12
$\begingroup$

S.E. have added a new help page on the use of generative A.I. to write answers:

https://stats.stackexchange.com/help/ai-policy

I think what's lacking is any reminder that whoever posts an answer is responsible for its being correct & relevant; merely acknowledging generative A.I. use doesn't legitimate recklessly posting potential nonsense. We've discussed this in a couple of Meta posts:

What's the policy on Human/AI collaboration for answering questions?

What to do about answers written by ChatGPT

We can edit the help page: I propose we do that after consideration of proposed changes here.

$\endgroup$
2
  • 6
    $\begingroup$ For once, my upvote means "Good idea" not "Good question". $\endgroup$
    – Nick Cox
    Commented Feb 23 at 13:39
  • 4
    $\begingroup$ The above should be interpreted this way: On Meta, upvoting a question is generally to be interpreted as "This is a good question to ask" and not necessarily as agreeing with any proposal or viewpoint contained therein. $\endgroup$
    – Nick Cox
    Commented Feb 23 at 15:02

2 Answers 2

5
$\begingroup$

I agree with Scortchi - Reinstate Monica's assessment that merely acknowledging the usage of generative AI in one's post doesn't give them the right to open the floodgates of blindly dumping the content, irrespective of gauging the veracity and soundness of the same.

Thus, it should be explicitly stated as mentioned in the question.

(I suppose one might upvote this post to show agreement; but again this can be done by directly upvoting the, in words of Nick Cox, "Good idea".)

$\endgroup$
12
  • 2
    $\begingroup$ I upvoted this too. $\endgroup$
    – Nick Cox
    Commented Feb 23 at 15:00
  • 2
    $\begingroup$ +1. Perhaps we want to include links to the relevant Meta threads on the Help Center page. I personally would advocate including a very conspicuous notification along the lines of "If you post GenAI-generated crap, it WILL be downvoted and deleted, and you WILL be banned." $\endgroup$ Commented Feb 23 at 17:27
  • 1
    $\begingroup$ Sure @StephanKolassa. I might be bit of pessimistic when I say I can't find any potential answer to be useful where someone used generative AI. Once we have a collective consent on the proposal, we can come to agreement the final statement (I second yours) that would appear in the page. $\endgroup$ Commented Feb 23 at 17:34
  • 2
    $\begingroup$ I think or rather hope that @StephanKolassa is joking a little. We shouldn't need or seek new criteria for downvoting, deleting or banning. Trouble is that none of us would recognise a really good AI-produced answer as it would be indistinguishable from any other kind. But garbage is always garbage, whether generated by the utterly clueless, the utterly confused, or some unutterable code. $\endgroup$
    – Nick Cox
    Commented Feb 23 at 18:47
  • $\begingroup$ Perhaps we could tone down the insolent style of the statement Stephan has stated without disagreeing with the implicit sentiment. Sure, it would be too much to warrant a straight banning, but certainly the statement could indicate resorting to the practice of adopting generative AI content without proper vetting is discouraged and continued usage of such might lead to ---- @NickCox. $\endgroup$ Commented Feb 23 at 19:09
  • $\begingroup$ @User1865345 Inam not sure you find it a useful answer but in this answer I used the help of AI to generate the image. I am not such a code monkey that I know the ins and outs of igraph so I asked the computer to make an initial sketch of the graph and it also helped me answering questions about how to modify colours text locations etc. $\endgroup$ Commented Feb 23 at 21:07
  • $\begingroup$ @User1865345: I took the liberty of submitting both "f you post GenAI-generated crap, it WILL be downvoted and deleted, and you WILL be banned" and "resorting to the practice of adopting generative AI content without proper vetting is discouraged and continued usage of such might lead to" to textinspector.com. The first statement got a Flesch-Kincaide Grade of 8.39, the second 13.91. I personally would rather err on the side of understandability. Of course, we can tune it down. Also, please note that I do not advocate banning for GenAI content, but for GenAI-generated crap. $\endgroup$ Commented Feb 23 at 21:50
  • $\begingroup$ Sure @StephanKolassa, I am all in what you stated and agree with that we are not against the whole exclusion of GenAI content but rather the practice of dumping anything based on those without proper checking. $\endgroup$ Commented Feb 24 at 1:41
  • $\begingroup$ @SextusEmpiricus, that's actually an interesting use and the output of the Markov chain turned out nice indeed. $\endgroup$ Commented Feb 24 at 1:43
  • $\begingroup$ @SextusEmpiricus: your comment sounds like "any content crafted, in part or in whole, using a tool that writes a response automatically based on a prompt it is provided", so it looks like you should have credited AI? $\endgroup$ Commented Feb 24 at 7:42
  • $\begingroup$ @StephanKolassa I see now that my answer may not have been such a great example of the use of AI that is intended in this meta-question. Isn't crediting only for creative original content? If I make a post with the help of spelling control and grammar improvements and I made calculations with the help of Wolfram alpha and I used Google to search for information, do I need to credit all three programs? $\endgroup$ Commented Feb 24 at 8:58
  • $\begingroup$ @SextusEmpiricus: that is an excellent question. For that specific question, I honestly don't think you need to credit anything, since your use of GenAI rather obviously did not go much beyond googling for an example of how to create this graphic. But by the text of our help page, you indeed seem to have to credit it ("crafted in part"). It's hard to find the golden mean here. $\endgroup$ Commented Feb 24 at 12:36
3
$\begingroup$

