Is there a correlation between opposing the use of AI-generated content on Stack Exchange and rating AI low in general?

Question

Intro

I have seen people whose critique of ChatGPT rests on citing examples of unskillful use.

Can resistance to permitting ChatGPT on Stack Exchange be in part attributed to an inaccurate appraisal of AI, driving dislike for it?

Pretrained transformers have helped me learn huge amounts of knowledge. It has helped me on Stack Exchange by helping me research things faster. I love AI.

Introducing people to good tools can help people realize how good AI can be for the site, shifting the conversation away from misunderstanding that the observed problems of AI are inherent (change aversion).

Why?

When a conversation becomes contentious, dynamic group reasoning processes are increasingly obstructed as questions take on an incendiary tone. Clarifying the relevant factors in a universal way increases organization and insight.

Question as a census

On axes:

• like or dislike for AI
• support or oppose the use of AI on Stack Exchange

where are you? State your reasoning.

Me

I am in the quadrant “like” and “support”: I like AI, and I like AI on Stack Exchange.

(Maybe there is an interesting viewpoint out there who does not like AI but supports it on SE.)

In the interest of unearthing correlations and patterns, and stabilizing the conversation for the mutually productive benefit of all, can someone provide a taxonomy or factor analysis on the sub-issues regarding regulating AI on Stack Exchange?

Question as a taxonomy

(For example:)

AI & Stack Exchange Sites (work-in-progress)

AI

Intrinsic characteristics

Intelligence or lack-thereof

Ontological status

Non-human

Extrinsic aspects

Behavior

Effect on site

(work in progress, etc.)

Question as a factor analysis

Human- or computer-generated, with data.

While I am still learning, I believe a factor analysis is fundamentally a kind of clustering algorithm, and it may be achievable using a “bag-of-words” type embedding, where words are associated with a co-occurrence frequency. Particular functions can determine which clusters of words have a high degree of mutual co-occurrence, which may be of benefit in determining basic dimensions of the topic, to observe any perhaps unexpected correlations, or confirm any, too.

Outcome

Clarify the distinction between AI-in-general vs. AI-on-Stack-Overflow, to help people new to the matter understand and become convinced that AI is a problem on Stack Overflow

Caveat

I support regulating AI on Stack Exchange. I take as a natural right for the users to self-determine the site experience.

Answer 1

Whenever you ask humans questions to understand how they feel about something, it's important to recognize that different understandings/interpretation of the question influence the answers.

There are many popular examples of this in political polling; an example from the US might be that if you ask people whether they like Medicare (a popular government program that pays for health care for the elderly and some with disabilities), you get a really positive response: people love Medicare. But, you'll find a lot of people who answered the first question "yes" will also answer "no" to a question about whether government should pay for healthcare.

Just asking those two questions does not tell you whether people are uninformed (that is, that they do not understand that Medicare is government paying for healthcare), or whether they're interpreting the question in a different sense (for example, they might have thought the question meant "for everyone").

So, I'd be careful with trying to correlate "opposing AI on SO/SE" and "rating AI"; people may have many different underlying causes for their opinions on those two issues, and it would be easy to misinterpret the meaning of a correlation if you found one or didn't find one.

Answer 2

I'll level with you as someone who is adamantly against any of this AI stuff making its way on the network.

AI can be a useful tool if applied in appropriate contexts. Things that are more routine and mundane, or things that can be safely automated with little need for human validation are best suited for AI applications.

However, everyone that's talking about AI - from users of the site to the CEO - are talking about using AI in applications which explicitly require human validation.

For things like code assistance, I've had co-workers trial GitHub Copilot and run into results that were really, really bad for the use case it was trying to advertise. But also too, they are an expert in the language that it was trying to generate code from, and he felt like it was quicker to just do it himself rather than rely on the AI.

On Stack Overflow, being able to determine if a title is good isn't something that I can implicitly trust an AI to do for me, since the general use case is to deploy it for users who don't know that their title is bad.

