Can version control protect students against allegations of ChatGPT use?

Question

There is a Guardian article about the dilemma when there has been possible use on generative AI to create student essays. It presents it as a hard problem for students to prove they really authored a document, and for educators to decide who to punish.

Version control is ubiquitous in the software world and can be used to demonstrate authorship in academic writing. I am most familiar with git, and a git repository of a document creation process, with commits on save and pushed to a university controlled server or github/bitbucket a few times would seem to provide ample evidence that the document was created by a human. Other options may be less refined but I know google docs has much if not all of the required functionality and I would assume that the cloud based version of Microsoft Office has something similar.

Would a student ensuring they had such evidence available protect them against such allegations? Would university rules or recommendations for students to do so help to combat cheating?

Answer 1

At my University (UK), while any final decision on plagiarism/academic dishonesty is made on a case-by-case basis after looking through all the facts and evidence, what you describe is usually the best form of evidence the students can provide.

(Typical cases in which this might help so far were when a student would work on a lab machine and leave their solutions open without logging out properly. Then, somebody else would come along and opportunistically copy their solution. Usually, this is about programming assignments, but it does not have to be).

A history of older file versions, with the timestamps, showing how your work was created, and especially if it doesn't otherwise contradict your narrative, would most likely be more than sufficient to disprove any academic offences allegations.

While I acknowledge the issue mentioned by Vladhagen, that ChatGPT could just as easily produce an early, 'draft' version, and several refined version, as it could the final submission, I would argue that this is more effort than most students attempting to use ChatGPT typically spend on their assessments. If all the five versions were updated at the same time (or with e.g. 30 mins between them), it would actually point to foul play: normally revising self-written text takes some time. So, to make it look realistic but still use ChatGPT, the person attempting to cheat would need to have substantial delays (1-2 days) between their repeated uploads.

While it's not perfect, and could be manipulated in many ways, again, I doubt many students who decided to use ChatGPT to cheat would actually put that much effort into the assignment (because, at that point, you might as well try to do the assignment instead). It probably won't be a trustworthy piece of evidence in e.g. a court of law, but it will likely go a long way to disprove allegations of academic dishonesty in the academic circles (in fact, academics seem to be considering it to disprove allegations of using ChatGPT in their 'essays', i.e. papers).

Answer 2

I do not see how VC helps student prove their authorship over the papers, at all.

Sure, it will dissuade people from pasting entire essays generated by GPT, but what is stopping people from incrementally writing essays from GPT? "Hey chatGPT, help me write an introduction for XYZ topic" is all it takes.

Furthermore, this is also unfair to people from a non software background, as VC is almost exclusively only used in software engineering. Git (and its variants) is not something that is very easy to understand, and can be messed up quite easily to someone that is inexperienced. It only takes a couple commands for someone to erase remote history. This is something that can happen even to seasoned engineers, imagine what would happen in the hands of a bunch of humanity majors that never touched a CLI terminal in their life.

Finally, and this is probably my biggest gripe with using VC as evidence of contribution, is the mere fact that sometime people just forget to make regular commit. For context, I am a professional Software Engineer with nearly a decade of experience, working with people that have 20+ years of experience, in a multi-billion dollar tech company. Almost all of us, when raising PRs, would have a huge "initial commit". Almost no one would actually have so called "incremental commits" because guess what, it provides next to no value and it is really annoying to have to type a minimum of 3 commands every 5 minutes. If we don't do it, then how can you expect a bunch of (presumably non tech) University students to maintain a pristine habit of having regular commits without fail?

Oh, and not to mention that git histories can be easily faked. this is a funny prank tool that demonstrates how unreliable git histories can be.

All in all I think forcing people to use Git just to write essays is a pretty horrible idea, and I guess just the concept in general is not really effective, for reasons I outlined. If you must use this approach, I highly recommend (for the sanity of your students) that you just stick with Google docs. It basically offers the subset of functionality that git offers which is relevant to you.

Answer 3

If ChatGPT could be used to complete a full essay, could it not also be used to complete a series of versions of the essay? This might take a bit more work I guess, but it still would be much easier to generate five gradually improving versions of my essay with ChatGPT. And, the reality is that students have been using the Internet and computers to make their work lighter for decades now.

Using version control platforms like GitHub to combat generative AI in student essays written using ChatGPT could be challenging and might not be the most effective approach for several definable reasons:

1. Dynamic and Adaptive Nature: ChatGPT's responses can be tailored to various prompts, making each essay unique. Unlike traditional text where direct copying can be detected, the dynamic nature of AI-generated text might make it harder to identify plagiarism using version control alone. Again, if ChatGPT can make one essay, it can write five progressive essays pretty much just as well. I do this myself, where I use ChatGPT to write me some boiler-plate code, then have ChatGPT progressively improve that code. (Which does at least require some human in the loop knowledge I guess).

2. Lack of Direct Copy-Paste: Students might not directly copy and paste responses from ChatGPT into their essays. Instead, they might use the generated content as inspiration or rephrase it in their own words, making it harder to detect through simple text comparison. This undermines the Turnitin type detectors.

3. Complexity of Content: GitHub and similar version control platforms are primarily designed for managing software development projects, not for analyzing the textual content of essays. Adapting these platforms to analyze and detect plagiarism in essays would require significant development effort and might not yield accurate results (e.g. like Turnitin). Moreover, making each student provide version controlled essays creates more work for the students not using ChatGPT, more work for those grading the work, and not much more work for those already relying on ChatGPT to do all the work.

4. Privacy Concerns: Students might have concerns about privacy and the security of their essays being stored on public platforms like GitHub. It could also raise issues regarding the ownership and sharing of their academic work.

Instead of relying solely on version control platforms, educational institutions should employ a combination of strategies to combat plagiarism and generative content, including:

• Assignment Design: Designing assignments that require critical thinking, analysis, and personal interpretation, making it harder for students to rely solely on AI-generated content. This might look like giving students assignments to analyze content created by ChatGPT. Suppose we handed students a packet of 10 essays on the importance of Jane Austen. Ask them to critique each essay. Some of them might be generative AI text. That could be a curious assignment and it might be harder for ChatGPT to address that assignment without some work on the end of the student.
• Individualized Assessments: Implementing assessments that require students to demonstrate their understanding and apply concepts in unique ways, reducing the likelihood of plagiarism.

Answer 4

University professors should agree you are asking the wrong question. Use of ChatGPT and the like is ubiquitous, and professors know its capabilities. Additionally, there is really no way for students to prove their work is 100% original unless it is done completely and entirely in view of university staff [edit: there is no "commercially viable" way. Professors will not watch hours-long screenshots of one student completing an assignment, at least not in traditional higher education; even this omits the possibility of cell phone use]. In fact, university staff might develop a policy in which students who go far out of their way to request authentication of their work must develop a fool-proof methodology which does so.

In any sense, version control might help students assess their writing and argument-creation process, but it does not offer hope of originality verification, like some other software try.