• If bad AI content gets upvoted, can we use that as an opportunity to understand & improve our voting system?
  • Should we focus more on the fact that bad AI content is bad, and not that it is AI?


There are multiple reasons Stack Exchange site users want to ban generative artificial intelligence from the site. I pinpoint three dimensions to hostility to AI: It is bad, it is unethical, it is non-human. The last two I defer to different posts. This post focuses on it being bad. I ask, (1) ‘Is AI being bad sufficient reason for banning it?’, as well as (2) ‘Does that improve site content?’

Is the problem ‘AI’ or is it bad content?

AI content can be bad because genAI is pseudo-intelligent; but the capabilities of generative AI is a matter to be settled in research. Like Google or another information tool, any person who can claim they have had a good use of it, has grounds to claim it is sometimes useful. Acknowledging that AI is contingently, and not categorically bad, allows us to shift the focus away from a deflected problem: why is Stack Exchange content bad? Why do users post bad content and it gets upvoted?

(3) Poor quality writing is an incidental feature of AI, and not a categorical one.

(Don’t -> Do) shoot the messenger

Bad content should be downvoted. Does it matter the method how it was made, if the poster will be held accountable regardless, through their reputation? People ultimately choose what they post. Someone who posts a bad answer can be handled without distinction between how they came up with bad content.

AI is a placebo

AI shows us something bad about our site, that was already there. If bad content is getting upvoted, there is something wrong with how people are voting, regardless of AI being involved. Like the replication crisis in science, embracing an attack on the firmaments of an established system is a progressive tack, instead of suppressing it.

(4) An automatic system for “false positives” is a benchmark for site content quality and vote-meaningfulness.

Let’s embrace that fact and make our site better.

This excellent evaluation of the state of the voting system does not focus on the combined topic of AI-and-voting.

  • 1
    Related: meta.stackexchange.com/questions/54494/…. And on Meta.SO: meta.stackoverflow.com/questions/332549/…
    – Marijn
    Commented Jun 12, 2023 at 12:32
  • 12
    Not my downvote, and I don't think this is provocative; but I imagine the downvoters might feel that we have had this discussion many times before.
    – tripleee
    Commented Jun 12, 2023 at 12:41
  • 10
    To answer your title: yes. To the solution: no. Something needs to change around answer recognition alright. Perhaps we could have "endorsements" or "recognitions". But like reputation, everything with a number on it will be gamed. There are tens of thousands of users who wanted to post GPT-generated answers. People like that will also falsify their reputation or SME-ness.
    – CodeCaster
    Commented Jun 12, 2023 at 12:46
  • 4
    "Why not talk about voting instead? It’s an opportunity for a more productive, longterm change to the site." Do you have a suggestion? Voting is also hard to talk about, since not even moderators know who's voting for posts or why. It's also not going to address the problems with plagiarized ChatGPT, which are more about integrity than reputation.
    – Laurel
    Commented Jun 12, 2023 at 12:48
  • 2
    Perhaps it is a problem. In my post about the reputation system I suggested raising the reputation threshold for eligibility to upvote answers to ameliorate this. I kinda doubt it will be implemented, though, given current management, which seems hell-bent on pursuing engaugement (= participation as measured by metrics easy to obtain, misinterpret and manipulate) at all costs. Commented Jun 12, 2023 at 13:10
  • 17
    This isn't a bad conversation to revisit occasionally. In fact, It's kind of essential. It's just really bad timing, I'm afraid.
    – user50049
    Commented Jun 12, 2023 at 14:09
  • 1
    High rep users won't upvote too easily. It's the low rep users, in particular, those with association bonus (+100) who would upvote the posts easily because they don't know that site very well. They just upvote the post they thought was good,
    – Nobody
    Commented Jun 12, 2023 at 15:05
  • I don't follow your logic at all. Would you mind spelling out your thought process more?
    – starball
    Commented Jun 12, 2023 at 15:59
  • 2
    Weighted voting: meta.stackexchange.com/q/240700/282094 based on tag badges as well as reputation: meta.stackexchange.com/q/281818/282094 has already been proposed.
    – Rob
    Commented Jun 12, 2023 at 16:47
  • 3
    Does this answer your question? Empowering tag-badge holders part II - let's look at silver?
    – Rob
    Commented Jun 12, 2023 at 16:48
  • 1
    Voting alone won't help us because votes are also a limited resources. Answerers must invest more time in an answer by themselves somehow. Maybe a higher negative impact of downvotes. We could talk about that. Commented Jun 12, 2023 at 17:03
  • 1
    your title is about the voting system, but then you go on to suggest changes to the reputation system? Those are two very related, but still, "reputation system" isn't automatically implied by "voting system". So what are you really trying to get at here?
    – starball
    Commented Jun 12, 2023 at 18:55
  • 1
    what are you really trying to get at here? vote scores? reputation? The privileges that reputation unlocks? representation of content quality?
    – starball
    Commented Jun 12, 2023 at 18:58
  • 2
    Your recent edits suggest you actually wanted to have a different question answered than the one people read into what was posted. Instead of shifting the goal of this question now that it already has answers, please consider to take the time to prepare a new focused question that reflects what you want answered and post it when ready. Commented Jun 14, 2023 at 9:49
  • 1
    "This one is about changing the focus from blaming AIs intrinsically, to holding people that post AI content accountable in a manner indistinguishable from how we hold them accountable." Note this already happened since day 1 of the AI ban. The SO ban claims a low but not zero rate of good AI answers and cites it is a problem due to people "without the expertise or willingness to verify that the answer is correct prior to posting". Since the entire AI topic is a pain point, please leave out such the parts not relevant to your discussion. Commented Jun 14, 2023 at 10:09

