Flagged
You are an assistant professor. Your university has run twelve student submissions through an AI detection tool. Each comes back with a probability score. You decide: flag or pass.
Some submissions are human. Some are AI. Some are somewhere in between. The detection tool is confident. The detection tool is wrong.
You can view each student’s file before deciding. You do not have to. Most people do not.
Flagged makes the argument that AI detection is pedagogically broken even if the technology were perfect, because every flag initiates a process that lands on a real person. The students most likely to be falsely flagged are non-native English speakers, neurodivergent students, and those using permitted assistive technology. The students most likely to evade detection are the most sophisticated users.
The alternative is not to do nothing. It is to design assessments that make the question irrelevant.
Built by Sam Illingworth as part of Slow AI.
Comments
Log in with itch.io to leave a comment.
I'm not sure I understand the situation being simulated here. One of the submissions is explanatory text from the student about how they used AI. Why would we "flag" that for malconduct, even if it's AI-generated? Or is it supposed to be from a student essay, despite the content?
Also, the "student profile" includes information such as "they ran the essay through an AI tool to make it sound more academic." How can we already know that? And if we already know the diction and grammar are from an LLM, what are we trying to decide? Whether the content is original to the student or not?
Thank you. The student profile is just to add additional context, and in many cases you would know that if you take the time to discuss the situation with a student.
In the final results, the text in the small grey box showed that I flagged at least one innocent student (my mistakes were passing 2 AI-generated texts) but on the image, it showed I had flagged 0 innocent students. I do think that flagging the ADHD case counts in the grey bex text.
Thank you. I will try and fix this now.
OK - so I just "learned" that if I go back, I can't change my response!! GREAT to know now! I didn't comment and needed to go back for that reason, and thought that this would "remember" my response from a few minutes ago?? Clearly it doesn't - so MY Comment for Marcus (Case 1 last time?) was here:
Unsure what I did last time – thought (BOTH Times!?) that his oral was better than his writing – so should have guessed that he would use AI for support?
Thank you Gail, I purposefully designed the game like that to help remind the player that oftentimes when these accusations iof malpractice are cast, there is no rewind button. :-(
Thank you for playing with such care, I greatly appreciate it.
Thanks Sam = new computer is arriving this morning, with computer guy to help set up - we had a HUGE thunerstorm last night here - almost right across Sydney! Will help with clean up here at our house - then get back to your game...
I'm keen to try to create something like you've done? Hope that is OK with you??
And will have comments and notes for each Case in your game, with my adaptations maybe - may take me longer, as I'm still learning this content - will see how I go! Also, I need to watch some of your webinars - which happen at 1am my time - please don't change their time - as I need to just catch up when I can? best always
Gail