6

The meta site is promoting a banner which would be used to warn a prospective user that AI generated content is not permitted. It must be requested to be installed on a per site basis.

We’ve recently run an experiment to test a banner highlighting the “AI-generated content” policy on Stack Overflow. Next week we’ll be graduating the experiment that ran on Stack Overflow, and adding functionality to allow the "AI-generated content" policy banner to be enabled on Stack Overflow and all other sites in the Stack Exchange network. The variant group had no significant impact on answer rates; however, it did see a reduction in posts flagged for AI-generated content. Please see the post on MSO for more details on the experiment results.

This banner would only appear at the top of the answer post tool. This change would also remove a similar banner on the answer post tool warning us that a new user is being replied to (when applicable).

Do we wish to request this banner change for our site?

For those wondering about other discussions of the use of AI on this site I would recommend looking at Should there be a policy on History SE for the use of AI-generated answers?

6
  • Thank you for bringing this up in meta! We did discuss this in the moderators chat a few days back, and I personally have issues with it. However, it if some users might be in favor, it would be better if this were discussed and decided on amongst our users at large. If you all want it, this is ultimately your call.
    – T.E.D. Mod
    Commented Jan 18 at 21:34
  • A note to voters, to accurately record users opinion on this policy, limiting to a single vote might be advisable. A downvote and upvote combination is unnecessarily biasing...
    – justCal
    Commented Jan 24 at 14:50
  • Is there a SE post or chat somewhere discussing why AI-generated content is bad here? How would one detect questions or answers generated that way? Both Q & A are expected to include basic research; does it matter how it was gathered? A native speaker can tweak GoogleTranslate text and improve the result. Wouldn't the analogy hold with SE History users and AI-generated content? I haven't tried this stuff, but there are apps popping up everywhere that purport to use it.
    – bgwiehle
    Commented Jan 25 at 14:04
  • 1
    @bgwiehle The issue with AI generated history answers has been discussed on other meta posts, both here and on main meta. The main issue is, at least in the current state of the technology, AI just puts word together, resulting in creating fictional responses. History is about preserving facts, so AI generating false information that appears well written (thats what AI is good at, language editing) can mislead people into believing this false information. That puts it on exactly the same level as Holocaust or pseudo-science trolls in my view. I'll add links to the other discussions...
    – justCal
    Commented Jan 25 at 15:34
  • A good example of an analysis of a fake answer from an AI user: Is this excerpt from an AI-generated answer re Portugal in WWII based on any real events?
    – justCal
    Commented Jan 25 at 15:48
  • 1
    I got a clarification from the SE Devs/mods on exactly what is being added and removed, and added to the question here, to clarify to everyone what is being decided.
    – T.E.D. Mod
    Commented Feb 1 at 19:05

2 Answers 2

Reset to default
6

Since it seems T.E.Ds reply would fit as opposing implementing the new banner, we need a counterpoint to actually allow a consensus of opinions to be reached here.

For arguments sake, I will say that any tool we can apply to stem the flow of AI generated fake answers is a step in the right direction. A banner such as this would bring awareness to the user that might not be aware of the issues AI generated answers have with generating fake responses. People are using AI in various forms across the internet, but any do not understand its limitations in generating factual answers. These people might gain from the placement of such a banner.

A banner will not stop the Trolls use of AI generated answers. But it might aid the Moderators by placing one strike against such answers in the first place. Any user that posts AI content, after having seen such a banner warning them that AI generated content is not allowed, can immediately be treated as ignoring site policy , and becomes eligible for moderator action. (I can't speak for the moderators on this, but I have seen violating site policy as a posted reason on many suspended accounts over the years).

I'd particularly like to call everyone's attention to this bit from the post:

The variant group had no significant impact on answer rates; however, it did see a reduction in posts flagged for AI-generated content.

A reduction in AI posts sounds great to me!

  These would seem to be the advantages of implementing this procedure: Awareness to the uniformed, and as a tool to help moderators control the sometimes belligerent users of AI.