5 Answers 5


I've spent significant chunks of my life -- measured in years -- thinking about this. I also have a personal perspective that is well-acquainted with not quite being able to do things as well as my able and neurotypical colleagues, so I'm in a good position to think about reasons that people would feel like their participation on its own merit would just be undervalued.

The voting system should be robust enough that, if attacked with sophisticated tomfoolery or targeted voting purely for the "lulz", that it can quickly normalize and ignore the noise. It struggles with this currently even with a lot of human help.

But the voting system can't really help issues that are systemic. If people are really struggling to find ways to see their good-faith attempts at participation on the site (perhaps to bolster their job hunt), then the product itself isn't providing them with enough varied ways to participate, or isn't sufficiently uplifting and bragging on the work of people that only participate around the fringes. Burying the visibility of those attempts won't help those users at all, or the custodians, or the site.

This is why lots of work went into helping people write better titles, better questions, better .. everything .., but it still hasn't been enough to close that gap.

Right now, AI does show an eventual promise of working almost as a cerebral prosthetic for people that need to overcome communication obstacles: be they not knowing a language very well or even aphasia like I have.

And, AI might even fill in that resiliency in the voting system in a few years. But AI isn't going to solve the problem of participation here being perceived as out-of-reach or too difficult any time soon. One can definitely see why you'd look at the reward system itself, but it's just a cog in a bigger machine.

If you scope just the voting system right now, you lose a lot of perspective about how it all works together, and I have to point out that the most vulnerable of these fringe users as well as the small minority of dedicated volunteers need to be served equally well as everyone else in these conversations as they move forward.

If we can't figure out the ground level of how the community of people that actually carry this water can be informed of, much even aligned with how the company expects to solve these problems, we're not ready to have the conversation yet.

When that relationship is repaired, then I definitely agree - we must make sure voting is still doing what it was designed to do in a very different time. But it's like ... way down the line, I think.

  • 17
    I should also point out that more people are mistakenly suspended for detected voting irregularities than ever suspended for GPT - and next to zero effort over the years has been extended into improving this because it was seen as too heavy of a lift. That might be a good place to start, but it's a very tiny corner of this picture.
    – user50049
    Commented Jun 12, 2023 at 14:40
  • 8
    I'm still a little miffed about having some votes reversed as serial voting after a user dumped a string of 5 or 6 low quality questions in a few minutes and I read them and downvoted them too quickly. There are things algorithms do (without the ability to appeal) that drive people acting in good faith away from the site. It would be good to reimagine the system from the ground up using lessons learned, even w/o implementing all of it. Comments, voting, reputation, privileges, badges, tagging, closing, deleting, elections all of it. If we could do anything, what would we do better than before?
    – ColleenV
    Commented Jun 12, 2023 at 15:04
  • 5
    @ColleenV I agree. It would be really great if everyone could get on board with something new called Stack Overflow again. I think it could even happen here! I'm just really worried about the inertia we're seeing now becoming the new normal and I hope that is overcome quickly. I hope the conversation happens.
    – user50049
    Commented Jun 12, 2023 at 15:12

Ignoring the legal and ethical problems for unattributed content for a moment, there's at least one other issue (there's actually a list of issues, but there's no point in re-iterating them here).

People get here to be helped by humans. By people who learned from experience or correctly interpreting the documentation what to do in a situation. If they had wanted a generated answer where they continuously have to tell the generator where it went wrong, they'd have gone to the generator themselves.

You don't fix that by adjusting the voting. That's fixing the wrong problem.

