Student use of generative AI for homework assignments is a growing problem at all levels of education. Learning management systems and online sites have developed tools to help educators detect if a document was “Generated by AI.” However, anecdotal evidence has shown that these tools are not always accurate, and that student claims of only limited use of AI for editing may be correct. This presentation reports on a study conducted by the authors that examined 12 different AI editing prompts a student might use for editing a document. These prompts were given to five popular AI tools: Chat GPT, Claude, CoPilot, Gemini and Grok. We applied these prompts to sample narrative, argumentative, and informative essays from 6th, 9th and 12th grade students at different proficiency levels. We then examined the changes made by the AI and the results of four AI generation checkers: GPTZero, ZeroGPT, Origniality.ai, and Turnitin. We found that students who may have thought they were only asking for grammar correction may end up with large amounts of AI generated text, which is then flagged by the online tools as being AI generated. The results indicate that minor changes to the prompts can lead to unexpected large changes in the results from the AI detectors, to the detriment of the students who thought they were behaving correctly. We conclude this presentation with guidance for both students and educators on the best path forward, enabling fair and effective student use of AI for editing.

Authors:
Jim Alves-Foss, University of Idaho, United States
Baby Grace Llego, University of Idaho, United States

About the Presenter(s)
Dr. Jim Alves-Foss is currently a Distinguished Professor of Computer Science at the University of Idaho, focusing on Cybersecurity.

Connect on Linkedin
https://www.linkedin.com/in/jim-alves-foss

See this presentation on the full schedule – Saturday Schedule

Bookmark