Is it OK to use GPT-3+ to rewrite your own paper to have better English?

Question

As an example, someone who's not a native speaker of English might come up with the following paragraph (generated using a second-class translation service):

Basic principle of gamification is providing of receipt of permanent, measurable feed-back from an user, providing possibility dynamic adjustment of user behavior and, as a result, rapid mastering of all functional possibilities of appendix and stage-by-stage immersion of user in more thin moments. Another method of gamification is creation of legend, history, provided with dramatic receptions, that accompanies the process of the use of application. It assists creation for the users of feeling of complicity, contribution to common cause, interest in the achievement of some invented aims.

One could then use GPT-3+ (aka ChatGPT) to rewrite the above into something closer to what a native English speaker might write:

The fundamental idea of gamification is to give users continual, measurable feedback, allowing them to adjust their behavior dynamically and quickly learn all the features of the app. Another way of gamification is to create a story with dramatic elements that accompanies the user's experience. This helps to create a sense of involvement, contribution to a shared goal, and interest in achieving the set objectives.

Ignoring the question of how good of a job GPT-3 did here, is such use of neural networks acceptable in academia? All the original ideas are still yours but GPT-3 helps you convey them using shorter phrases and better English.

Update: Since the example I've provided generated a lot of debate, here's a slightly better original text with a correction provided by GPT-3:

On enormous territory of park there are lakes, rivers, canyons and caves. Lake Yellowstone, one of the greatest alpine lakes in North America, is located in the center of Yellowstone caldera, greatest supervolcano on a continent. Caldera is considered a dozing supervolcano; he erupted with enormous force several times for last two million years. Greater part of territory of park is covered by hardening lava; in a park there is one of five existing in the world of the geyser fields.

Corrected to:

On the enormous territory of the park, there are lakes, rivers, canyons, and caves. Lake Yellowstone, one of the greatest alpine lakes in North America, is located in the center of the Yellowstone caldera, the greatest supervolcano on the continent. The caldera is considered a dormant supervolcano; it has erupted with enormous force several times over the last two million years. The majority of the park's territory is covered by hardened lava; within the park, there is one of the five existing geyser fields in the world.

As you can see, an extremely poor text is probably a bad use-case for GPT, but a few errors here and there are well within its capabilities.

Answer 1

Edit: the question after the edit is so different from the original version that I wrote a completely new answer. My original answer is preserved at the bottom of the post.

1. New answer (added following OP's update to the question, which effectively turns it into a new question)

Your added example shows the use of ChatGPT to improve a text written in intelligible, but unidiomatic, English. ChatGPT apparently outputs a grammatically flawless version of the same text, without any of the meaning having been altered in any way.

So, is it okay to use ChatGPT in such a way? Yes, absolutely — as I said before, it's exactly as "OK" as it is to use Google Translate, Grammarly and similar tools. That is, as long as you are not submitting the text as a class assignment meant to test your English writing abilities, in which case it would obviously be cheating. Otherwise, I can't think of any plausible objections anyone could raise here. The issues of accidental plagiarism, giving credit to the creators of ChatGPT and the texts it was trained on, etc, that were raised by some people previously, seem completely irrelevant in this scenario. All you are doing is improving grammar, and that's perfectly fine if indeed ChatGPT is capable of making such improvements.

To be clear, my analysis above pertains only to the types of usage illustrated by the specific example you posted in the updated question. I still have some doubts about whether putting too much trust in ChatGPT's capabilities is a good idea, and can imagine situations where a writer with a poor grasp of English language nuances would not realize that ChatGPT was making their text worse in some ways, or even modifying the ideas of the text or committing plagiarism. But that's a secondary concern that's not so pertinent for the current discussion. Ethically, there is nothing inherently wrong in using automated language improvement tools, as long as we remember that at the end of the day, writers are responsible for the content they are putting out into the world, in the same way that the driver behind the wheel of a Tesla running on (the misleadingly named) "Autopilot" mode is responsible for what their car is doing on the road. If ChatGPT messed up and you ended up putting your name to an ethically compromised text, that's on you. With responsible use, this can be avoided.

---

2. Original answer and edits prior to OP's update to the question. This addresses the "bad" example in the original question and seems no longer relevant for what OP actually wants to know

It's exactly as "OK" as it is to use Google Translate, Grammarly, or any of the other automated language-transformation tools that purport to turn all of us into budding Hemingways with zero effort on our part.

To be more precise, there isn't anything unethical about it, but as of the time of this writing, it doesn't work well enough to be anything but a complete waste of time (as your own example illustrates). So if by "is it OK" you mean "is it a good idea", the answer is no.

