Plagiarism & ChatGPT: Is Cheating with AI the New Normal?

[anorlunda]

The following is an excerpt from https://reason.com/volokh/2022/12/28/plagiarism-and-chatgpt
It talks about exams. Other profs talk about its use with homework.
ChatGBT answers are not identical to other things posted on the Internet. Normal plagiarism detection will not work.

If you haven't tried it yourself, it's worth a try. Be aware that after the initial answer, you can criticize it, or give more detailed requirements and it will produce improved answers. @Greg Bernhardt even tried feeding it some broken HTML code, and the AI found the bugs and gave back improved correct code.

https://chat.openai.com/chat
You'll have to create a free account.

Is this something to worry about? What is the defense?

JOSH BLACKMAN | 12.28.2022 12:06 PM

Since I began teaching, I have only given essay exams. No multiple choice. No short answers. Rather, each exam has two, complex issue-spotter essay questions. The exam is completely open-book. I always tell my students they can bring whatever they want to the classroom–nothing will help them. I also issue a regular warning: do not cheat, because I will spot similarities in writing very quickly. Over the years, I only had once incident. I found that two students had very similar answers to a particular essay question. I inquired further, and found out that the students were study partners, and had pre-written answers to questions in advance, based on what I had asked in the past. And they inserted those answers into the exam. The pre-written answers were not exactly on point, and did not receive full credit, but there was no plagiarism in that case.
Enter ChatGPT. This "chatbot" uses sophisticated technology to generate answers to questions. These responses are written in plain English, that are easy to understand, and incorporate information from a massive neural network. These responses are not perfect, but may pass muster with professors who are short on time. The temptation for cheating is real. And one professor in South Carolina caught plagiarism. He wrote about it on Facebook, and the New York Post followed up.
This technology should strike fear in all academics. ChatGPT does not work like TurnItIn, and other plagiarism detection software. The software generates new answers on the fly. And each time you run the app, a different answer will be spit out. There is no word-for-word plagiarism, or poor paraphrasing. Each answer is unique. And ChatGPT is constantly evolving. It gets smarter as more people use the system, and the neural network grows. The system was only launched three weeks ago. By May, the system will be far more sophisticated, as it incorporates everything that comes before. Like the Borg, students will assimilate; resistance is futile.

 

[Vanadium 50]

This is not new. Students have been able to buy essays since Plato. What is new is the drastically lower cost in time and money.

Some universities will accept the situation.
Others will increase the cost to cheat, most easily by increasing the severity of penalties.

 

[PeroK]

There's an article in today's Guardian newspaper on ChatGPT. Apparently, they are working on a scheme to identify AI-plagiarism. Here's a lecture by Scott Aaronson about it:

https://scottaaronson.blog/?p=6823

 

[Hornbein]

Henry Ford II failed to graduate from Yale. Tucked inside his senior thesis was the bill from the guy who wrote it.

Ted Kennedy got caught having some other guy take a test for him. Was expelled from Harvard, wound up at Dartmouth.

 

[anorlunda]

There's an article in today's Guardian newspaper on ChatGPT. Apparently, they are working on a scheme to identify AI-plagiarism. Here's a lecture by Scott Aaronson about it:

https://scottaaronson.blog/?p=6823

Thanks for sharing. All of us should read that blog post. Most interesting is the following explanation.

So what is GPT? It’s a text model. It’s basically a gigantic neural network with about 175 billion parameters—the weights. It’s a particular kind of neural net called a transformer model that was invented five years ago. It’s been trained on a large fraction of all the text on the open Internet. The training simply consists of playing the following game over and over, trillions of times: predict which word comes next in this text string. So in some sense that’s its only goal or intention in the world: to predict the next word.

The amazing discovery is that, when you do that, you end up with something where you can then ask it a question, or give it a a task like writing an essay about a certain topic, and it will say “oh! I know what would plausibly come after that prompt! The answer to the question! Or the essay itself!” And it will then proceed to generate the thing you want.

So it was serendipity. The goal was word prediction, not intelligence per se. But the "amazing discovery" was that it seems intelligent.