  • 2
    I wonder why people think you cannot legally use ChatGPT's output. OpenAI's terms of use are indeed pretty clear on whether you can or cannot use the content generated by ChatGPT. Commented Jun 12, 2023 at 14:44
  • 6
    @Tortliena Because more often than not, chatGPT will provide you with text or code blocks identitcal to work used as its training model without attribution. ChatGPT doesn't really "quote" nor provide source for its output.
    – Tensibai
    Commented Jun 12, 2023 at 15:17
  • 2
    @Tensibai Like people rarely tell their sources when they used one, either. On Worldbuilding SE, it's frequent people use pictures without stating they have the right to do so. Often they don't. And I'm skipping everything which is too short to be remotely claimable (was it print("Hello World") a famous programmer line?) Or fair use. Commented Jun 12, 2023 at 15:44
  • 7
    @Tortliena Legality is the wrong standard to apply when discussing ethics. Many unethical things are not illegal, or at least not illegal everywhere, and vice-versa. For many people in many cultures, presenting someone else's work as your own is unethical. If an AI wrote it, you didn't, so it's unethical to pretend you did. There's an additional but separate issue that current GenAI is not able to attribute its output to the relevant training data. Commented Jun 12, 2023 at 15:52
  • 1
    @BryanKrauseisonstrike I agree, but I'm talking about the legal problem this answer claims there is, not ethics. It's a lot better when the morale thing to do is backed up by law(s), but this answer (and other random online posts) claim said law definitely exists without proof or reasonable explanation, unlike some references I can provide against that. Guess it's because courts are still in the fog there 🧑🏻‍⚖️☁. Commented Jun 12, 2023 at 16:22
  • @Tortliena The problem is not about law, it's about SE rules as stated in the help-center about referencing. All GPT answers, by nature, fail to comply to that rule here.
    – Tensibai
    Commented Jun 13, 2023 at 7:00
  • 1
    @Tensibai That's again deviating the topic : It's then a SE rule problem, not a pure law problem. Commented Jun 13, 2023 at 7:03
  • 2
    Ok, I get what you mean, well, there's things pretty law infringing that is a problem for a bunch of code helpers too (not just chatGPT): even the fair use you stated above is only ok with proper attribution. With ChatGPT you can not even tell what the source was, so regardless of if people do it or no at the end of the day, there's a problem using its output as you can not really attribute the original author when you're supposed to.
    – Tensibai
    Commented Jun 13, 2023 at 7:49
  • @Mast_-_on_strike Thank you for your perspective - I respectfully would like to know, what do you think of my post’s assertions about the increasing ubiquitousness of AI, with regards to prohibiting it? Do you merely advocate the post cannot contain explicit material written by an AI, and beyond that, there is no regulation? In other words, the group on strike only demands the right to ban users who basically conduct the equivalent of “plagiarism” - claiming words as their own which are not? Because honestly, I’m totally on board with that. If so, I’m glad we could have this conversation, Commented Jun 14, 2023 at 8:56
  • so I could understand what the essence of the problem might be - because this is coming from somebody who loves AI, but still approaches it from a critical point of view, and does not at all think many of its answers are good - I agree totally that a) they can be genuinely junk and b) that they get upvoted, possibly for surface reasons like readable prose style and formatting and so on. Commented Jun 14, 2023 at 8:57
  • 1
    @hmltn No, there are a great many problems with generative AI (such as ChatGPT) posts being posted by users. Some of those are they're not accurate or useful to the author of the question at all. It's a system that will hallucinate when it doesn't know the answer or when there is no answer (like asking who won the soccer world cup in a year that didn't have one). this problem is not new, the Google AI research group published something similar about NMT in '18. Many answers look like a well-versed essay while containing bogus info.
    – Mast
    Commented Jun 14, 2023 at 11:28
  • 1
    There have always been users spouting nonsense, but it got easier to generate nonsense you could get away with. Because it didn't look like nonsense at first glance.
    – Mast
    Commented Jun 14, 2023 at 11:29

There are two problems here.

  1. Users who are actively gaming the system will use colluding accounts to upvote their own content; or several users will agree to form a voting ring where they upvote each other's posts regardless of their merits.

  2. Users who are not subject-matter experts will upvote what seems like a good answer. This is in some sense similar to social media account owners who like and forward links to content which they have not actually read. "Sounds good and might be useful" is a reaction which for some users translates to an actual upvote.

Under normal circumstances, the system is reasonably self-correcting most of the time, and the net votes end up in the negative for bad content. But if you have spent some more time here, you will have seen seen answers which are bad or incorrect which somehow ended up with dozens or hundreds of upvotes which turn out to be pretty much impossible to remove from the site using the normal curation mechanisms.