7
  • Not poking too hard here (because you're exactly right that there needs to be pro and anti arguments here to vote for), but it would be nice if this answer went past simply stating that doing something would be great (the earlier answer said that too), and made an argument that the benefit will, or even may, outweigh what will be lost. I wish I could remove my diamond on this comment. (waves hand) You did not see the diamond
    – T.E.D. Mod
    Commented Jan 24 at 14:40
  • I wish a moderator would have made this answer. As far as I know, the entire procedure for removing AI material and users is still hidden from (us) regular users. So I feel very awkward and ignorant trying to expand upon Information I have no access too.
    – justCal
    Commented Jan 24 at 14:43
  • Well, in that case, I may (when I get a moment) add the content I'm bummed isn't here.
    – T.E.D. Mod
    Commented Jan 24 at 14:44
  • Thanks. That would be helpful.
    – justCal
    Commented Jan 24 at 14:46
  • The problem is, as any experienced essay writer or debater will tell you, I can usually make a much better case for the position I personally disagree with. :-)
    – T.E.D. Mod
    Commented Jan 24 at 14:47
  • By the way, beyond the need for two answers for a policy decision, I personally would prefer a banner which can be placed on request, such as the need for sources banner we can currently flag answers with. But if I included that in my answer then it would be unclear what any votes were for...
    – justCal
    Commented Jan 24 at 14:54
  • Edited. Hopefully its an improvement.
    – T.E.D. Mod
    Commented Feb 1 at 21:24
0

Here's what I wrote when this first got mentioned in a chat. I think it stands well unedited here:

I'm interested in other opinions, because I know mine on subjects like this tend to be weird. Personally, I like the idea of doing something in general, but not this implementation.

First off, it looks like its only a banner posters get when posting an answer. It feels like half our problem is people using AI content in questions.

Secondly, I'm pretty sure anyone posting AI content into an answer as their own work knows they are doing something shady. A warning about not doing one specific shady thing seems unlikely to deter them much.

Third, it gets rid of the prompt to new users to check our help center, which seems a much more pressing need, and to be advice much more likely to be followed. Plus, using the old new user banner means a new user who dismisses it has at least seen the words "help center", so when prompted to look there by users, they can't really plead ignorance.

Now as a clarification, I thought they were talking about the prompt given to new users on their first post to visit the Help Center, but it looks like it instead will just replace the little "new user" banner on the answer tool that you get when replying to a new user's question. Also, they do claim that on SE it in fact seems to have reduced AI posts. I'm dubious, but also a believer in hard numbers over feelings. So perhaps my second and third points aren't as strong as I thought.

Just to reiterate, this isn't just a question of having the banner or not, but rather a question of **if we want the new user prompt/tag or the anti-AI prompt on the posting tool for answers. That's the choice.

4
  • Could you expand on "It feels like half our problem is people using AI content in questions"? Do you mean questions like history.stackexchange.com/questions/72278/… where the question is "An AI generated this statement; is it true or false"? Or is there some other use of AI in questions that I'm failing to think of?
    – Quuxplusone
    Commented Jan 29 at 19:00
  • @Quuxplusone- Yes, 100% that. There have also been some questions where we had to draw it out of OQ that the "confusion" behind the question arose from AI content, and there have been some where I've been suspicious but OQ never owned up to it (but of course if OQ can't reference the content that prompted the question, the question generally gets closed for that reason alone).
    – T.E.D. Mod
    Commented Jan 29 at 19:44
  • 1
    @Quuxplusone Another recent example of Question body generated by AI is this deleted question concerning the Vatican. That content scores at 98.36% fake. Our long-time troll even included a paragraph fake AI generated death rate statistics to support a recent Denialist question, since destroyed of course.
    – justCal
    Commented Jan 31 at 14:32
  • @justCal: Ah, I see the threat model now (although it's not T.E.D.'s model at all). Your attacker posts a "question" that's really an AI-generated essay containing a few affiliate links, like this one from the same attacker. An off-topic "question" like this will likely stay up longer than an off-topic answer, especially if (as in that case) it engages legit users to post answers. But the original sin there isn't "use of AI"; it's "posting spam to advertise your own site/storefront/channel, and not asking any actual question."
    – Quuxplusone
    Commented Jan 31 at 16:13

You must log in to answer this question.

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