Edit: to be clear, it’s possible that some ways of using ChatGPT, when used with caution and some effort by a person with already reasonably good English language proficiency, may actually help the person produce writing of slightly better quality. See @anjama’s comments below. I make no claim one way or the other on this issue. My answer above was addressing OP’s specific suggested use case, as illustrated by their example.

Edit 2: I took a closer look at OP’s example input text for the “improvement” workflow. Honestly, the text is pure gibberish, and the idea that an intelligence, whether artificial or human, could penetrate through the poor writing and fathom the writer’s “actual ideas”, then express them more eloquently, is simply absurd. Garbage in, garbage out, as they say. I suppose you could interpret it as a sort of general prompt for someone to discuss gamification and what it’s about. If I had to rewrite it, I could use my own knowledge of what gamification is to write something plausible, but it would bear little relation to whatever it was that the writer of the text actually wanted to express.

To summarize, OP specifically asked for his premise not to be criticized, but the question simply makes no sense in its current form: either the premise is completely false, or the example OP used to illustrate the situation he had in mind is atrociously bad. In the absence of another, better example that lives up to OP’s premise, the question cannot be given a meaningful answer.

Answer 2

Ignoring the question of how good of a job GPT-3 did here…

I think a bit of a frame challenge is necessary here. If GPT-3 or some other language model was capable of preserving meaning exactly (not adding, not removing, and not altering) while making purely stylistic changes, that would be one thing. But that's not what the GPT family does. It produces convincing-sounding waffling that is in many cases not quite semantically equivalent to the input, or, when asked to answer a new question, not quite correct. In academia, the devil is really in the details. So while asking GPT-3 for writing help can be useful (just the other day, it helped me discover the term "pleading necessity"), copying and pasting its responses is a dangerous road to go down. Think of it as a very diligent but overconfident and intellectually shallow assistant.

Answer 3

I would treat this approximately as I would treat proofreading by a human. If it results in substantial changes to the presentation, it should be mentioned in the acknowledgments ("Our thanks go out to the developers of GPT-4, which made numerous useful suggestions on the presentation of the ideas put forward in this paper. In addition, it spotted a crucial error in an earlier version of the proof of Lemma 5, without which Theorem 7 would not have been possible, and assisted us greatly in formalizing the proof of Theorem 8...."). But there is nothing unethical about getting feedback on a paper from anyone, irrespective of whether they are a person or a software.

Answer 4

No!

GPT does not understand the input it is given. You can have no confidence that GPT will not change a correct statement into an incorrect one, possibly in subtle ways that will be difficult to spot for our hypothetical non-native speaker. For example, in your sample text, GPT changed the word "permanent" into "continual," which doesn't mean the same thing. It also changed the first sentence from the original, which was about collecting feedback from a user, into one about providing feedback to the user.

You may argue that these changes make the text more correct. That is a reasonable argument. However, the reason GPT is able to do this is that it is trained on a broad corpus of texts. Not knowing any better, it will combine the input text it is given with appropriate selections from its training data, thereby incorporating elements that are not the author's original work. Unless these sources are identified (likely impossible) and cited appropriately, this may constitute plagiarism.

Answer 5

The answer for me is Yes,

Using neural network to "rephrase" your passage is acceptable in academia. As far as I know, no rule of journal states that you can't use a language tool to revise your paper. And it doesn't matter if that's ChatGPT or other NLP model.

But acknowledge that ChatGPT is called ChatGPT for a reason, no matter what the true intention is, using a model optimizing for dialogue to an academic scenario just because that is "closer to native English speaker" is truly not a wise idea.

Just be careful that ChatGPT may(actually probably) lose content that help convey your idea of original work, and the rest of the stuff is take your own risk and mention in the acknowledgements

Answer 6

If the outcome really only includes your own ideas and they haven't been extended by the language model then it should be fine. Then it's like a German English translation but bad English into good English. This is all based on the assumption that writing a good English text isn't part of the rating.

Answer 7

Instead of transforming the text directly, you can ask it for points on how to improve the text.

That way, you can decide for yourself whether the points it suggests are worth incorporating into your text or not. It also reduces the probability of GPT messing up your original thoughts/arguments in the text.

When generating text from scratch or rewriting, you need to be very careful as it can make things up and/or change the meaning of what you originally intended. It can be similar to reviewing text from an inexperienced student who is good at bullshitting. If you are confident in your ability to spot the bullshit, fine, but that's your responsibility.