Proposal (to be inserted after the 1st paragraph):—

Furthermore, by posting AI-generated content in an answer you vouch for its being correct and relevant; if you are unable to do so, do not post it. You wouldn't quote from a journal article unless you'd read and understood it—the same applies to quoting a ChatGPT response. We are, of course, tolerant of errors made in a good-faith attempt by a human to answer a question, but not of those negligently introduced by mindless duplication of GenAI output.

Revision 2:

Furthermore, by posting an answer you vouch for the relevance and accuracy of its content; if you are unable to do so, do not post it. We are, of course, tolerant of errors made in a good-faith attempt by a human to answer a question, but not of those negligently introduced by mindless duplication of GenAI output.

(The help page is quite long: we can explain the problems with GenAI in detail further down, if we want to; but need to provide clear & pithy guidance in the introductory section.)

$\endgroup$
7
  • 7
    $\begingroup$ "You wouldn't quote from a journal article unless you'd read and understood it" I hate to break this to you, Scortchi... $\endgroup$
    – mkt
    Commented May 31 at 19:54
  • $\begingroup$ But seriously, this is nicely written text on a complex topic. +1 $\endgroup$
    – mkt
    Commented May 31 at 19:56
  • 1
    $\begingroup$ @mkt: So you're saying people don't always read ... Oh! In fact that explains a lot. $\endgroup$ Commented May 31 at 20:07
  • 3
    $\begingroup$ @Scortchi-ReinstateMonica there is a lot of quoting without understanding. But when we quote a journal article, then at least there has been some scrutiny by others. With a generativeAI this has not been the case. It is like a non-peer reviewed article and by an author that is not very reliable or cares about telling the truth or using logic. $\endgroup$ Commented Jun 1 at 8:51
  • 2
    $\begingroup$ So I think that the particular sentence making an analogy with quoting a journal articles is not so strong. $\endgroup$ Commented Jun 1 at 8:53
  • 1
    $\begingroup$ An example would be to quote a particular theorem without understanding or having read it's proof. I myself may have quoted several times a result form Wolfram alpha without being entirely sure about it (but in that case I always make at least clear with a reference that it is based on Wolfram Alpha). $\endgroup$ Commented Jun 1 at 8:55
  • 2
    $\begingroup$ Doesn't the first sentence apply equally to content we post without AI? How about removing "AI-generated", and adding a sentence "This also holds for content generated with the help of AI." Per the discussion here in the comments, I would remove the current second sentence. $\endgroup$ Commented Jun 4 at 6:24

You must log in to answer this question.

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