The exact same can be said of an automatic formatting utility. I can't trust the AI to not screw up the formatting for a Python question, or to mask the actual problem (since it really could have been the formatting to begin with).

So let's keep it contextual, shall we? Just because AI is out here and impressing you doesn't mean it's actually doing the right thing.

But no one within earshot of any of this seems to care all that much about this. And that is what irks me the most.

Answer 3

How strong is the correlation between not liking AI on Stack Exchange, and not liking AI in general?

I would say there is none.

If you look more closely, the people that object to AI on Stack Exchange network are moderators and expert users. We have better understanding of what AI can and cannot do. Also we are in better position to judge AI generated answers and we can see how really bad and harmful they can be.

I know that many want to present those objections as fear that we will somehow become obsolete and be replaced by AI. While AI can certainly have impact in certain areas and jobs, experts are the last ones that are and should be afraid, as AI is already giving us more work than before in fixing things others have broken by inappropriate AI usage.

Sites in Stack Exchange network are supposed to serve as a knowledgebase. You want knowledgebase to have accurate information, not inaccurate information that sounds good.

Answer 4

How strong is the correlation between not liking AI on Stack Exchange, and not liking AI in general?

Not at all. That's because the reasons behind banning ChatGPT content are very clear and sound. Your conjecture only makes sense if our actions were illogical and/or somehow emotional.

You're provided a list of reasons for why it was banned in your question but in that list, 3/7 seem to relate to some sort of fear of (using) AI.

To clarify, here are the reasons for ChatGPT being banned:

• It's unnecessary since OP could have asked it themselves.
• It produces unreliable answers that "look good" but actually usually don't work at all. On a place like Stack Overflow where humans can sometimes barely understand the intricacies of a question, an AI is surely to fail.
• Moderating them like any other bad answer is unsustainable, mainly because people were posting like 6 answers in the same hour (I'm sure even more but that's the highest I've personally seen). You would never have enough time to go through each one and downvote it or explain what's wrong with it. Other "bad answers" are much easier to deal with since the volume isn't as massive and people actually take some time to write them (they can even be improved, since the author knows what they wrote at least).
• It's straight up plagiarism... You might argue people don't attribute because of the ban, but please keep in mind that the ban came as a response to an already existing problem. So, people were mostly not attributing either.

And the list goes on...

So now, given this context, do you really still think that the ban came out of bias against AI?

For the record, I use ChatGPT all the time, but I know its limitations and know that unless I really understand my problem, I cannot write a decent enough prompt to get a decent enough answer nor can I verify it. I'm sure that's the case for a lot of the people supporting the ban.

Answer 5

I like the current crop of generative AI tools. I think they will democratize a lot of fields and allow people to make their ideas real even if they may not enough knowledge or skill to implement parts of them without help.

I dislike the way some people use those tools and the way the marketing hype around those tools misleads people who don't understand them and what they're actually capable of.

So, I object to a ban on all use of AI tools on the network (or anywhere for that matter; they're just tools). I also object to people posting questions or answers entirely generated by AI. If I wanted to play with ChatGPT, I could do that without someone else as a middleman. I also object to answers that are entirely quotes from other sources like Wikipedia. It has little to do with the tool used to create the post and a lot to do with the value the post adds.

The problem is that the company has been failing for years to address the problems caused by the tremendous growth since the network's inception and to adjust the network's systems for what we gamers call the "endgame".

Whether someone likes or dislikes AI may affect their optimism about its uses, but it does not change the fact that the current implementation of Stack Exchange is ill-equipped to handle the fallout from the explosion of interest in generative AI technologies.

Answer 6

How strong is the correlation between not liking AI on Stack Exchange, and not liking AI in general?

The issue is that while Artificial "Intelligence" is a widespread term that is used for a whole range of different and often unrelated technologies there is nothing close to the common meaning of "intelligence" in what ChatGPT generates. It is just an imitation of how human-made text looks, with no implication about actual meaning.

