29 Sep 2024 Search

Tantek Çelik suggests that Creative Commons should add a CC-NT license, like the existing Creative Commons licenses, but written to make it clear that the content is not licensed for generative AI training. Manton Reece likes the idea, and would allow training—but understands why publishers would choose not to. AI training permissions are becoming a huge deal, and there is a need for more licensing options. disclaimer: we’re taking steps in this area at work now. This is a personal blog post though, not speaking for employer or anyone else. In the 2024 AI Training Survey Results from Draft2Digital, only 5% of the authors surveyed said that scraping and training without a license is fair use.

Tantek links to the Creative Commons Position Paper on Preference Signals, which states,

Arguably, copyright is not the right framework for defining the rules of this newly formed ecosystem.

That might be a good point from the legal scholarship point of view, but the frequently expressed point of view of web people is more like, creepy bots are scraping my stuff, I’ll throw anything at them I can to get them to stop. Cloudflare’s one-click AI scraper blocker is catching on. For a lot of the web, the AI problem feels more like an emergency looting situation than an academic debate. AI training permissions will be a point where people just end up disagreeing, and where the Creative Commons approach to copyright, where the license deliberately limits the rights that a content creator can try to assert, is a bad fit for what many web people really want. People disagree on what is and isn’t fair use, and how far the power of copyright law should extend. And some free culture people who would prefer less powerful copyright laws in principle are not inclined to unilaterally refuse to use a tool that others are already using.

The techbro definition of fair use (what’s yours is open, what’s mine is proprietary) is clearly bogus, so we can safely ignore that—but it seems like Internet freedom people can be found along both axes of the fair use alignment chart. yes, there are four factors, but generative AI typically uses the entire work, so we can ignore the amount one, and we’re generally talking about human-created personal cultural works, so the nature of the copyrighted works we’re arguing about is generally similar. So we’re down to two, which is good because I don’t know how to make 3 and 4d tables in HTML.

  Transformative purist: work must be signficantly transformed Transformative neutral: work must be somehow transformed Transformative chaotic: work may be transformed
Market purist: work must not have a negative effect on the market for the original Memes are fair use AI business presentation assistants are fair use A verbatim quotation from a book in a book review is fair use
Market neutral: work may have some effect on the market AI-generated ads are fair use AI slop blogs are fair use New Portraits is fair use
Market chaotic: work may have significant effect on the market for the original AI illustrations that mimic an artist's style but not specific artworks are fair use Orange Prince is fair use Grok is fair use

We’re probably going to end up with alternate free culture licenses, which is a good thing. But it’s probably not realistic to get organizations to change their alignment too much. Free culture licensing is too good of an idea to keep with one licensing organization, just like free software foundations (lower case) are useful enough that it’s a good idea to have a redundant array of them.

Do we need a toothier, more practical license?

This site is not licensed under a Creative Commons license, because I have some practical requirements that aren’t in one of the standard CC licenses. These probably apply to more sites than just this one. Personally, I would be happier with a toothier license that covers some of the reasons I don’t use CC now.

  • No permission for generative AI training (already covered this)

  • Licensee must preserve links when using my work in a medium where links work. I’m especially interested in preserving link rel=author and link rel=canonical. I would not mind giving general permission for copying and mirroring material from this site, except that SEO is a thing. Without some search engine signal, it would be too easy for a copy of my stuff on a higher-ranked site to make this site un-findable. I’m prepared to give up some search engine juice for giving out some material, just don’t want to get clobbered wholesale.

  • Patent license: similar to open-source software license terms. You can read my site but not use it for patent trolling. If you use my content, I get a license to any of your patents that would be infringed by making the content and operating the site.

  • Privacy flags: this site is licensed for human use, not for sale or sharing of personal info for behavioral targeting. I object to processing of any personal information that may be collected or inferred from this site.

In general, if I can’t pick a license that lets me make content available to people doing normal people stuff, but not to non-human entities with non-human goals, I will have to make the people ask me in person. Putting a page on the web can have interesting consequences, and a web-aware license that works for me will probably needs to color outside the lines of the ideal copyright law that would make sense if we were coming up with copyright laws from scratch.

More: notes on Humans Commons