1

The use of ChatGPT and other services and tools to generate answers to questions on Project Management, or any Stack Exchange site, comes with risks. Please use these tools carefully.

If you do decide to use such a tool or service:

Attribution is needed, even for machine-generated content. The Stack Exchange network's rules about attribution are not just about respecting copyright, but also helping people to determine the validity of an answer and to understand how to use the content in other ways. Please see our guidelines on referencing material written by others for more information.

Review the content before posting it. These tools are not flawless. They can sometimes generate content that, on the surface, looks correct. However, a deeper analysis finds errors. A person asking a question may not have the necessary background to perform that analysis and determine that an answer is wrong or misleading. This also increases the burden on other users, especially if they are unaware that the content is machine generated.

Do not use the tools and services to generate large numbers of posts in a short time frame. Flooding the site with content is not considered civil behavior, especially with quality concerns around machine-generated content. Doing so increases the burden on the humans who take part in curation and moderation efforts to keep this site full of high-quality content. Be respectful of the time and effort others are putting in.

Plagiarism or flooding the site with low-quality content are actions that can lead to the suspension of your account.

For more discussions around the SE network, see this Meta Stack Exchange answer.

2
  • If you go down this path you will also need attributions for software spellcheckers, grammar checkers, thesauri, or any tool that makes word choices or semantic changes. Practical example: Did you cite all spelling and grammar suggestions or corrections made by your browser when posting? I understand the problem you want to solve, but that doesn't make an incorrect interpretation of copyright law or accusations of plagiarism factual. SE can do anything it likes under its terms of service, but legally indefensible definitions or accusations remain indefensible. Q.E.D.
    – Todd A. Jacobs Mod
    Commented Dec 31, 2022 at 5:30
  • @ToddA.Jacobs No, you don't need to cite all spelling and grammar suggestions. A single word or a small word phrase is different than entire sentences or paragraphs. But if you have concerns, you should raise them directly with SE staff, perhaps on SE Meta. The policy is quite clear from my perspective: copying content from anywhere requires attribution, no exceptions for machine-generated content. You don't copy large amounts of content when using spelling and grammar detection tools and no content is uniquely generated.
    – Thomas Owens Mod
    Commented Dec 31, 2022 at 10:32

1 Answer 1

Reset to default
0

Users are Responsible for the Content Quality of Their Posts

I agree with most of the key points in the original top post, such as:

  1. The error rate of many AI-generated materials.
  2. The need to edit and fact-check material.
  3. Not using automated tools to farm reputation.
  4. Avoiding low-quality content blighting our site.
  5. Not allowing users to spam questions or answers (see rep-farming).

However, as someone who works professionally with AI systems, I disagree with two of the points because they are related to specific tools as well as to end-user license agreements to which we are not privy.

  1. AI-generated text is not intrinsically plagiarism in a legal, scholarly, or literary sense although it certainly can be.

    High-end AIs and even high-end text spinners are generally plagiarism-free from a legal standpoint. Ideas cannot be plagiarized, and cannot be placed under copyright. Only unique expression can; there's a whole area of the law that deals with intellectual property, but while the use of AI-generated text (especially a low-quality one) might not pass all plagiarism checkers (e.g. Grammarly or Copyscape) it's not actually plagiarism in the technical or legal sense unless it's essentially a tract of unattributed material from another author.

  2. Most commercial AI tools do not claim copyright for output.

    While a few companies that I know of offer art-generating tools where the user has free use but the SaaS company retains the copyrights on the images, I'm presently unaware of any high-end text tools that do the same. Even if they did, we as a community are not parties to those licenses, and have no responsibility to police for copyright infringement. For example, unattributed quotes from the Scrum Guide are pretty obvious and should be attributed, but we aren't obligated to search the Library of Congress to ensure that every line or paragraph is original or cited regardless of whether they are written by humans or AIs. The burden here is on the posting user, and the community should use common sense.