This exacerbates the problem, as some users will then regard the number of upvotes as a signal that they too should upvote.

This is not to say that there cannot be ways to fix the mechanics of voting (and ideas for how to do that are welcome); but in the meantime, we just have to live with the fact that the votes are only a rough approximation of quality.


Countering upvotes via experts does not address the most pressing problem of bad AI answers: that there are a lot of them. Even if experts have, say, 100x times vote power they won't have 100x vote volume. A "super powered" expert cannot handle significantly more AI answer volume than a "regular" expert can right now.

This is why bans such as on Stack Overflow are explicitly not about individual answer quality. They are a blunt instrument that can be wielded by non-experts, which significantly increases volume.

  • 1
    Vote volume is a problem, especially with the current system that limits the number of votes that you can cast on someone's posts before they get reversed. What's the point of an expert making an effort to determine whether a dozen or so AI posts are useful if they can only downvote a small number of the bad ones?
    – PM 2Ring
    Commented Jun 12, 2023 at 21:59
  • It was also a problem on Stack Overflow before ChatGPT (but on a smaller scale): A lot of answers on Stack Overflow are plagiarised and/or completely bogus (much more than people realise). But most go undetected, especially the code dumps (practically impossible to check for plagiarism), unlike ChatGPT-generated answers which are extremely easy to spot. Commented Jun 13, 2023 at 18:19

If bad content gets upvoted, doesn’t that mean the voting system is the actual problem?

No? How did you even come to that?

You seem to be missing that the system is a piece of machinery operated by people.

  • It might mean that the recommended guidance on how and why to vote is unclear to people.

  • It might mean that people are unable to distinguish fact from fiction in answers, including askers of questions, who by definition have a relevant knowledge gap.

  • It might mean that people are trolling (welcome to the internet)

  • It might mean that a person was unable (perhaps due to a knowledge gap- I.e. did not know how) to apply information in an answer to their particular scenario.

There is an imbalance in the system: downvoting is unlocked after upvoting (125 versus 15 rep). A lot of people don't have downvoting privileges (just look at the reputation leagues). So it's natural for that to manifest in more upvotes than downvotes. We could talk about changing that, but the voting system is tied to the reputation system, which is tied to the privilege system, so any innocent-looking change could come with a fair share of negative implications / unforseen consequences.

In relation to the above point, I've been mulling on the idea of feature-requesting that information about votes from users without voting privileges get displayed to all users- such as the ratio of upmod to downmod votes. But my thoughts on that are still in the works.

  • 1
    Aside from that tiny quote, the rest of the suggestion seems to be supported by the situations this answer describes. People who are unfamiliar, inexperienced, or disruptive to the site should usually be those of lower reputation, so weighing their votes less would lessen the negative impact. Commented Jun 12, 2023 at 18:51
  • 3
    I feel this answer is a bit unfair with the asker. By definition, the score was always supposed to reflect content quality, bad content should end up with a low score and good content with a high score. Individually people would differ in their views but collectively and in the limit of many trials, the voting would sort content by quality (somehow). I agree that the question is not clear, but this answer only highlights the noise in voting. For all practical purposes it still is used (by most users) as a measure of quality. Otherwise we could remove the feature right away. Commented Jun 12, 2023 at 19:51
  • 1
    @Trilarion I guess it didn't come out that way, because what you've stated is my implicit point. When people know how to use the system, it works. I'm questioning the asker's supposition that the system (in its mechanics and not in its human operators) is the problem. At least- that's what I assume is meant when the word "system" is used. Perhaps that understanding of mine is wrong.
    – starball
    Commented Jun 12, 2023 at 20:41
  • 2
    @starball I was worried my question would have an incendiary effect on people for whom the topic is important or sensitive; luckily, people were extremely polite and constructive, and thoughtful, in their answers - your post / comments are the only ones with a slightly offended air - if you agree that we all benefit just from understanding each other’s viewpoints and approaching one another positively in doing so, if you could just edit the tone of your post to clarify your point of view without subtly implying.. astonishment or something at my mental.. faculties - it would round out the Commented Jun 14, 2023 at 8:32
  • 1
    answers as all being genuinely interesting food for thought and a deeply civil, constructive and mutually beneficial conversation. I would thank you for this. Thanks. Commented Jun 14, 2023 at 8:32
  • 2
    @hmltn I'm not offended, but would like more clarification on your question. I asked some clarifying questions under your question post and would appreciate if you addressed them (I'm not sure if you have yet, though I notice you keep drastically changing your post). I've edit my post here to add more of my own thoughts.
    – starball
    Commented Jun 14, 2023 at 9:35

You must log in to answer this question.

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