24

A comment to one of my answers:

This was a really creative way to circumvent the problem. Bing GPT gave me the exact text that you wrote here(i searched and came here to give you the very well deserved upvote). I think that BING GPT is damaging, since people will not actually come to S.O. to view your answer, upvote or interact with S.O. in a meaningful way.

Chandradhar Koneti Multiple conditions in switch case?

If "Bing GPT" or any other "nonsense with superpowers" out there is giving the exact text of a Stack Overflow post without an attribution, isn't this some sort of violation of some sort of copyright?

So, apparently people are picking up the habit of asking A.I. oracles for answers instead of coming to Stack Overflow. I think I have noticed a recent severe reduction in the number of upvotes I am receiving on popular, old answers of mine. It is even possible that this can be seen in reputation graphs. Here is mine:

Reputation graph.  The plot is roughly linear, but exhibits a slight bend around 2023

Note that the line has started following a noticeably less steep slope since some unknown moment in 2023.

When I once tried asking ChatGPT a programming question, it gave me an answer that was surprisingly — no, let me reword this — astonishingly close to what I was asking for, and it almost worked, but working code is like pregnancy: there is no such thing as "almost pregnant"; code either works, or it doesn't. So, my experiment led me to the conclusion that A.I. is not really damaging to Stack Overflow. However, verbatim copying of code, especially without attribution, as the comment above attests, can definitely be damaging.

Are there any plans to do anything about this?

Can anything be done about this?

12
  • 1
  • 8
    The "Bing GPT" incident looks like an issue of merging search+GenAI without due diligence to check if the results are (almost) exclusively from just one component. An (almost) search should not pretend that results are generated; using GenAI to file off serial numbers is fishy... Nov 6, 2023 at 8:11
  • 10
    Sue Microsoft. Please.
    – Joachim
    Nov 6, 2023 at 10:49
  • 5
    @Joachim When the company gets off this bandwagon, I do hope they pursue this sort of thing: seems like class-action lawsuits might be a more reliable way to get revenue from “generative AI” than paying OpenAI money (a large chunk of their current strategy).
    – wizzwizz4
    Nov 6, 2023 at 12:20
  • 2
    That graph doesn't show a reduction in upvotes does it? Nov 6, 2023 at 14:29
  • 3
    If you want to get rep numbers (and, thus, rate of increase/slope) from a graph, it's probably easier to do that from your reputation graph on the main stackexchange.com site. However, given that you're active on SO (i.e. posting Q&A in the timeframe of interest), just looking at reputation isn't sufficient. To see an actual difference due to AI, your reputation gain must be adjusted for changes in your own participation. You definitely have different participation levels corresponding to the recent knee in the SO rep graph.
    – Makyen
    Nov 6, 2023 at 15:40
  • I should also note, that you can get the actual datasets for those graphs by looking at the raw data sent in the AJAX calls those pages make, rather than getting the data from the graphical representation.
    – Makyen
    Nov 6, 2023 at 15:42
  • @user1271772 The graph shows that the slope becomes less steep. The slope represents upvotes over time. So, it shows a reduction in the rate of upvotes. In other words, if I used to receive 10 upvotes per month, now I might be receiving 7 or 5 upvotes per month.
    – Mike Nakis
    Nov 6, 2023 at 16:32
  • 1
    Of course there could be other reasons for this, for example it may be that the technologies on which I answer questions are slowly becoming irrelevant, as the world is being consumed by script kiddies monkeying in scripting languages. I just wonder what others see in their own graphs.
    – Mike Nakis
    Nov 6, 2023 at 16:34
  • @Makyen yes, that's true. My incoming upvotes are mostly on old answers. Every once in a rare while I get in the mood for answering new questions, and that may be what caused the spike. In any case, I think the graph is receiving an undue amount of attention: I only included it for illustrative purposes; I do not think that the problem is us missing out on reputation; I think the problem is copyright infringement.
    – Mike Nakis
    Nov 6, 2023 at 16:40
  • 4
    My graph flatlined/became negative a few months earlier than yours, it's only began to recover recently as i've significantly reduced my participation, and thus stopped losing as much rep to downvoting low quality content.
    – Kevin B
    Nov 6, 2023 at 16:41
  • If the slope is less steep, it looks like that is because it's correcting for your anomalously high slope near where the horizontal axis has the "2023" label. The overall slope really does not look like it has changed much since 2020. Nov 6, 2023 at 16:54

1 Answer 1

8

Yes, copying without attribution is a violation of the license

Yes, if someone/some company is copying user generated content from Stack Exchange without providing attribution sufficient to meet the CC BY-SA license requirements (link is to an answer of mine on Meta Stack Overflow listing the requirements) and the use is not covered under "fair use" (or other similar copyright exemptions in jurisdictions other than the US), then they are violating the copyright of the content creator.

