What should teachers look for to identify the use of AI in students' work?
BACKAs the semester is starting, lecturers are facing a big challenge: identifying the use of AI generated content among their students’ work. The LIS team has gathered here a few tips to support the lecturers who are not always very familiar with generative AI tools.
First, lecturers should start by familiarizing themselves with some of the existing generative AI tools, like ChatGPT, Bard, Bing, or Perplexity. By taking the time to test them and evaluate the outputs, they will easily be able to recognize the writing styles and the quality of these tools (accuracy, bias, depth, references if any, etc.).
Additionally, lecturers should familiarize themselves with the new or updated policies in place at GIHE (updated Academic Rules and Regulations, and Online Examination Rules and Regulations), and make sure to use the new Project Outline Template, as well as the new Assessment Cover Sheet accordingly.
Taking the time at the beginning of the semester or before the first assessment, to discuss the use of generative AI tool with the students, and explaining exactly what is permitted or not, as well as the consequences, can be beneficial and discourage a few students.
Now to detect the use of AI in a student’s work is another challenge. TurnitIn provides an AI score which indicates the percentage of human-written vs AI-written content but isn’t 100% accurate. It’s the same with all the other AI detectors available like ChatGPTzero, Originality.ai, Scribr, etc.
Here’s a list of what is typically missing from AI generated outputs:
- AI doesn’t cite its sources. When prompted AI will provide references, but experience has shown that these are fictive or wrong (wrong year, wrong publication, wrong author, etc.).
- AI isn’t a creative writer; it will use the same tone of voice and stay very generic. This can be noticeable due to the limited scope of research in the student’s paper.
- AI will use a higher level of vocabulary and doesn’t make spelling mistakes. Comparing the ‘suspicious’ work with previous work submitted by the student can already be of great help.
- AI cannot include personal experience and anecdotes to illustrate its writing.
- AI doesn’t fact-check. It is not always accurate about dates, people, and past events. It mixes them up sometimes, which makes it easy to detect AI written papers.
- AI-written text can lack logical progression and the student’s paper will seem like a patchwork of several pieces of work put together.
Iowa State University offers valuable insights on integrating Artificial Intelligence (AI) into education. They highlight its potential, limitations, and the need for responsible usage.
Among other recommendations and best practices, they provide the following guidelines to identify misconduct without AI tools:
- Enter the assignment prompt into generative AI-tools to identify sample responses and gauge the level of complexity and similarity of the created response to the concerning student work.
- Compare the submitted work to prior student assignments noting significant changes in tone and style of writing.
- Be mindful of AI “hallucinations” or “confabulations” in which generative-AI tools create realistic sounding information that is not accurate.
- Identify lack of citations, where these were required, or lack of accurate content matching citations provided. Some generative AI tools produce content that references “others,” “critics,” and “research,” without attribution.
- Identify references or topics that veer far outside of the class content or expectations for the assignment.
The Transparency in Learning and Teaching in Higher Education (TILT Higher Ed) project have put together several examples and resources to support faculty: https://tilthighered.com/tiltexamplesandresources
One of these resources is a Checklist for designing transparent assignments, providing questions to reflect on.
TILT uses the ‘Transparency Framework’ which includes Purpose, Task and Criteria. These will help students produce higher quality work as they won’t lose time trying to figure out exactly why, what, and how they should do the assignment:
Purpose includes:
What skills will the students practice from doing this assignment?
What knowledge will they gain from working on this assignment?
Task includes:
What do teachers expect students to do and how should they do it?
Criteria includes:
Checklist of expectations and examples of what a work would look like when meeting those expectations.
More information about the Transparent Methods is available here: https://tilthighered.com/transparency
References:
AI for Education (2023). Won’t my students use ai to cheat? https://www.aiforeducation.io/blog/deg5axlmx45rdwz7k7yxvj1tv4mtw2
Center for Innovative Teaching and Learning (2023, August 29). Hote to productively address AI-generated text in your classroom. Indiana University Bloomington. https://citl.indiana.edu/teaching-resources/academic-integrity/AI-Generated%20Text.html
Centre for Excellence in Learning and Teaching (n.d.). AI (Artificial Intelligence) in teaching. Iowa State University. https://www.celt.iastate.edu/resources/ai-teach-learn/
Jeanne, P. (2023, May 29). 12 ways to identify ai-generated text in student work. ILLUMINATION. https://medium.com/illumination/12-ways-to-identify-ai-generated-text-in-student-work-617d8950c2d6
MAWinkelmes (2017, March 14). Mary Ann Winkelmes: Transparency framework 1 – Purpose [Video]. YouTube. https://youtu.be/PxndXE5_qMc?si=Eohdzu57wuG-ds7h
MAWinkelmes (2017, March 14). Mary Ann Winkelmes: Transparency framework 2 – Task [Video]. YouTube. https://youtu.be/c_DMDRUSedU?si=mEGRbCVaBuIx-sMv
MAWinkelmes (2017, March 14). Mary Ann Winkelmes: Transparency framework 3 – Criteria [Video]. YouTube. https://youtu.be/DXjR9qbvzF8?si=aHwL0a9Nk-kYNZ4z
Transparency in Learning & Teaching project. (n.d.). Checklist for designing transparent assignments. TILT Higher Ed. https://tilthighered.com/assets/pdffiles/Checklist%20for%20Designing%20Transparent%20Assignments.pdf
Transparency in Learning & Teaching project. (n.d.). TILT Higher Ed: Examples and resources. TILT Higher Ed. https://tilthighered.com/tiltexamplesandresources
Transparency in Learning & Teaching project. (n.d.). Transparent methods. TILT Higher Ed. https://tilthighered.com/transparency