  3. Attribution requirements are rooted in copyright law and licensing, and don't necessarily apply.

    Attribution should be required when quoting from a source, but there are all sorts of exceptions to this. As an example, Neil Gaiman doesn't have to attribute line and verse in his book Norse Mythology because:

    • The source material is all in the public domain.
    • He's not plagiarizing any works under copyright.
    • It's not a scholarly or academic piece.
    • The source material he's using isn't under a license such as Stack Exchange's use of a Creative Commons license that requires attribution.

    This is no different than if someone posted a generated cryptogram, crossword, or Sudoku puzzle. Unless copied from a copyright holder such as the New York Times or required by software license, such things are simply content and should be treated as such.

None of this is to say I agree with people rep-farming with such tools, or posting low-quality or highly derivative material. Nonetheless, I want to ensure we tilt at the right windmills (quality and abusive behavior) without getting side-tracked by legal fictions or scope creep in our collective responsibility to avoid "broken windows."

What the Community Can and Should Do

In my opinion, posts that are low-quality for any reason should be dealt with the same way. We deal with low-quality posts here all the time, and whether it's Google Translate, Grammarly, or some new tool such as ChatGPT our obligations are still the same:

  1. Downvote, flag, or edit low-quality posts.
  2. Flag abusive behavior such as flooding sites with low-quality or non-factual posts to limit the problem.
  3. Flag answer- and question-spam, and moderate repeat offenders.
  4. Handle real plagiarism through existing community and moderator processes, although eventually I think SE will have to implement tooling at the systems level to handle widespread or cross-site plagiarism (along with the inevitable false positives that such tools generate).

There are always edge cases, and that's what community and diamond moderation is for. Short of XAI becoming more commonplace, or specific tools creating unambiguous and specific problems, trying to treat this as a user-side tool-usage problem instead of a posted quality problem is going to quickly become an exercise in futility, especially as such tools evolve, improve, and (eventually) add value.

Let's not play whack-a-mole with evolving tech. Instead, let's stay focused on keeping the content quality of our site high. While there's a Venn diagram with a currently-large overlap between the two things, they are not synonymous.

See Also

4
  • 2
    "Attribution requirements are rooted in copyright law and licensing, and don't necessarily apply." - I don't agree with this. Our attribution requirements are independent from copyright law and are for the health of the site, not just copyright. (Related: Academic environments have a notion of plagiarism which is distinct from and independent from copyright.) Our attribution requirements relate to copying any material you didn't write yourself. Mods are not empowered to enforce copyright law, but they are empowered to enforce the attribution requirements.
    – D.W.
    Commented Dec 29, 2022 at 22:28
  • 1
    By suggesting dealing with ChatGPT answers the same way as we deal with any other low-quality post, this post does not seem to engage with one of the big challenges of dealing with ChatGPT. One of the major challenges is that ChatGPT breaks our current systems for quality. Right now we rely on voters to vote. However, voters often vote based on superficial impressions. ChatGPT breaks the correlation between superficial impressions (e.g., does it look like it is well-written?) and substantive correctness (are the claims correct and useful?).
    – D.W.
    Commented Dec 29, 2022 at 22:34
  • 1
    As a result, ChatGPT posts often end up being low-quality but highly upvoted. So while your arguments make sense in principle -- if we had reliable mechanisms for detecting low-quality answers and downvoting them, we wouldn't have a problem -- in practice, ChatGPT breaks our current mechanisms for dealing with low-quality answers. Also, I think there is value in setting expectations. There have been a significant number of answers from people who don't have bad intentions, but just aren't aware of the harm their actions have caused.
    – D.W.
    Commented Dec 29, 2022 at 22:35
  • Per SE staff, unattributed machine-generated content is not allowed on the network.. Network-wide policy is to consider unattributed machine-generated content as plagiarism and use the standard actions to handle it. I'd even consider it worse than traditional plagiarism - not only does it perpetuate garbage-in/garbage-out should other algorithms be trained on SE content, but the fact that it looks legitimate on the surface only makes it more dangerous should be it used by someone who doesn't know better.
    – Thomas Owens Mod
    Commented Dec 30, 2022 at 1:23

You must log in to answer this question.

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