Are answers written by AI Chatbots suitable for the site?

Question

Relevant question and answer

What is the grammar point that we changed the structure "S + V and V" into the structure "S + V, Ving"?

Is an answer written by an AI chatbot, such as ChatGPT, Bard or others acceptable?

Subquestions:

• Should such answers be deleted?
• Should they be flagged? How?
• What if the post acknowledges that the answer is generated by an AI?
• What if the post is generated by an AI but checked for accuracy by a native speaker?
• What if the post is a correct answer?
• What if the question is about text produced by an AI?
• What about self-answers?

As far as I can tell, we haven't had this discussion yet. Over on ELU there is a blanket ban on AI generated text in question, answers and wikis - Do we want this?

There has previously been a question here on meta about Is it on topic to ask about English advice given by ChatGPT? The community decision was that "Yes, it is okay".

Answer 1

We want original content, not copy-pastes.

I think maybe the best way to frame it is by asking something like "Do we want answers that just link another forum" or "do we want answers that are literally just a copy-paste from wikipedia." If someone asks a question and I find an answer on Reddit that looks right, can I just copy-paste it here?

No, we'd prefer you didn't.

We want answers that are original content (created by you, our loyal and trusted user that we give all these medals to) and if you're pasting from any other site then you're not generating original content.

Really I don't think the rule should be "we don't want chatbot answers". The rule should be "we don't want copy-pasted answers, regardless of where you got it from". It's certainly good (and usually preferred) to cite a notable source but chatbots (or "some post I found on reddit") is not a "notable source" and thus should not be copy-pasted as an answer.

Answer 2

NO

(mostly)

Should such answers be deleted/flagged?
A post with unattributed content from a bot is plagiarism and should get flagged for immediate deletion.

The rest of my answer assumes proper attribution.

What if the post is generated by an AI but checked for accuracy by a native speaker?
Let's take a step back first to what FF actually did, which is to provide a bare quote of what ChatGPT said without any other framing context, that's to say, with no stated support for the content. This kind of answer has no place on ELL or the SO Network at large.

Any learner who wants a direct answer from ChatGPT can install the browser extension or get an OpenAI account and ask the bot themselves. The reason people continue to ask questions here is that, for better or worse, they want answers from humans. A bare quote of an answer from a chat bot is roughly on par with a bare quote from an answer on Quora or Yahoo Answers. None of those sources are at all reliable, so merely quoting what they said doesn't even qualify as an answer in my books, and should be flagged as "not an answer" and deleted.

Now, back to OP's actual question. If FF had said they'd verified and approved of the content of ChatGPT's answer, then FF would be asserting the content as their own answer, though in someone (something?) else's words. This is a value-added version of a bare quote as it comes with the explicit approval of a high-rep user with a positive history of answering questions.

I'm divided on whether an explicitly supported AI bot quote is acceptable as an answer, and leaning heavily (please, please) towards unacceptable.

I can support it in that it's roughly equivalent to saying, "The answer is 'Yes'. User123 said it perfectly on Quora:...", which we do allow here.

But I also oppose it because it may reasonably lead to a flood of answers from people who claim to have carefully read the content, but in fact have not. Finding the exact answer the user is looking for on a site like Quora requires some labour and luck. You cannot build a high-rep account by Quora-mining. You could, however, build a high-rep account by posting unverified quotes from AI bots, but claiming to have verified them. Modding this behaviour would be be nearly impossible. The easiest way to deal with this is to ban AI answers altogether.

What if the post is a correct answer?
This is irrelevant. We don't deal in correct or incorrect answers on this network, only upvotes and downvotes. Correct answers will generally get upvoted and incorrect ones will get downvoted. But anyone with a couple months' experience here knows that a highly upvoted answer can attract an expert who shows that it's not correct after all.

What if the question is about text produced by an AI?
Questions about AI-produced English or AI-produced statements about English are just as valid as those produced by native speakers, English teachers and English student classmates, all of which are welcome here. In other words,

ChatGPT says X. Is that good grammar/true?

is as valid a question as

My friend says X. Is that good grammar/true?

