Forum: support

RE: Addressing AI-Assisted Cheating: Image-Based Question Text? [ Reply ] By: Achim Zeileis on 2024-11-19 02:46	[forum:50019]
Yes, I completely understand your reasoning. Personally, I'm always worried, though, that using ineffective security options might even encourage cheating and "tit for tat" thinking. Instead I usually try to open about the limitations of the system and explain the code of conduct. But it's a difficult situation and none of these options is ideal.

RE: Addressing AI-Assisted Cheating: Image-Based Question Text? [ Reply ]
By: Jeff Pisklak on 2024-11-19 02:34

[forum:50018]

Thanks for this - I wasn't aware of the tex2image() function!

I had already known that OCR would be a simple workaround for my image-based strategy, especially since OCR-capable apps are becoming increasingly common (a quick search for 'math' in Apple's App Store confirms this trend). Nonetheless, I figured it was still better than nothing, given the nonexistent security options currently available on Canvas.

RE: Addressing AI-Assisted Cheating: Image-Based Question Text? [ Reply ]
By: Achim Zeileis on 2024-11-19 01:10

[forum:50016]

It is not directly built-in but it is not difficult to create such questions using the tex2image() function provided in the exams package. See my answer in this discussion on StackOverflow:

https://stackoverflow.com/questions/74715438/prevent-copy-paste-of-code-chunks-in-r-exams/74716450#74716450

The main problem with that approach is that it will be very uneffective if students can anticipate it. Running such an image through optical character recognition (OCR) is very easy and effective as shown in one of the other answers in the StackOverflow discussion above.

Addressing AI-Assisted Cheating: Image-Based Question Text? [ Reply ]
By: Jeff Pisklak on 2024-11-19 00:23

[forum:50015]

With the growing prevalence of generative AI, it's now incredibly easy for students to copy and paste quiz questions into tools like ChatGPT and receive (often) accurate answers almost instantaneously. Because of this, strategies like reducing quiz time limits have become less effective in maintaining assessment integrity.

Currently, learning management systems like Canvas do not have built-in options to counter this issue in any respect. One potential workaround could be displaying quiz questions as images, with alt text provided for accessibility purposes. This approach could add a layer of difficulty to copying and pasting text directly into AI tools.

I'm curious if future versions of R exams could consider incorporating a feature that automatically converts question text into images for presentation. I don't have an in-depth understanding of the R/exams back-end and the complexities involved in implementing such functionality, but I wonder if there might be ways to automate this process efficiently within the framework.

I realize this may not be a high priority for package development, and I’m not sure how much interest or demand there is for this kind of feature. Still, I think it could be a valuable addition, so I wanted to share the idea for consideration.