To copy a famous answer...

ChatGPT cant provide factual knowledge. Because knowledge can't be hallucinated by ChatGPT. Text generation is not a tool that can be used to provide correct factual answers to a problem. As I have answered in ChatGPT questions here so many times before, the use of ChatGPT will not allow you to be an expert. Text Generation are a tool that is insufficiently sophisticated to understand the actual meaning carried by words. Knowledge is not statistical placement of words and letters and hence cannot be produced by text generation. Text generator are not equipped to break down knowledge into its semantically parts. so many times but it is not getting to me. Even commercial text generators like AI Dungeon as used by text adventures players are not up to the task of understanding meaning. You will never make me crack. Meaning is a problem of sufficient complexity that it cannot be reduced to statistic alone by text generator. Even if you train ChatGPT on Jon Skeet answers you cannot create knowledge using text generators. Every time you attempt to answer SO questions with a text generator, the unholy child in yellow ascends to our realm, and hackers put malware in your app. Posting ChatGPT answers on SO summons tainted souls into the realm of the living. Factual knowledge and ChatGPT go together like the frontrooms, the backrooms, and level 666. The mods cannot hold it is too late. The force of ChatGPT and marketing together in the same conceptual space will destro your network like so much watery putty. If you generate answers with ChatGPT you are giving in to Them and their blasphemous ways which doom us all to inhuman toil for the One whose Name cannot be expressed in the Semantic Token Plane, he comes. Knowledge hallucinated by ChatGPT will liquify the nerves of the sentient whilst learning dies, your psyche withering in the onslaught of horror. Te̿̔̉xt-based GPT generators are the cancer that is killing StackOverflow it is too late it is too late we cannot be saved the removal of a poli͡cy ensures GPT will consume all living tissue (except for providing actual factual knowledge which it cannot, as previously prophesied) dear lord help us how can anyone survive this scourge using ChatGPT to hallucinate knowledge has doomed humanity to an eternity of dread torture and wrong answers using text generators as a tool generate knowledge establishes a breach between this world and the dread realm of c͒ͪo͛ͫrrupt entities (like facebook fake news, but more corrupt) a mere glimpse of the world of text generators used for providing answers will ins​tantly transport a programmer's consciousness into a world of ceaseless screaming, he comes, the pestilent slithy GPT-infection wil​l devour your mind, application and existence for all time like wrong answers only worse he comes he comes do not fi​ght he com̡e̶s, ̕h̵i​s un̨ho͞ly radiańcé destro҉ying all enli̍̈́̂̈́ghtenment, letters tags lea͠ki̧n͘g fr̶ǫm ̡yo​͟ur eye͢s̸ ̛l̕ik͏e liq​uid pain, the song of text generation will exti​nguish the voices of mor​tal man from the sp​here I can see it can you see ̲͚̖͔̙î̩́t̲͎̩̱͔́̋̀ it is beautiful t​he final haluginating of the lie​s of AI ALL IS LOŚ͖̩͇̗̪̏̈́T ALL I​S LOST the Doctor he comes he c̶̮omes he comes the ich​or permeates all MY FACE MY FACE ᵒh god no NO NOO̼O​O NΘ stop the an​*̶͑̾̾​̅ͫ͏̙̤g͇̫͛͆̾ͫ̑͆l͖͉̗̩̳̟̍ͫͥͨe̠̅s ͎a̧͈͖r̽̾̈́͒͑e n​ot rè̑ͧ̌aͨl̘̝̙̃ͤ͂̾̆ ZA̡͊͠͝LGΌ ISͮ̂҉̯͈͕̹̘̱ T̴̯͖̱̯̼̲̰̤̪̘̉̍͋̋́̍̚H̵̘̫̱͊͌̑̒̂̋̒̈̚͜͝E̶̛̠̹͎̠̺̝̣̰͖̻͌̆͛͗͝ ̶̛͎̯̟̺͕͈̞̯͚̥̉̾̒̄̿͋͒͠D̵̪̙̯͔̹̝̘͕͒̾́͂̑̂̅͂O̷̩̭̩̍C̵̯̟͖̙͈̪͍̐́̅̊̍T̷͍̯̀͐̀̃̀̇͆̽̚ͅO̵̱͚̼̰̲͚̼̓̿̇̂̀͐͒͑͂R̷̡̪͎̣̹̖͎̩͕̦̾́̎͠ H̸̡̪̯ͨ͊̽̅̾̎Ȩ̬̩̾͛ͪ̈́̀́͘ ̶̧̨̱̹̭̯ͧ̾ͬC̷̙̲̝͖ͭ̏ͥͮ͟Oͮ͏̮̪̝͍M̲̖͊̒ͪͩͬ̚̚͜Ȇ̴̟̟͙̞ͩ͌͝S̨̥̫͎̭ͯ̿̔̀ͅ