Not asked, but I'm answering it: What if the language of the question itself is generated by an AI bot?
While I wouldn't delete a question if the asker used AI to help them phrase it, I would strongly discourage it because a learner couldn't be sure it says what they think it says, and might assume that because a bot made it, it's perfect.

The trouble I see is people gaming the system by telling ChatGPT things like, "Write a hundred questions typical of ELL" and posting a few every day and building rep. Even with attribution, this violates the intent of the site, which is to help learners, not to generate question content. Since there's no way to determine whether a user is of this type or of the type in the paragraph above, I think we need to ban all questions written by AI.

Overall
By banning all questions and answers generated by AI, we're not depriving anyone of anything, we're keeping closer to the purpose of this site, and we're making our lives easier, so I support a blanket rule of no questions or answers generated by AI. (This wording fully allows for questions about content generated by AI)

Answer 3

No

Answers should be primarily written by humans.

Answers that appear to be written by AI should be deleted. I'm unsure if "Very low quality" or "In need of mod attention" is the right way to flag such questions.

They should still be deleted if there is no attempt to hide the fact that they were written by an AI

They should still be deleted if the answer has been checked by a human, and they should still be deleted if the answer is correct.

Moreover, an AI cannot be considered a source for a claim about English Learning. You cannot use a quote from an AI as evidence to back up your claim about a point of English grammar or the meaning of a word.

AIs do have a role. Questions about the text produced by AIs are on topic. Questions which include, as part of the prior research, the advice given by an AI are on topic. And learners may use AI to draft a question in good English Thus we should be more allowing of use of AIs by Learners asking questions than we are of answers. In this way our policy should be subtly different to that of ELU.

Why

AIs like ChatGPT can produce convincing clear text. They can help learners by always being ready to chat. Their English is grammatically correct (at least it is more accurate than most native speakers) But they don't actually know anything about grammar!

They are, however, always willing to answer a question even if they don't know the answer. And they never know the answer, because they don't actually know anything.

There is a role for AIs in helping Learners to learn. But there isn't a role for them in answering questions. If one wants an answer from an AI, it is easy to get one. The fact that a question is being asked here means that the OP wants a human answer.

Using AI to write answers that are upvoted means that you're getting credit for something you might have zero knowledge of, which flies in the face of the Stack Exchange model.

This isn't about the quality of the answers per se. If a native speaker checks and confirms that the answer is reasonable that doesn't make an AI an appropriate tool here.

A blanket ban is fairer and easier to understand, and therefore enforce. It would be ridiculous to say, "Only native speakers are permitted to use Chatbots to write answers".

Answer 4

It depends on the answer.

AI is a tool, and using it doesn't inherently make an answer bad either in content or form. We already have mechanisms in place to handle poor quality and plagiarized answers.

Do we have to note when a grammar checker was used on our answer text? Why not? Grammar checkers are a more primitive version of natural language AI.

If we allow people to reference webpages without further checking the source of the text from that website we are already allowing AI generated text in answers--it's just been laundered by posting it somewhere else.

I think we should discourage AI answers because of the quality issues that abusing the tool causes. I think banning the use of any tool is a waste of focus and energy. Why shouldn’t I be able to use AI to help me generate example sentences to illustrate a particular usage in my answer?

On Stack Exchange answers ideally can be judged entirely on their content, not by who wrote them or what tools they used to write them. We ask people to cite their sources not just to avoid plagiarism, but also to help people judge credibility.

We should have more confidence in the system. If it can't handle an influx of poor quality answers, we should make the system more robust, not make essentially unenforceable rules.

We don't need to ban AI. Maybe we just need to clarify in How do I write a good answer? that answers need to consist of the author's own work and not be copypasta from other sources no matter how accurately it is attributed.

