ChatGPT

[ian10]

So with the recent release to the public of ChatGPT, not sure what it is then have a look at  https://openai.com/blog/chatgpt/

At or college we have been discussing it and trying it out a bit to discover its capabilities.  The really concerning thing is that students could potentially start to use it to produce evidence for assignments, we discovered it will even write program code for you.

I noticed that Turnitin are currently working on a detector for it, see here https://www.turnitin.com/blog/sneak-preview-of-turnitins-ai-writing-and-chatgpt-detection-capability

But I was wondering, is there much of a difference between the full blown version and the tool bulit into Canvas.

Hoping that when Turnitin release the ChatGPT detector that it will be accessable via the Canvas API.

[ian10]

On the subject, one of my collegues came across GPTZero 

Tried it on a section of text generated by ChatGPT and was informed that the text was created by an AI.

Might not be the best solution but until Turnitin fully implement their solution it could be another tool for us educators to use to identify cheating.

[yeelinda]

I hope Canvas integrates a ChatGPT detector throughout the entire platform, not just Turnitin. I'm seeing students using AI for discussion posts and other assignments that aren't submitted to Turnitin.

[chriscas]

As much as I know some people want an "AI detector" of sorts, both me personally and the institution I work for would be very against the idea.  Unlike traditional plagiarism detection, where there is at least some hard evidence, like sentence X appears on website Y, AI detection is all algorithms and best guesses.  Even with a "low" number like 1% false positives, that would be over 200 false accusations against students for us (just using the number of submissions that go through Turnitin, which is a somewhat small percentage of overall work).  How do students argue against the accusation if they are innocent?  It is going to become a teacher said vs student said situation, which does not really benefit anyone in the end.

I'll add that I'm a Canvas admin, but if I was a faculty I would come down very hard against cheating/plagiarism/etc, but to do so with what is essentially guessing (and maybe very very good guessing) would not be something I'd do.

[Copyleaks]

Quick commercial:  AI content detection, including ChatGPT4, is accessible via our platform which integrates with Canvas.  As Yeelinda touches upon, we're seeing significant growth in the use of AI content across student populations worldwide, particularly high school students.

Feel free to try it out:  https://copyleaks.com/ai-content-detector

[andrew_ippolito]

what is your % success rate for detection?  Currently the 'cheat detection' software out there is terrible (close to 50:50 from what I've read).  Student forums are filling with complaints that they can't rewrite their essays and papers without it looking like it came from an AI, and I imagine lots of students are dealing with false accusations.

(granted, 90% of the alleged submissions are assuredly GPT-crafted and very obvious, I'm talking about those good-faith students within the mix).

Systems should be designed around the good-faith users, not the ones who take advantage.

[Copyleaks]

We aim to ensure our users feel confident with their scan results.

In our most recent round we tested over 20K human-written papers to warrant a confident false positive rate for the AI Content Detector. The outcome was .035%, the lowest false positive rate of any platform. 

Nevertheless, no matter how low a false positive score is, we encourage discussion around results. In education, sharing the results with students supports learning and imbues transparency. Accordingly we tell teachers to consider meeting every scan result with the belief of positive intent and let it be the gateway to a conversation that can lead to further education and empowerment for the student. 

[Guz_MPC]

@Copyleaks I hope you can help answer these three questions.

1. My main question is: Does your service support detecting AI in Canvas Discussions?
2. What about detecting this in other Canvas assignment areas (all across Canvas)?
3. If not, are there any plans for something like this? Will this be something we can see perhaps in Fall 2023/Spring 2024?