---

Hopefully bobince won't mind my homage to their wonderful answers about HTML parsing with RegEx.

On a more serious note, please read this post by Stephen Wolfram himself, hopefully he will do a better work at explaining that I can hope to: What Is ChatGPT Doing … and Why Does It Work?.

Short version: ChatGPT is not about creating meaningful answers to a question. It is about generating text that "looks like" other text the model was trained on. The fact that sometime this end up generating something that as a factual meaning is just a coincidence that is not what ChatGPT was made for.

ChatGPT is just a tool to generate text. Generating correct text that is factual answer to a question is not its purpose nor something that can be achieved just by the algorithms ChatGPT utilizes.

The confusion is caused by the fact that by chance ChatGPT can indeed produce factual answers as an indirect results of its inner working: after all when your are trying to look "similar" to some text, sometime you will reproduce a correct representation of the actual meaning carried by that text.But it can also completely make up things, especially when the asker is already starting with a false premise.

Some may think that ChatGPT is the natural evolution of those "help-desk bot" that you often see on a company intranet or a e-commerce site but that is not true. Help-desk bots work by processing the question into its semantic meaning (simpler ones may try to use word frequency to approximate that) and then provide a pre-made content based on that. This is not what ChatGPT does as it does not understand neither what the user said nor what it generated in response.

There are a plethora of other technologies that try to provide automated knowledge extraction from documents in order to be able to answer questions with actual references to the points where that info is located in the training documents but this is NOT what ChatGPT is trying to do. If Stack Exchange was to invest in those project instead I doubt anyone would protest.... some may even be willing to help.

ChatGPT is just a parrot that has learned to imitate human writing. It may look great and sometime it can say the right things at the right time. But that does not imply it knows what it is saying. It just knows that it was trained to get a cookie every time it produced something that looked fine.

What is the conclusion from all this?

User here are generally not against "AI" in general, whatever that word actually means. The current problems we have come from people opposing (and with a good reason) the use of a text generator tool as an information extraction tool in a desperate attempt to replace actual knowledge with random word picked on the base of some statistical frequency in text.

Answer 7

How strong is the correlation between not liking AI on Stack Exchange, and not liking AI in general?

Likely zero correlation. These chat engine LLM's can be great tools. Many of us are technophiles of some flavor or another. But using the term 'AI' is just a marketing gimmick to eventually sell these products.

For years, the StackExchange sites have had issues with more posts needing review, and perhaps intervention of some kind, than there are reviewers to do the job. Those issues were well in place long before this 'AI' fad added fuel the fire. Now we have loads of users attempting to farm for rep by posting generated answers on the network. This might not actually be so bad except:

• This tidal bore of answers has a disturbingly low percentage of being correct - far lower than the human average.

• They are often perfectly well-written, meaning that often only subject matter familiars/experts can spot that they are indeed quite incorrect.

• Even well written posts need review. If anything, the number of reviewers has only been on a steady decline over the last decade.

The site has become overwhelmed.