"Who is buried in Grant's Tomb?" well a search of the world's literature shows that the most likely word to follow those words is Grant." Well duh! But we can tell the world that it is AI because of the neural net.

That story also suggests the way to make a version 2.0 bot, focused on science an engineering. Instead of using all the Internet text to train it, use only the text from the most trusted textbooks and journals, and don't forget to include @PeroK 's great answers on physicsforums.com.

It would be great fun to make a grad student project trying that idea using only physicsforums archives. It is interesting because on PF many questions and answers are highly repetitive over the years. Make a ChatPDM (Peter Donis Machine) and a ChatJHM (Jim Hardy Machine). What would Jim Hardy say to that question? He would probably say roughly the same thing he said to similar questions in the past. But it is too tedious to do a manual ad hoc search of JH's history for each question. It is less tedious to use all JH's history and put it in a neural net.

 

[Office_Shredder]

"Who is buried in Grant's Tomb?" well a search of the world's literature shows that the most likely word to follow those words is Grant." Well duh! But we can tell the world that it is AI because of the neural net.

This is a gross underestimation of what is happening here.

 

[martinbn]

Wouldn't it be obvious, that they cheated, when you ask them questions on the topic.

 

[MidgetDwarf]

Henry Ford II failed to graduate from Yale. Tucked inside his senior thesis was the bill from the guy who wrote it.

Ted Kennedy got caught having some other guy take a test for him. Was expelled from Harvard, wound up at Dartmouth.

im sure his father made them an offer they could not refuse.

 

[Office_Shredder]

To be a bit too specific, is pasting in a chatgpt answer to a question prompt *plagiarism*? Plagiarism implies you are stealing someone else's work - whose work are you stealing? The us copyright office recently ruled that ai generated content can't be copyrighted because it's not created by anyone, this feels kind of similar. You're not presenting your own work, which is the point of an assignment, but perhaps the thing being done here is less bad than plagiarism.

 

[anorlunda]

To be a bit too specific, is pasting in a chatgpt answer to a question prompt *plagiarism*?

Like many other things in the modern world, the definition of Plagiarism may need some adjustments.

If the spell checker assisted you in writing your post, does the spell checker need acknowledgement in the credits? It is a continuum, and whatever lines we draw are necessarily arbitrary.

 

[Office_Shredder]

If the spell checker assisted you in writing your post, does the spell checker need acknowledgement in the credits? It is a continuum, and whatever lines we draw are necessarily arbitrary.

This is an excellent analogy to consider.

 

[Vanadium 50]

I don't think analogies are helpful. All analogies are imperfect and the imperfections sometimes make all the difference.

I would focus on actions, not technology. Technology impacts cost, not whether it is right or wrong. AI, Mechanical Turk, TaskRabbit or Poindexter down the hall should make no difference.

 

[Office_Shredder]

.
I would focus on actions, not technology. Technology impacts cost, not whether it is right or wrong. AI, Mechanical Turk, TaskRabbit or Poindexter down the hall should make no difference.

3 out of the 4 have you taking uncredited work from another human, the other has you using a tool to generate new content.

Consider the following sentence:

Every time I talked to a teacher, I talked to their TA, and talked to their principal, and talked to there superintendent as well.

That's a totally original sentence I wrote. Consider the following options:
1.) Grammar check points out I wrote there instead of their.
2.) Thesaurus check complains I use the word talk a lot, and changes some of them to other synonyms.
3.). Readability bot suggests I write teacher's assistant instead of TA since I haven't defined the acronym yet
4.) I put that sentence into Chatgpt and ask it to rewrite it for me. Here's what I got:

Every time I spoke with a teacher, I also communicated with their TA, principal, and superintendent

Which if any of these may I use in my paper without referencing the robot? If they are all OK, can I do a paragraph into chatgpt instead of a sentence? Is it OK if the new paragraph is longer, or contains more content?

What if instead of asking chatgpt, I asked a friend to review what Iwrote and they made some suggestions? Do i have to quote them now?

I agree that sticking a one sentence prompt into the bot and getting back a 5 paragraph answer feels wrong, but there's a lot of gray area here.

