12

SE Inc. wants to sell our content to some companies for AI training. However, some of our content is not allowed to be used to train AIs. E.g., since ChatGPT hasn't been banned on all SE sites, some of our content contains some ChatGPT output, which OpenAI explicitly bans (2c-iii) from being used to train some AIs.

This makes me wonder: Is SE legally allowed to sell user content for AI model training despite the fact that some of this content is not allowed to be used to train AIs?

4
  • 10
    That is an interesting business model: OpenAI is going to sue SE because SE happened to have ChatGPT generated content while that content could only appear on an SE site because ChatGPT was trained on data from the SE network.
    – rene
    Commented Apr 24, 2023 at 7:50
  • 9
    @rene the ouroboros model
    – VLAZ
    Commented Apr 24, 2023 at 8:26
  • The data is publicly available on the Interwebs... it is available for free, there is nothing to sell. Google has been using this for their search engine (which is at least partially AI-based afaik) since the dawn of SO.
    – Lundin
    Commented Apr 24, 2023 at 13:10
  • @Lundin techcrunch.com/2023/01/27/… Commented Apr 24, 2023 at 15:35

1 Answer 1

9

Welcome to licensing hell.

IANAL and this is my best interpretation, but:

  • Any content published on the Stack Exchange network is bound by CC-by-SA.
  • In order to publish something under this license, the site presumes that you are able to do so. Disputes to the contrary are handled through DMCA requests.
  • Stack Exchange can sell the data they have licensed under CC-by-SA because there's no restriction on being able to, and we're also giving them tacit permission to underneath the secondary license we provide Stack Exchange Inc. when posting content to the site.

So if someone put some ChatGPT content on the site that was then resold to ChatGPT? Someone should've put in a DMCA request about that situation since the company can't know a priori that it wasn't legally licensed to them.

12
  • Thanks. That line of defense ("we trust that our users respect the law/license/etc. so I'm not liable for anything about their content") didn't work well for RapidShare and a number of other sites. Also, since SE didn't forbid the use of ChatGPT, it means they are ok with users adding ChatGPT output in their answers. Commented Apr 24, 2023 at 15:55
  • 6
    Just wait 'till we get AI lawyers on the case....
    – W.O.
    Commented Apr 24, 2023 at 16:23
  • "Someone should've put in a DMCA request" How would that work? Only the copyright holder (or someone working for them) can issue a takedown notice with any legal force.
    – Laurel
    Commented Apr 24, 2023 at 16:30
  • 3
    @Laurel: Yes. That's the joke. The only way that this whole thing could work or even begin to work is that the rights holder has to be the one to take action. Stack Exchange Inc. has repeatedly stated that they're not in the position to do anything about someone taking your data without attribution, so I really don't think that them policing ChatGPT content fed to a third party is in their bailiwick, either.
    – Makoto
    Commented Apr 24, 2023 at 16:32
  • 2
    @FranckDernoncourt: There's no network-wide prohibition, but prohibitions exist in general across the network. Most notably one with copyright of ChatGPT itself, but...
    – Makoto
    Commented Apr 24, 2023 at 16:33
  • Not prohibited on all SE sites. Commented Apr 24, 2023 at 16:56
  • 3
    @FranckDernoncourt: Something about plagiarism does give a more blanket-esque prohibition, as plagiarism is not permitted anywhere on the network. And yes, the use of ChatGPT does require attributions.
    – Makoto
    Commented Apr 24, 2023 at 16:58
  • 1
    @Makoto attributing ChatGPT still doesn't make it ok to use to train some AIs, and using ChatGPT output while attributing ChatGPT typically isn't plagiarism. Commented Apr 24, 2023 at 17:01
  • 2
    @FranckDernoncourt: I should probably state my role. I don't really care about this awkward licensing mess that leadership has gotten themselves into, I'm only exploring the hypotheticals. Basically, attribution would be a way for content to self-identify as not fit for training, but it's really up to whomever's feeding their LLM to pay attention to that. It's not the responsibility of Stack Overflow Inc. to do much beyond their initial assertion of trusting that users are compliant.
    – Makoto
    Commented Apr 24, 2023 at 17:20
  • 1
    @Makoto " It's not the responsibility of Stack Overflow Inc. to do much beyond their initial assertion of trusting that users are compliant" -> en.wikipedia.org/w/…: "In March 2012, the Hamburg higher regional court upheld three earlier decisions that the file hoster could be held liable for publication of copyright protected material by third parties." Can't always put all the liability of users. Commented Apr 24, 2023 at 17:41
  • 1
    Isn't the actual content on SE "owned" by the respective authors? As with any published written work. The license only gives SE the rights to distribute it infinitely, they don't "own" the actual content and so they are in no position to sell it - rather didn't they agree to distribute it free of charge? And through the license model anyone is free to "scrape" the content, including adapting the content for pretty much any purpose, including commercial AI training. As long as attribution is given... I guess the AI could have a long list of attributions available to dodge that concern.
    – Lundin
    Commented Apr 25, 2023 at 6:42
  • 4
    @Lundin: CC-by-SA is crystal clear on this point. Works covered by this license can be used commercially, so long as the work is attributed, and if the sold content is remixed, it must also be distributed under the same CC-by-SA license. The selling part is not in question. It's the attribution.
    – Makoto
    Commented Apr 25, 2023 at 6:47

You must log in to answer this question.

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