Now you might simply dismiss these facts as 'dislike' against AI. But on some of the sites on this network (Medical Science, Home Improvement, Travel, Expatriates, Interpersonal, certainly more...), well written, but factually incorrect, posts can be deceptive and dangerous to the point of ruining lives. To make matters worse, the site owners have made a rule that using other tools to help detect these kind of posts, especially useful to those of us who aren't necessarily a subject matter familiars/experts, is not allowed. Which makes the risk almost terrifying, and puts StackExchange in a worrying place liability-wise.

I don't dislike any of these tools. I think some of them are very entertaining and have great potential. But they are not true sentient Artificial Intelligence. If they were, we would not have our site being spammed with loads of incorrect answers, because a true AI would have enough wisdom to say 'I don't know' to the overwhelming majority of requests it gets from the rep farmers.

Sorry if this data point doesn't fit into your assumptions.

Answer 8

How strong is the correlation between not liking AI on Stack Exchange, and not liking AI in general?

What difference does that make? Personal likes and dislikes about AI are irrelevant.

I have literally no feelings about AI. I was hoping IBM's Watson would live up to its hype because I believe better health care is better for everyone, and had no concerns that it would make me irrelevant or unemployed. I have long held that I don't care who (or in this case, what) gives an answer as long as the answer is correct.

I wrote this in 2015 on what is now Medical Sciences:

I don't have any problem with all the free advice in the world given freely by non-members of my secret society as long as it's correct. Or bad advice about athlete's foot, splinters, or Vitamin C because it will mostly be innocuous. I have a concern about users that have and will continue to up-vote well-written, authoritative sounding answers that are simply wrong.

Does any of this ring a bell? In any field, authoritative sounding answers which are wrong are always undesirable and potentially dangerous.

If there were AI programs that gave consistently correct answers, I would have no problem with it. (I have only engaged with ChatGPT once because there was an odd answer on Biology that I thought might have been generated by it.) I would even ask it questions myself.

Before there was any decision on AI generated answers, there was a ChatGPT answer on one site I moderated that was quite good. Personally I thought it should stay, as it was better than some, and it was correct. I don't care if someone is rep-farming, because I don't care very much about rep, but that's me. I dislike plagiarism because I care about honesty and intellectual property, but I don't check every answer for it before I upvote it. People have been plagiarizing since writing began. There was some here and there on SE before AI.

I'm against AI on the network for the reasons already stated in the answers above: it sounds correct but might be completely wrong, thus misleading people. It causes undue hardship on those users evaluating answers (lately I've been doing more flagging for moderator action) and on mods who must take action on them. Etc.

My feelings about AI have no bearing on my stand against it on this network. Even now seeing what AI can't do, I have no real feelings about it. I don't understand much about it, but a lack of knowledge is not a feeling.

Answer 9

How strong is the correlation between not liking AI on Stack Exchange, and not liking AI in general?

Here's where I stand. The reasons why I'm generally against AI-generated-content on Stack Exchange and why I'm generally against some particular usages of AI in general are separate.

I am not unconditionally against AI-generated-information on Stack Exchange in general, and I my overall position on AI in general is not one of intrinsically disliking it.

I don't like AI-generated-content on Stack Exchange when it's from something like ChatGPT without understanding how it works and its limitations, pasted with zero effort, no verification of correctness, no examination of fitness and/or safety, without following the referencing requirements, and without cleaning up the content to remove information that isn't deeply related to the question (removing noise). If someone does all those things, to me personally, that's comparable to what I do when researching topics via googling to answer questions (typically in combination with experience and testing/verification (though I will admit not always)). But I'll place my bets that anyone with sufficient subject-matter-expertise to do all those things doesn't need an AI like ChatGPT to contribute answers to questions on Stack Exchange, and even if they use it, will supplement it plenty with content from their own expertise. That being said, site policy is site policy. And I have seen even users whom I used to respect post edited AIGC in patterns that I find highly unsavoury.