 

[PeroK]

If the spell checker assisted you in writing your post, does the spell checker need acknowledgement in the credits? It is a continuum, and whatever lines we draw are necessarily arbitrary.

Yes, if you contracted in some legally binding way to use the spell checker only for your own personal use and not for professional or academic purposes. The lines are ultimately legal.

Using ChatGPT can only be plagiarism if the owners of ChatGPT put legal constraints on its use. They could require you to pay by the word for anything you submit professionally or academically. In the same way that you potentially pay a license fee for use of any software.

Whether a professional body or academic institute accepts essays written by an AI engine is then a separate issue.

For example, if you are a freelance programmer, you can use whatever tools you have bought or are licensed to use. Your client is unlikely to care how you generated the code, only on the quality of the code.

Similarly, if you are freelance author, then I doubt a magazine or newspaper would care how you generated an article - although if ChatGPT does too much of the work, then you're at risk of being cut out of the loop.

 

[martinbn]

I think that plagiarism in the context of exams/tests/homework is meant as cheating. The problem is not whether you stole someones' work, who deserves the credit, but whether you did the work yourself and thus showed if you have learnt what you were supposed to. In this case I'd say make the exams/tests/homeworks in such a way to avoid that. If it comes to actual plagiarism, as in you publish your work, but you didn't do the work it was the bot that did, then I don't know. On one hand usually you don't credit software for producing your graphs or doing the tedious calculations. But if the software did all the work, then why is it your name on the paper! May be, if you think that the results are worth sharing with the community, you can publish them anonymously or say how it was done. Personally I don't see the problem here because as of now it seems that the bots produce only garbage.

 

[Frabjous]

I heard an interview with an English prof a couple of weeks ago. He said that his real worry was using AI for the first draft which he viewed as the hardest part of the essay. The subsequent rewrites would correct AI foibles and probably eliminate most watermarking attempts.

 

[anorlunda]

More from Professor Blackman. This time about ChatGBT trying to pass the Multistate Bar Exam (MBE).

https://reason.com/volokh/2023/01/02/can-gpt-pass-the-multistate-bar-exam/

Based on anecdotal evidence related to GPT-4 and LAION's Bloom family of models, it is quite possible that this [pass the bar exam] will occur within the next 0-18 months.

 

[Andy Resnick]

The following is an excerpt from https://reason.com/volokh/2022/12/28/plagiarism-and-chatgpt
It talks about exams. Other profs talk about its use with homework.
ChatGBT answers are not identical to other things posted on the Internet. Normal plagiarism detection will not work.

If you haven't tried it yourself, it's worth a try. Be aware that after the initial answer, you can criticize it, or give more detailed requirements and it will produce improved answers. @Greg Bernhardt even tried feeding it some broken HTML code, and the AI found the bugs and gave back improved correct code.

https://chat.openai.com/chat
You'll have to create a free account.

Is this something to worry about? What is the defense?

A few department colleagues and I starting talking about this exact issue last semester; we don't have a definitive response yet but are thinking along the lines of treating AI-generated text by students similarly to how the student use of graphing calculators and symbolic math software has been successfully incorporated into our classes.

As always, the first line of defense against cheating is to reduce the motivation to cheat.

 

[anorlunda]

A few department colleagues and I starting talking about this exact issue last semester; we don't have a definitive response yet but are thinking along the lines of treating AI-generated text by students similarly to how the student use of graphing calculators and symbolic math software has been successfully incorporated into our classes.

That sounds enlightened. And you might learn how to teach the students how to use critical thinking to detect bad AI answers, and/or how to elicit improved answers from the tool. In other words, how to use it skillfully, just like any other tool.

 

[Filip Larsen]

https://arstechnica.com/information...-chatgpt-fearing-negative-impact-on-learning/ (as also referenced in another ChatGPT thread):

New York City Public Schools have blocked access to OpenAI's ChatGPT AI model on its network and devices, reports educational news site Chalkbeat. The move comes amid fears from educators that students will use ChatGPT to cheat on assignments, accidentally introduce inaccuracies in their work, or write essays in a way that will keep them from learning the material.

