13
$\begingroup$

See Ban ChatGPT network-wide for reference.

Briefly, ChatGPT is a language model that uses the statistics of words in a training dataset to generate grammatically correct responses to plain language prompts.

That means you could take a Stack Exchange question, put it into ChatGPT, and with no effort have a lengthy text answer generated.

The main problems with ChatGPT for a site like this are:

  1. Attribution to ChatGPT is required, but most people using the content ignore this.

  2. Nothing ensures the output is accurate, it's just based on a statistical model that can go horribly wrong at times. Worse, even wrong output is written in a way that can seem accurate/knowledgeable. ChatGPT is very good at "faking" answers, which means it takes far more human effort to identify bad ChatGPT answers, while it takes almost no effort to post one.

  3. Nothing prevents ChatGPT from plagiarizing. It does not reference the material it is trained on, and very easily could lift entire passages from specific items in sufficiently niche areas.

  4. Even if the sentences it produces are not lifted, ChatGPT doesn't reference the sources used, which is a requirement for most answers here.

  5. No value is added. If someone wants to get an AI answer to their question, they can just go to ChatGPT and ask it themselves. Posting a ChatGPT answer is not really any different than pasting a question into Google and copying a list of the first several entries. If someone wants the first 5 results to a particular Google search they can just use Google.

For simplicity, I recommend we simply ban this content here. Moderators here have already removed ChatGPT-sourced content that infringes on existing site rules, and I doubt there will be many ChatGPT posts that meet our other standards, so this policy change won't really affect much except to make it clear and avoid any argument about what is and isn't acceptable.

$\endgroup$
10
  • 3
    $\begingroup$ ChatGPT generated content should be banned network wide. I agree, such posts should not be allowed here. $\endgroup$
    – theforestecologist Mod
    Commented Dec 12, 2022 at 16:02
  • 2
    $\begingroup$ Same for me. Well written, thanks for it. The main problem will be to detect this content. $\endgroup$
    – Chris Mod
    Commented Dec 12, 2022 at 17:00
  • $\begingroup$ @Chris I think it will be easier here due to us requiring references for most answers. If posters do go back through and edit references into AI-generated content, then that might be harder to detect but will also get around some of the biggest issues. There may also be software countermeasures, but I'd prefer to not discuss those publicly in too much depth. $\endgroup$
    – Bryan Krause Mod
    Commented Dec 12, 2022 at 17:07
  • $\begingroup$ @BryanKrause That is a good point. $\endgroup$
    – Chris Mod
    Commented Dec 12, 2022 at 19:05
  • $\begingroup$ Also I agree -- I have brought the practical implications to SE Meta as to how to go about this. Basically my question there was how to identify it; banning is one thing, recognizing/enforcing the ban quite another. $\endgroup$
    – AliceD Mod
    Commented Dec 13, 2022 at 8:17
  • 1
    $\begingroup$ Wholeheartedly agree with this. I'm quite active on Stack Overflow, and GPT-based answers have been coming fast and furious since the public version was released (end of November, I think?). The problem is identifying them. Once you've read a bunch you begin to get a fairly good idea of what they look like, and you can tell from the code provided (if you take the time to test it) that the answers are just plain wrong. Hopefully with the number of SMEs here we'll be able to tell what's bad from what's good, but it's going to involve a lot of community education to flag suspicious content. $\endgroup$
    – MattDMo
    Commented Dec 14, 2022 at 16:09
  • 2
    $\begingroup$ One light at the end of the tunnel is that the owners plan to turn the AI into a commercial service, AFAIK eliminating the free public version, so this wave won't last forever. $\endgroup$
    – MattDMo
    Commented Dec 14, 2022 at 16:11
  • 2
    $\begingroup$ Interesting Q&A on Artificial Intelligence that just happened to be in the HNQ bar next to this post. It explains how GPT works, and why the accuracy can be all over the place. $\endgroup$
    – MattDMo
    Commented Dec 14, 2022 at 16:18
  • $\begingroup$ I wonder if we should generate some kind of banner or similar publicly/explicitly condeming and banning its usage here on bio.Se. It wouldnt stop bad actors, but it might reduce its usage making a little easier to moderate $\endgroup$
    – theforestecologist Mod
    Commented Dec 15, 2022 at 13:19
  • $\begingroup$ I don't think there are enough "good actors" using chatGPT to post on SE to worry about it that much, and it's so low effort to post one that we can just inform them by removing the content. $\endgroup$
    – Bryan Krause Mod
    Commented Dec 15, 2022 at 19:21

1 Answer 1

Reset to default
4
$\begingroup$

Given the score on the post here and the lack of any expressed dissent, I think we can presume until further notice that the policy on Biology.SE is against AI-generated answers.

Personally, my plan will be to delete this content with a comment linking to this post, and I'd ask users to flag "in need of moderator intervention" for content they suspect is generated (please explain your suspicions if possible). I don't expect it will be necessary to suspend users unless they're posting copious quantities of generated content (for example, if it appears they are using scripts to directly submit content) or ignoring warnings - this is pretty much the same way we handle other problem content besides blatant spam or egregiously offensive material.

I think it's better that we not discuss too much openly what strategies we'll use to detect this content for now, and for users that are unsure I wouldn't worry about seeking it out - you can also flag unreferenced answers simply for being unreferenced and downvote wrong answers as you've always done.

$\endgroup$

You must log in to answer this question.

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