We are an asynchronous program, and believe we are starting to see the use of bots running new quizzes.
- We know that for $5 a month you can pay for a Bot to run on a Chromebook and do a pretty decent job running New Quizzes.
- There are multiple types of Bots, some that use screen readers, others that take snapshots.
This being the case:
- In Moderate (New Quizzes) we have seen Junior is taking a 20 question within 8-10 minutes.
- Looking at time between answer submission is 2-5 seconds (typical of bot use).
- Depending on New Quiz Settings (one at a time or not) we see the typical start and restart.
None of the above are definitive other than its amazing how fast Junior can answer questions!
- Not feasible to use something like lockdown browser, due to the nature of our program and clientele.
We are looking into the CSVs for more concrete (not necessarily proof.. but evidence). Specifically we are looking for:
- IP Hopping which would indicate potential 2nd agent logged in accessing the quiz.
- Is this necessarily a red flag?
- Rapid API bursts.
- This could just be Canvas talking back and forth saving responses, and/or confirming connection with School District Domain & User?
- Suspicious User Agents
- What specific User Agents would we want to look for in a CSV.
When we have run student SCVs through an AI like Gemini or Claude (taking out student id info) we have had mixed results as to prompting if there is evidence of student cheating and/or bot use.
We have policies in place and best practices to address cheating/bot use. But we are looking for more definitive proof other than Junior took a 20-30 question New Quiz in under 10 minutes with a near perfect score (2-3 sec) time between responses.
My questions are:
- Anyone addressed BOT use?
- Successes or Failures in addressing it.
- Suggestions as to how to dig deeper into a CSV if possible?
- Other methods of BOT detection besides the 2-3 second between answer submission on a New Quiz?
- Suggestions as to promps to use when running a student access CSV through an AI, where it would omit normal API calls or other normal background processes?
Lastly I know that there is not a definitive smoking gun (like with plagiarism all checkers) which are 99.9999999999 effective, in detecting cheating & should not be used for punitive measures. But any advice would be appreciated.
