Beyond the gut feeling: Evidencing academic misconduct effectively

As we all become more familiar with artificial intelligence, views about appropriate use are also changing.  Learning and Teaching in the Age of AI, a new University resource, reminds us that ‘we are approaching AI not as a novelty or a threat, but as a tool to enhance the educational experience’.

Educators across the University are incorporating AI into a variety of teaching and assessment activities. This is an excellent opportunity to learn more about how your students are using AI. Resources are available via the Artificial Intelligence and Learning webpage to guide your advice to students, such as the AI in Assessment Guidelines, and to assist with thinking about AI in learning activity and assessment design. In addition, the Library has now released a comprehensive guide to citing and acknowledging AI.

Academic integrity in the changing AI landscape
Even though use of AI might be permitted, academic integrity and assurance of learning are crucial. It’s becoming increasingly complex to detect inappropriate uses of AI, or to understand exactly how AI was used. In contexts where AI use is permitted and encouraged for certain tasks, how do we know when students might have crossed the line?

While an initial concern about AI use might stem from a Turnitin AI score or a ‘feeling’ about AI-like writing style or grammar, this is not sufficient evidence, especially in the absence of other reasons for concern about misconduct.  If you are planning to lodge an academic integrity report relating to AI, you need to provide evidence from your expert perspective, and ideally multiple pieces of possible evidence. This might include:

• References and source materials. Many large language models (LLMs) generate fabricated references. Check if the references in the reference list are genuine. Do the cited scholars actually make the arguments/quoted statements that are being attributed to them? Are there no references at all?
• Consistency with assessment guidelines. Does the assignment answer the question, use expected sources or concepts from the course? Are there irrelevant or incorrect concepts, off topic arguments, or use of inappropriate case studies (e.g. using American examples when the course is explicitly about Australian issues).
• Document metadata. Does the document authorship information suggest that somebody else helped to prepare the assignment? Look for the author name and the ‘last modified by’ name if the submission is a Microsoft Word document.
• Editing time/unusually fast completion time. Does the document metadata show very short editing time? Remember that this could be due to the student creating a ‘save as’ final version of their assignment or using legitimate cloud-based writing tools. If the submission is a quiz or other timed assessment, is the completion time improbably short?
• Turnitin Similarity: Are there any Turnitin Similarity matches which are cause for concern, indicating possible plagiarism or copying?
• Student engagement. Do MyUni Learning Analytics show that the student has not engaged with course materials?
• Answers available online. Check known ‘homework help’ websites for your assignment question. Is the answer provided by the student available online via contract cheating websites?
• If you use Cadmus, does the report which may show the writing process for the final submission, raise any flags?
• Style. Is the writing style consistent with the student’s previous work, and your experience of their participation and communication? Inclusions such as odd uses of commas or frequent uses of dashes are not sufficient in this context.
• Formatting. Does the font size or style change for sections of the written work? This can be an indicator of text copied from elsewhere.

For detailed information in interpreting Turnitin Similarity reports, including advice on what text matches and the AI score mean, watch this recent ADEPT recording.

Make A Report
While these pieces of information individually may not be sufficient proof of misconduct, collectively they give the Academic Integrity Officer avenues for investigation and discussion with the student, to ascertain whether they completed the work themselves or whether a breach of the Academic Integrity Policy may have occurred. This can lead to an educative discussion about the importance of original work. The pointers above are useful for any suspected academic misconduct, not only for investigations involving generative AI.

Provide as much information as possible when making your academic integrity report.