 

[anorlunda]

New York City Public Schools have blocked access

Now compare that with @Andy Resnick 's reply. Which approach do you think will win in the end?

Also, let's react slowly rather than rapidly. Google search in the 1990s was revolutionary. It changed the world. But it did not accomplish that change in the first 30 days after release.

 

[Vanadium 50]

And by Google, do you mean AltaVista?

 

[anorlunda]

And by Google, do you mean AltaVista?

Definitely not. Search was ho hum before Google, and users took several tries to find what they were looking for. Yahoo's yellow page approach was still popular. Google's PageRank was much more effective and it lit the field on fire.

I used to happily use AltaVista, but it was my attorney who alerted me that this much better Google search existed.

 

[Mark44]

To be a bit too specific, is pasting in a chatgpt answer to a question prompt *plagiarism*?

The problem is not whether you stole someones' work, who deserves the credit, but whether you did the work yourself and thus showed if you have learnt what you were supposed to.

Which is really not so different from the types of testing where no calculators/no graphing calculators are permitted.

 

[Office_Shredder]

Which is really not so different from the types of testing where no calculators/no graphing calculators are permitted.

That's not plagiarism, that's cheating. I said I wanted to be specific :)

 

[Mark44]

That's not plagiarism, that's cheating. I said I wanted to be specific :)

That was my point, too, that it wasn't so much about plagiarism, but rather, cheating -- taking credit for work done by someone/something else.

 

[Wrichik Basu]

https://www.timesnownews.com/educat...at-exposes-ai-written-essays-article-96809855

Princeton student Edward Tian has developed an application that can simply identify text curated by artificial intelligence. On January 2, Tian shared a few concept videos which highlights GPTZero's capabilities.

Bloomberg reported, “The app was so popular that it crashed "due to unexpectedly high web traffic," and currently displays a beta-signup page. GPTZero is still available to use on Tian's Streamlit page, after the website hosts stepped in to increase its capacity.”

The website https://gptzero.me/ is currently showing "Connection timed out" for me.

 

[Demystifier]

And ChatGPT is constantly evolving. It gets smarter as more people use the system, and the neural network grows.

That's actually not what ChatGPT does. ChathGPT, in it's current form, stopped learning in 2021.

 

[Vanadium 50]

ChathGPT, in it's current form, stopped learning in 2021.

Like so many students....

 

[mysterium]

Have you used GPT-4 yet? It seems that in the future, educational institutions must have programs like Turnitin to detect generative AI.

 

[scottdave]

Have you used GPT-4 yet? It seems that in the future, educational institutions must have programs like Turnitin to detect generative AI.

Turnitin now has the ability to detect AI-written content. I don't know if it is standard or optional.

 

[cbance]

The following is an excerpt from https://reason.com/volokh/2022/12/28/plagiarism-and-chatgpt
It talks about exams. Other profs talk about its use with homework.
ChatGBT answers are not identical to other things posted on the Internet. Normal plagiarism detection will not work.

If you haven't tried it yourself, it's worth a try. Be aware that after the initial answer, you can criticize it, or give more detailed requirements and it will produce improved answers. @Greg Bernhardt even tried feeding it some broken HTML code, and the AI found the bugs and gave back improved correct code.

https://chat.openai.com/chat
You'll have to create a free account.

Is this something to worry about? What is the defense?

I would like to leave some of my own input and feedback on this one. - Well, it depends on how you look at it and how it's being used. Once example that really gets up my nose is Essay writing services that use ChatGPT to provide paid Essay writing which is basically cheating on education. Not to mention it produces really poor output and out of date information.

 

[WWGD]

Henry Ford II failed to graduate from Yale. Tucked inside his senior thesis was the bill from the guy who wrote it.

Ted Kennedy got caught having some other guy take a test for him. Was expelled from Harvard, wound up at Dartmouth.

Dartmouth: For when you get caught cheating at Harvard.

 

[WWGD]

Actually, there's a contest, award included, to detect cheating through using GPT.

 

[jack action]

When I was an undergrad, there was a teacher who gave us the example of a student who was so lazy at copying someone else's homework that they only bothered to put correction fluid over the original author's name and write their own over it. These people apparently still exist with AI: