It seems as if one user has been posting a bunch of answers generated by ChatGPT today, and I was wondering what Workplace.se's stance regarding AI-generated content is. Stack Overflow has temporarily banned ChatGPT-generated answers, but Stack Exchange has left it up to individual sites to determine what their policy regarding such answers is. Of course, if it ends up being allowed, it would have to be properly attributed/cited, but is it allowed? If so, to what degree? If not, what do we do if we find an answer that seems to have been written using ChatGPT or some other AI?
I don't think there is any SE site that explicitely allows such answers. There are just a lot of sites (like this) that haven't seen any (until now).
I flag and report AI generated answers on all sites I'm part of, whether they have an official policy about it or not.
We trust to create quality answers by asking experienced experts, to give their opinion not just based on abstract facts, but also based on their experience with those facts.
An AI cannot do that. Posting an AI generated text is in my view nothing else than copying from wikipedia, linking to a blog or posting an ad... it's garbage.
If the person asking the question wanted to have an AI generated text spit at them, they would have asked ChatGPT themselves. They did not. They came here. They deserve a real answer from a real human.
Unless we have a policy saying otherwise, I will remove posts that are clearly generated by a machine. Feel free to flag anything suspicious.
Users that want AI-generated answers right now can go to Bing, Bard, or Quora, which provide them instantly. Copy-pasting questions into a public AI engine and copy-pasting the answer back only helps people who haven't heard about these engines yet.
Workplace often invites creative solutions, which LLMs don't generate, they offer a blend of what others have said on the subject. There's a useful service, but it's not specific to SE.
The purpose of StackExchange is to provide human-generated answers, which have a chance to put new knowledge on the web, that hasn't been typed in and published before.
I believe it's better to do that one thing well. Maybe at some point, AI answers could be auto-generated sitewide, clearly distinct from human ones. They shouldn't be represented as user answers.
I would like to give an update on the situation. I have not yet fully formed my own opinion on the matter, but there are facts nonetheless:
SE has handed down a new set of policies, that make moderating AI generated posts nearly impossible. This is not a side effect or a policy accidentially covering AI generated posts. It is explicitely saying "AI generated posts are fine and must under no circumstances be moderated".
I think the voting of -250 and counting on a staff member post says all there is to know about how this new policy is going.
Since not having AI generated crap on our platforms is indeed a policy that mods and userbase agree on, it seems a very weird move on part of the company, to force their mods do not take any actions on this.
Some moderators are going on strike starting today.
There is no common guideline or understanding on this from our moderators here. This is not a group post, it is my take on the topic.
I can speak for myself, I am very unhappy not only what this new policy does to the SE network, but also how it was communicated to us moderators.
There were strong undertones of "the tools have a racist bias" including the implicit conclusion that what we do is racist, because we only blindly follow tools. That is obviously not the case, nobody does anything just because a tool says so. But no proof or even data was offered for the claim.
I am very much opposed to people telling me what to do, "or else you are racist", without actually giving me any data or explanation. I am open to correcting my behavior when educated about my behavior not being up to standard, but not with such a vague blanket sprinkling of "racist".
The fact that the public policy is not identical with what we were privately told on how to act is not a good precedent either. I am happy with being a company's policeman. Even if I don't agree with all the "laws". And I don't. But I am not happy with being a company's secret police, doing stuff outside the laws. The laws are for everybody. Especially the police. Because the police already has the factual power to do anything they want, it is super important that they are not above the law (or have secret agendas with the law).
So that said, why am I undecided?
Well, since march, we had one (in words: o-n-e) case. I deleted the post, I did not even suspend the account, I just sent a mod message to please not do it again.
So as a mod, it doesn't seem to be a big problem. We would have easily survived if I hadn't done that. So should I make a fuss about whether or not I deleted one answer?
And that is where I am now. Undecided.
I really don't care for the actual actions in the last months. But I don't think they were wrong or racist and I don't like how it was communicated. I also don't like what the policy will make of the rest of the network, where trying to gain green internet points is more tempting and more widespread.
So just to let you know, right now, we will not moderate AI answers, even if you flag them. Not because we don't want to or are too lazy. Because our corporate overlord has decided to tell us to stop immediately and one way or another, we will. Because that is the contract we have.
There are several problems with AI-generated content.
First, it's effectively a form of plagiarism. Copying and pasting something that someone else has written (even an AI) and presenting it as your own is fundamentally dishonest. Apart from the ethics of this, plagiarism has never been allowed on Stack Exchange sites.
This leads to the second problem: when you post an answer here, you're implicitly "promising" the OP (and future readers) that your answer is based on your own knowledge, research, and experience. Copying and pasting from ChatGPT doesn't do that - you're really not sharing your own knowledge or experience and didn't do any of your own research.
Third, AI generates plausible text, not necessarily correct (or even relevant) text. I've seen plenty of cases where a ChatGPT was subtly irrelevant or didn't quite answer the question (it answered something that was quite similar to the question but not the actual question). For example, I posted a question on Stack Overflow on why Python reference counted the Singletons for Boolean values when they're used. Someone posted a ChatGPT answer explaining that the Boolean values are Singletons whose reference count need to be incremented when they're used - all of which I had already mentioned in the question. In other words, it came up with a bunch of facts related to the question, but it missed the entire point of the question.
Also, on Stack Overflow, we've found that the answers generated by them are often factually wrong. For example, I recently read an article where an author looked himself up in ChatGPT; it claimed that he had a "complicated" relationship with another author and had once publicly accused said author of plagiarism. In reality, the other author was one of his closest friends, and he had certainly never accused him of plagiarism (publicaly or otherwise).
Fourth, this has already been pointed out, but if someone really wanted a ChatGPT answer, they could just go to ChatGPT themselves - they don't need us to do that.
I'm going to give a different perspective on this than the other 2 answers - I expect that it will get a little bit of pushback - but then I'm a firm believer in healthy debate.
First and foremost - to make it clear - I do not use any form of AI generation in my answers - everything you read from me is a labor of love from my fingers - and the ubiquitous random capitalization of my writing style is a pretty clear indicator that I'm a human with bad habits.
I am not opposed, conceptually, to AI generated answers.
Consider the following premise - AI Chat generation looks upon thousands and thousands of things humans have written, and tries to write a meta-answer, compiling that collective wisdom, selecting the common elements and then repackaging those common elements into something useful.
In many ways, this isn't much different to what we do
Now, does an AI have 'Life experience' - of course not. But if an AI can accumulate or parse the life experience of 10,000 people - does this invalidate a potential answer?
I would be opposed to someone automating an AI response as a way to game the system and increase their status/rep and get medals - but in the same breath, I don't think I would mind if someone took a question to AI, read the answer and went:
"Huh. I wouldn't have worded it like that... but that's a pretty good way of phrasing it" and then pastes that answer in.
Let me continue with this example:
Have we, as an SE community lost anything by allowing this? I mean, we have good answers, great answers and I've even got an accepted answer that probably has some of the most down-votes for an accepted answer (still waiting on my prize :D) - The strength or weakness of an answer is not contained within answerer, but within the democratic process of reading and voting on them.
If an AI Chat gives a good answer to a question and it helps the Asker with their problem - has our function not been fulfilled?
I appreciate that I'm talking high-level and conceptually and that should we allow these, the likelyhood is that we will be inundated by low-quality answers that put significant strain on the Mod team to filter out.
However, I also don't like the absolutism here of 'NO AI ANSWERS!'.
My sense therefore is that a good answer is a good answer, regardless of it's source. If it's a Human who has directly experienced something similar and found a novel way of navigating through a problem - Great.
If it's an AI that has taken the meta-experience of multiple different people and given us a unique insight based on that extraction - then I don't see this as a problem.