site., rubrics that prioritize authenticity and critical thinking, discussing the strengths and limitations of AI with our students, prioritizing process over product, student-centered options to demonstrate understanding of key concepts, non-punitive late-work policies, creative projects, multi-media presentations, and so on.
External Detection Tools
External AI Checker Disclaimer
For those who plan to use external sites to identify AI-generated writing:
Firstly, please keep in mind that student privacy must be paramount; to avoid violating FERPA, no personally identifiable information can be input into detection sites.
Disclaimer Suggestions for Disclosure of AI Detection Tools, If Used:
- Transparent Communication: Just like students need to click a checkbox to consent to it every time they upload an assignment to TurnItIn in Canvas, if you are using any AI detection tools, please communicate this transparently to your students. This includes detailing what application will be used, its purpose, and how it works.
- Data Handling Information: Clearly outline what student work will be uploaded, to which platform, and how their data will be protected. This is necessary for maintaining trust and respecting students' privacy rights.
- Rationale for Use: Explain the reason behind using such software. Emphasize that it is a tool to assist in maintaining academic integrity, and why you believe it is important.
- Student Rights and Options: Inform students of their rights regarding their work and data. If there are options for students who are uncomfortable with their work being uploaded to a particular platform, these should be clearly stated.
If used, experts state AI detectors should not be the sole basis for judging the originality of a student's work, and instead be the springboard to a meaningful conversation. Adjusting to the new reality of generative AI is challenging; it can also be an opportunity to critically evaluate a new technology alongside our students as we demonstrate the kind of curiosity and lifelong learning we all try to pass along to our students.
Plagiarism/Academic Integrity
- Generative AI, Plagiarism, and Cheating Links to an external site. by Leon Furze examines the complexities of using Generative AI in academic settings, highlighting the limitations of detection software and the inadequacy of some common approaches.
- For a script to talk to students about potentially AI-written text, check out the last page of A Toolkit for Addressing AI Plagiarism in the Classroom Links to an external site.
- SDSU's AI citation page for students Links to an external site.