Only the copyright holder, or their agent, has legal standing to do anything

A major stumbling block is that only the owner of the copyright, in this case the user who created the content, or the copyright owner's agent has legal standing to do anything about it. So far, Stack Exchange/Stack Overflow has chosen not to include text in their Terms of Service which would make them the non-exclusive agent for each user, for the purpose of enforcing copyright/license violations, so Stack Overflow has no legal standing to actually do anything about the situation.1

Thus, only individual users can take any actual action to enforce the copyright and licensing conditions. That's time consuming and costly for individuals to do. While it might be possible to resolve a single such breach of the licensing terms by the copyright owner sending a Cease and Desist letter to the company that is responsible (i.e., Microsoft, per the question), getting actual resolution, and damages, would require filing a lawsuit.2 Even if that lawsuit was able to be filed in your local Small Claims Court, which the rules for your local court may or may not permit, it's still quite a bit more work than most users will be willing to do.

It, probably, is possible to try to directly initiate a class action suit against the company who is in breach of the licenses (Microsoft, in this case), but that's a major undertaking. Such lawsuits generally require huge amounts of time from lawyers and from the people who are primary in the lawsuit. Unless there's a massive settlement (possible, but unlikely), the only real winners of such lawsuits are the lawyers, who get paid for their time.


  1. This is complex. If the company really wants to be selling the user generated data to creators of AI models, then the company probably should at least seriously consider adding such language to the Terms of Service in order to be able to put legal pressure on the companies they want to sell the data to. Without the ability to do that, their leverage in selling the data is minimal and relies on those other companies' fear of judgements in either other lawsuits, other lawsuits creating precedence, future lawsuits about this content (which are unlikely without Stack Overflow taking the initiative), and/or, eventual, legislation on the subject.

  2. While a DMCA Takedown Notice might work, such notices are actually not valid in this situation, because DMCA Takedown Notices only apply to situations where it's some other individual user that has violated the copyright/licensing and posted the content on a service provider's system. A DMCA Takedown Notice informs the service provider that the copyright owner, or their agent, believes the content to be in violation of their copyright. If the service provider removes the content, then the service provider can't be held liable for the infringement (i.e., they get a "safe harbor" that's free from liability).

    In the situation in the question, the underlying structure of what's happening doesn't match those circumstances. It is the actual service provider choosing to infringe the copyright/license, rather than an independent individual user. Thus, the safe harbor provided by compliance with a DMCA Takedown Notice doesn't exist, so such a notice isn't applicable. It might, however, be sufficient to be considered a Cease and Desist letter.


I am not a lawyer. I'm definitely not your lawyer. This is not legal advice. It's only my opinion and understanding of the situation.

3
  • As you know, it's claimed that "Stack Exchange currently dual-licenses user content, which allows them to sell user content without attributing users".
    – PM 2Ring
    Nov 6, 2023 at 22:48
  • @PM2Ring IMO, it's difficult to argue against the Terms of Service, as they have existed since 2010-06-09, creating a second license. However, if the second license allows SE to sell a license for user-contributed content in order for others to make derivative works (e.g., AI generation models) without attribution to the original contributor is unclear to me. In earlier versions, they were granted the right to "create derivative works". In later versions, that wording was replaced by "commercially exploit".
    – Makyen
    Nov 7, 2023 at 2:09
  • 1
    While "commercially exploit", IMO, gives SE the right to make copies and sell those copies to other parties, and could be reasonably argued to give SE itself the right to create derivative works and sell them, it does not appear, IMO, to allow SE to sell or transfer the right to create derivative works, which is what AI-generation models are, IMO, and the right which AI-generation companies need. IMO, SE having the ability to sell a non-CC BY-SA license for the right to create derivative works is, at best, arguable or require legal structures more complex than a simple sale.
    – Makyen
    Nov 7, 2023 at 2:10

You must log in to answer this question.

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