Answer 5

No

The site is about learning English. While there is certainly no prerequisite that answers must come from native speakers, it is specified on the homepage that questions must be 'practical' and therefore answers must also have authority on practical use. This can only come from native speakers or those who have learned English to a sufficient degree to be able to answer authoritatively having put what they learned into practical use. Artificial Intelligence cannot fall into that group yet, and possibly never will.

AI draws from information found on the internet. Even with advances in machine learning and people saying how 'eerily human' it now feels, it is still just drawing from what other people have written. It may be able to 'learn' what is correct and incorrect but it isn't able to discern it. My own experience with ChatGPT has shown me that it will give me incorrect information, then 'apologise' and admit it was wrong when I challenged it. If it were to use unidiomatic language it cannot see the other party's response in real-time - it only gets a calculated response from the user. This is not the same as putting what you have learned about a language into practice. In fact, AI couldn't even be said to be a 'native' speaker of any language. A native speaker is someone who has a first language learned socially. Native speakers are sometimes said to 'think' in their language, even though they may speak others. A computer does not 'think' in English. If you asked it the same question in two different languages it may give you the same answer run through a translator. Some AI algorithms may just select texts in the language you speak to it in order to collate a response, meaning you could get conflicting answers depending on how you posed the question.

There are so many different opinions about AI at the moment. Those who are most impressed (or scared) by what AI can suddenly do tend to be people who don't know anything about it and are worried about how it may affect their livelihoods as artists, writers, teachers etc. Those who actually work in AI - which as a data analyst, I do to an extent as it is one of the tools available to me - tend to have a different opinion, that it isn't that sophisticated or remarkable and is nowhere near the degree of 'sentience' that some are suggesting, and certainly isn't capable of replacing human responses.

Answer 6

Yes, but ...

To me as a non-native English speaker, an AI chatbot is useless because I don't know whether its answer is correct or not. But if some native speaker makes use of it, then I will know the answer is checked by him personally - that is, he as a native speaker confirms the correctness of an AI chatbot's answer and, therefore, I can trust it.

But it's very important to note that the AI's answers are often very vague. So they usually have little value.

Answer 7

No.

AI is inherently plagiarism from better-qualified authors who deserve credit for their work, and who do not deserve to have the fruits of their work endlessly recycled, without attribution, without any request for permission, and without any offer of payment in return for the vast financial gain accrued from the use and abuse of their intellectual property.

Nobody denies that AI depends on source material to create a facsimile that appears to be genuine. It cannot create anything. It has no ideas, no understanding and no actual intelligence. Even the AI art we all saw first a few months ago is not art, even if no artwork on earth looks exactly like it. It is infinite varieties of plagiarism and pastiche of other people's original work, stirred up and spat out as "original" art. But it cannot exist without the unsourced, unacknowledged, unpaid source artists whose work is fed into the AI meat grinder as the one indispensable ingredient in the unhealthy AI burger.

Nobody claims that AI can generate original thought. Yet. If all AI content is recycled versions of human creativity, it is a) work based on the unpaid labour of millions of humans and b) is nothing other than an unsourced and unverified pastiche of actual human intellect, originality and creativity. It is not possible that it can be "better" than the human sources it is based on. But it can be worse.

Ban it, not just because it sucks, but because it is an act of theft by large corporations sold back to the authors as a "free tool".

Answer 8

Yes

I'm the user who posted a response from chatGPT as an answer in the question linked to by OP here.

Before doing so, I had already posted a relevant "answer / comment" myself, and carefully read the Answer posted by the current OP to that question.

I don't think it should matter that the actual text of my answer came from chatGPT (apart from the initial disclaimer pointing out the source, which I think we should insist on if such answers are to be permitted). The point is I think everything chatGPT says there is accurate and well-phrased - and probably more elegantly expressed than what I might have written myself.

---

EDIT - I think it's important to note that I'm only endorsing AI-generated Answers (that have been "curated / verified / endorsed" by competent native speakers OR "advanced" non-native Anglophones). I don't think it's a good idea to allow Questions primarily or solely focused on the idiomatic validity or factual accuracy of some AI-generated text.