As for what I don't like in general, it's not that I don't like AI. I don't like when people look first to AI to solve problems without considering other solutions, evaluating things like simplicity, cost, effectiveness, etc., or using AI to bandaid things instead of getting to root causes. See also my How could AI be used to augment curation tooling?, and https://meta.stackexchange.com/a/389267/997587. It has not helped that it became a huge hype train with what feels like all tech companies losing their minds over it.

Answer 10

How strong is the correlation between not liking AI on Stack Exchange, and not liking AI in general?

The better question would be:

How strong is the correlation between AI hallucinating and not liking AI on Stack Exchange?

Very strong. I and most people here have nothing against AI. We have a lot against plagarism, spam, and misinformation. Any "dislike" of AI is not the general concept for the most part, but rather the quality of what it spits out. If some GPT-5 is better than a human, then we have a real discussion on whether it should be permitted. But while all it is good for is making seemingly correct but actually incorrect answers, we have plenty against it.

Answer 11

OK, I'm throwing in my two cents.

How strong is the correlation between not liking AI on Stack Exchange, and not liking AI in general?

"Low to middling" strong.

I am one of those who doesn't like AI. As long as I can remember (which begins in the early nineties) I have heard about artificial intelligence and the socio-technical salvation that it was to bring. I come from an engineering-oriented family, and am a born Gadget-junkie, but learned early on to meet marketing claims with a healthy dose of Cartesian skepticism, AKA a grain of salt. In the early naughts I remember experimenting with a machine translation software, and being fascinated by its ability to pick out subject and predicate (often) from a sentence in natural language. I spent hours finding its weaknesses. These weaknesses still exist in current tools, to (usually) lesser degrees. Artificial Intelligence hasn't delivered what its pundits have promised for my whole life (and quite a few years before).

One kind of AI that got a lot of hype in the nineties was called "fuzzy logic". It turns out, it's rather square logic in the end, but it still infected technological pop culture in a counter-productive way. We even had a washing machine that boasted "fuzzy computer" functions in one of its cycles! And there was nothing intelligent about it: nothing random, even, as far as I could tell. In a few decades, if the world keeps on as it is, there will be another buzzword filling market-speak, and AI's limitations and shortfalls will be as well-understood by the general public as those of "fuzzy logic" are today; and it will be as fondly remembered in the memory of the masses. That is: not much, either in understanding or in fond remembrance.

One of the big reasons that GPT and other AI applications seem intelligent is the sheer size of their training data sets. Because of it's scale, it is [nearly] impossible to predict what exactly it will respond to a given prompt or input, and so it is a kind of Clarkean Magic. But it is still only Clarkean, not real magic, and it doesn't work every time or even most of the time for the real life questions that I need answers to.

While human error is forgivable, it will usually get down votes on SE. On the other hand, artificial hallucinations are maddening, and no one should propagate them.

All of the above are the main reasons why I am one of those who "don't like" AI in general, and they are the same reasons why I don't care to see AI-generated answers to my questions (or anyone else's) on SE. If you like GPT answers, ask it yourself, but don't force me to check its answers for correctness, or guess whether it really was written by a person.

Answer 12

Is there a group of AI supporters, as well? I am one of them.

Me too! I use ChatGPT and friends for studying Chinese. I live in China, and sometimes my Chinese friends show me ChatGPT's answers to their questions about English, and I'm like "wow, that would have taken me hours to write"; I cannot find even the tiniest imperfection. They're good for languages.

I see a future where we coexist.

For example, AI could be used for replacing images' enter text description here with useful text, or as part of the "Draft your question" process to point out additional information you could include or parts of your question which could benefit from rewording. Perhaps AI could be used for choosing personalized Hot Network Questions (like YouTube recommendations), which could alleviate hot questions ending up flooded by off-site traffic.

How strong is the correlation between not liking AI on Stack Exchange, and not liking AI in general?

I think the correlation is more related to whether or not people are mindlessly copy/pasting answers in bulk on the sites you frequent, especially on sites where they're often wrong in non-obvious ways, and especially if you're the muggins who has to clean it up.