Ultimately, I don't think it is possible to ban the use of these language models, and every person needs to decide for themselves whether it creates value for them or not.

Answer 8

Generally speaking, the answer is an emphatic no. The main reason for this no answer is as follows: the purpose of writing a paper is to demonstrate your learning and your work. Asking some tool or some automation to take what you can produce, then "improve" it, and finally present that "improved output" as "your work" is misrepresenting the reality. When it gets discovered that "your work" was not really yours, it will result in adverse consequences. These consequences will be much harsher than when your work was just not "good enough" but at least it is "yours". So, it is in your best interest to not mis-represent some one else's or something else's work as your work.

That said, it is ok to use ChatGPT or any other tool for improving you. The main reason is as follows: every tool that analyzes your work and provides a feedback for possible improvements is helping you in improving yourself. An improved you will produce a better quality work. So, one should use these tools to "learn", rather than "copy". That learning will, over time, result in a better quality output from oneself. By using these tools to "learn", you will have gained skills. By using these tools to "copy" you miss an opportunity to make yourself better.

Answer 9

It's as okay as is using Google Translate, automated spellchecking, finding synonyms with MS Word, etc. I don't see any downsides.

Yes, a person may appear more fluent than they are, but if they do it consistently what's the issue? Yes, it may hamper someone's language development, but that's their choice. Yes, it may give you an edge over competitors, but it's their own fault for not using all the tools at their disposal.

All the Luddites who think this is a bad idea will go the way of the dodo. Don't mind them.

Answer 10

I have not used ChatGPT, but I have used a different, less intrusive tool called LanguageTool, in particular when I write German (to help with die/der/das/dem/den/des), but to a (much) lesser degree when I write English. LanguageTool is an automated checker for spelling, grammar, and style, but it also suggests how to reformulate a sentence, such that a strict language teacher might argue the text is no longer entirely my own.

Check the rules of the journal. Chances are they allow automated spelling, grammar, and style checkers, but not text generators such as ChatGPT.

Whatever you end up using, you should mention in the acknowledgements. You would thank a colleague for proofreading your English; the same applies to artificial intelligence or automated corrections based on other technologies. Such an acknowledgement also provides full disclosure, protecting you from allegations of academic dishonesty by people who don't think you should use those tools or that you must be open about it if you do. The terms and conditions of the tool and/or the journal may also require such disclosure.

For example: Parts of this text were processed using aspell/LanguageTool/ChatGPT/Deepl Translate/Google Translate/... to improve language.

Answer 11

As long as we are only limiting our usage of AI to paraphrase and rewrite our own text and original work (That is not plagiarised), It is not merely OK, but important and essential to leverage and utilise every tool available to you to improve your writing and the final output.

Your writing has to evoke interest, be precise, short and convey the ideas, findings and such with as much flair as possible.

AI tools can help non-native writers overcome language, idiom, grammar and syntax challenges. And this is welcome.

The caveat at the beginning of this answer stands at the end too - given the power of AI to create text based on prompts, it is challenging to say a clear yes or no. So, as long as we limit the use of AI for paraphrasing, it is a welcome use.

Answer 12

At the time of this writing, the answer is likely to be fluidly changing, and dependent on the exact journal in question. For example, it was reported today that the International Conference on Machine Learning (ICML) has banned the use of large-scale language model (LLM) programs such as ChatGPT in conference papers, writing in their newest call for papers:

Papers that include text generated from a large-scale language model (LLM) such as ChatGPT are prohibited unless the produced text is presented as a part of the paper’s experimental analysis.

As a follow-up to public interest queries, the ICML wrote a longer clarification, including the note:

The Large Language Model (LLM) policy for ICML 2023 prohibits text produced entirely by LLMs (i.e., “generated”). This does not prohibit authors from using LLMs for editing or polishing author-written text.

However, it's debatable whether LLMs like ChatGPT are actually suitable for that purpose of editing and polishing text. For example, interviewed by the Verge, Deb Raji (AI research fellow, Mozilla Foundation) said:

I see LLMs as quite distinct from something like auto-correct or Grammarly, which are corrective and educational tools... Although it can be used for this purpose, LLMs are not explicitly designed to adjust the structure and language of text that is already written — it has other more problematic capabilities as well, such as the generation of novel text and spam.

Answer 13

Let's avoid the discussion about the content/quality (of Chatwhatever AI) and let's stick to the formal part.

The worst thing you are doing is transferring some text that should be confidential (where "some text" is "your review") to a service over the web on which you have no control where/how is storing and accessing said text.