6

Back in December 2022, an AI chatbot ChatGPT was released to the public. It is really sophisticated that it can also answer many kinds of questions, regardless if it is factually correct or wrong.

It has been banned on Stack Overflow and made official due to the disruption it caused. Another discussion on the main Meta concluded with a staff stating "we hope that folks on network sites feel comfortable establishing per-site policies responsive to their communities' needs."


Since then, there were a few users (ab)using ChatGPT by posting their responses as answers as-is without mentioning the source. Currently, these have been deleted as plagiarism (as per our help center, How to reference material written by others).

Moving forward, how should these kinds of AI-generated posts, including answers and questions, be treated in this community?

2 Answers 2

4

I vote for a NO for bot generated posts1.

(1) Our question authors' base consists of users who know nothing about what Android is, to the people who writes code for the Android OS itself, and everything between that.

(2) The devices these askers' use; their Android versions; each having many undocumented tweaks and configurations with no commonality between two different devices from same vendor; each device receiving some update that suddenly breaks consistency with no explanation available anywhere;

(3) Both (1) and (2) pose a serious problem for us regular answerers whose strength in numbers is largely insignificant.

Why did I bother mentioning that?

(A) When an author from (1) asks a question we are given a block of text that is missing requisites. We often need device related info, and we need to seek clarification on what is acceptable as a solution and what is not; furthermore, because of the problems of (2) we often have to tailor our solution to non-generic2 problems; we sometimes have to use trial and error until a solution emerges that is acceptable to the asker; on top of that, we answerers come to know that some new Android version or app has made changes that reduces the usefulness of our answer. So we now gotta test out solution again, or delete it, or mark it as a note that it won't work for this or that version.

(B) Either way, we answerers know what we are doing. Our solutions work. And we take guarantee that our solution was tested on a specific environment with particular specifications. This eliminates blind guesswork in the answer. We strive to work precisely. Posting only what works, and making a note on what might not work or why it won't work.

It's just that when a human posts a bot generated post:

(C) they never bother to figure out the problem statement like we do now. Being slapped with a generic answer to a question that needs a non-generic solution, I think it would annoy the asker. Such a thing would definitely infuriate the hell out of me if I put myself into asker's position;

(D) Bots make guesswork. one pattern in bot generated posts I have seen so far here is that the human poster has no idea what is correct and what is not in that answer. They don't even grasp the question sometimes. Said otherwise, the user is doing a copy-paste having no idea if the answer would even work or not. This is no different from guesswork based answers we sometimes receive. I despise guesswork personally. A guesswork post ought to inform the reader in the beginning that the post is a guesswork and is tested on nothing! An answer reader can save their time and preserve their sanity that way.

(E) Bots breed ignorance. Since a human user merely copy-pastes a bot-generated post it is natural that when confronted about the accuracy, completeness, correctness, or just about a clarification on any particular instruction in the post, the answerer would try to defer answering hoping that the matter dies down, or they might after some probable struggle and fight simply accept that it is a bot-generated post and that they cannot explain what is being asked there. Either way, this would be a collosal waste of time of the asker, a site vistor, and us reviewers. I decline to have my time be expended on something where a human cannot explain the stuff they have written.

(F) Plagiarism is a weak criteria. The problems mentioned in (C), (D) and (E) won't go away merely because the human user posts a note attributing the text to whichever bot created it.

(G) Signals low-quality question. If a bot generated post answers the question to asker's satisfaction than we have a known problem. The problem that the asker did not do sufficient search on the net. Such a question would be no different from one whose answer is just one web-search query away from their favorite search engine (otherwise known as LMGTFY). In that case, the question is low-quality and relevant site policy ought to be applied there.

I do provide a solution to the human users who think that they are trying to "help" by posting bot-generated posts. Given commenting privileges, they can post a link in the comments to their preferred bot which in their opinion can answer the given question. If the suggestion works than it is proof enough that the question is low-quality, and if it doesn't then we at least won't have to read the garbage ourselves!

All in all, I say NO. We should not accept bot generated posts.


My personal policy: I currently treat these posts as spam and deal with them that way. I care not if the answer somehow works. It gets deleted by my diamond. That's just it. Of course if the consensus here emerges to keep these posts, than I would use that policy.


  1. A bot generated post is a post created by technologies which are often referred to as some sort of AI. I use the term "bots" to refer to front-ends of these technologies.

  2. To me a non-generic problem is any problem where a custom tailored solution is needed. The opposite is where one answer applies to all related questions. For example, "How does Android's Virtual Machine works" is a generic question. It can be answered once and all related questions can be marked duplicate to it. However, "My Redmi XYZ device automatically restarts after last OTA update? How do I fix this?" is a non-generic problem. The solution to that is not guaranteed to work on other devices.

4

How should we treat ChatGPT (and other AI-generated) posts?

Just like any other answer, i.e. upvote if good, downvote if bad, flag if useless. See my answer here for my rationale.

2
  • 1
    Perhaps you might want to clarify or at least quote what is considered "good", "bad", or "useless" on this answer specifically, e.g. your rationale mentioning "proper attribution" which is also what SE staff has declared.
    – Andrew T. Mod
    Mar 27, 2023 at 4:53
  • @AndrewT. Good point, I was just referring to the content. Regarding the attribution, it's quite rare that large language models regurgitate content verbatim ai.stackexchange.com/q/38431/4 My guess is that large language models are less prone to plagiarism than the average SE user. Note that SE could add automated plagiarism detection, but AFAIK never cared enough to do so. I believe the SE staff in question just mentioned that one should simply attribute the large language model. Mar 27, 2023 at 4:59

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .