One of the things that happens with any new pool of cultural value is that people who create it want a social contract for it. Creative Commons licenses are often used for this. Which is wonderful, because that’s what they’re for. They provide clarity around norms and expectations. But people often want more.

The Creative Commons licenses were written at the start of the century and have been updated several times since then, improving each time IMO. They’ve also taken account of developments in the law over that time, carefully and in-keeping with the intention of the licenses (do not ask me about the EU DB right lol).

They grant additional permissions that are not automatically available under copyright law (and some associated laws). As legal documents, they take from the author to give to the public. The author is part of the public, so they also benefit in sum. Increase sharing, reduce enclosure.

They do not add restrictions over baseline copyright law. And this is where the tension with their use as a social contract comes in. Because people often want to exclude bad actors, variously conceived. They want to restrict them.

But the license modules don’t do that unless your definition of “bad” broadly lines up with the four-factor test for fair use in US copyright law (I have it on good authority that this wasn’t by design, it’s just my personal mental model).

This is because the licenses are copyright licenses that use the strength of copyright law’s restrictions against them, opening up what they close down. This is the free software license strategy applied to cultural works. And if you don’t want anything else, it works so well it’s almost invisible.

I’m not trivializing the “anything else” category here. It includes personal reputation, politics, clean water, healthcare, the sacred, everything other than paperclip maximizing the freedom to remix. I take this seriously, and I’ve helped work on serious solutions to it. But AI is not part of this “anything else”.

One answer to “I use open licensing because I want to be read. Google proposes to ingest my work for free and then not show it to anyone. Why would I consent to that?” is “Get back to me when you stop using a NonCommercial license and apply it to your most recent books.”

Another answer to “I use open licensing because I want to be read. Google proposes to ingest my work for free and then not show it to anyone. Why would I consent to that?” is “Photocopiers copy your work without reading it, let’s ban them!”.

“Ingesting” sounds bad. If we wish to mediate the harm of “ingesting” using copyright licenses, then we need to point to the act restricted by copyright law that “ingesting” performs. What is this act? Is it reading? Is it remembering? Is it learning?

It is training an LLM.

OK, so how do we train an LLM? We copy lots of texts onto lots of SSDs, copying them over the network to do so. We then copy them into a computer’s memory, copy that to the cache, copy that to a GPU, and run some math on it. Lots of copies! Copyright law (still not a lawyer) should be fine here!

But copyright law restricts particular acts and/or contexts (have I mentioned that I am not a lawyer) of copying. And what non-lawyers, particularly those used to digital media or software development, don’t get is that Common Law is a poetry slam, not a recipe book. Genies are less capricious.

Do people like LLMs being trained on their copyrighted materials? Some don’t. Can they express what the problem is in a way that a copyright license can be used to prevent that and only that, or at least prevent that and only an acceptable (to them) number of other cases?

Mayyyyyyybe?

I spent a lot of time helping to improve the NC licenses despite not liking NC as harm reduction and epistemic modesty. I spent years explaining OSM’s and CC’s positions on the EU DB right to each other. I did all of this as a community member (disclosure: I am proud to have later worked for CC).

But I don’t think that the problem with LLMs is well conceptualized, and I don’t have to reach for “what happens when I remember a CC-licensed novel I read and I have Neuralink installed” to establish that.

Because unless we can express the problem in terms of rights restricted by copyright law that LLMs uniquely traduce either the reserving or the returning of, we are going to cause more problems than we can possibly imagine.

And we can’t because they don’t. They just don’t. “Oh but they can copy my work” so can a pencil. “Oh but they store my work” they do not. “Oh but they can produce work in my style” that’s not an act covered by copyright, and so can a pencil. “Oh but it’s industrial scale copying” under fair use.

Do LLMs displace the market for your work by copying your copyrighted materials without permission or exception? Possibly! Do they displace-the-market-for-your-work-by copying-your-copyrighted-materials-without-permission-or-exception? They do not. But am I making a teleportation argument here?

A teleportation argument is “I did not rob the bank, I just teleported the contents of its vault into my truck”. Or “I did not copy that song file, I reassembled it from encrypted fragments sent over the network separately”. It’s a logical fallacy in which means and ends are confused.

It’s the opposite of “let him have it”, it’s “I did not shoot him, I merely moved my finger causing kinetic energy to be imparted to a projectile that some time later encountered a block of organic matter in an exchange of inertia”. But without realising.

So “OK but which restricted act of copying does the culture theft machine perform” meets “it doesn’t matter how the bad thing happens, it happens and we must deal with it”. And we all talk past each other and our own contradictions.

If I’m so smart; is there a problem, can it be addressed with licenses-as-social-contract, and is doing so better or worse than an embarrassed silence?

There is a problem but it is not captured by “stop Google copying my work without reading it”. It is a political problem wayyy larger than this.

This cannot be addressed with licenses-as-social-contract. Because it doesn’t have any levers to pull that won’t knock over actors other than Google (Google??!!).

And attempting to do so in this instance will snatch defeat from the jaws of victory for the free culture movement. It will do this by failing, by disappointing people, and by rendering the licenses less effective elsewhere. This is not going to be popular with people who quite reasonably want a social contract for their community.

But I mentioned free software licensing earlier as the obvious inspiration for free culture licenses. And so what about “the last free hacker”? If someone wants to withdraw from cultural circulation to work as they wish, why shouldn’t they?

Because this is a key difference between culture and software.

It’s time for some game^H^H^H economic theory. Executable software has both use and exchange value. It is a tool. Culture only has exchange value. Reading isn’t use in this sense. No, your textbook and your reference work are not tools *at the moment of your reading them*.

An index is a tool. Reading it provides immediate utility. You forget how much people hated search engines compared to card indexes…

But, anyway, yes, nobody is stopping you being the last free hacker of anything by training an LLM on your work. You are however stopping them by claiming you can restrict their reading.

The American and French legal models of cultural value get mixed up here. The surplus value of my soul-expressions are being economically alienated from me, therefore I must be paid and the alienators must be coerced by the state. Pick a lane.

Or don’t, and generate more clicks than light.

What if I told you that getting mad at LLMs for copying your work (sic) when you tell them to is three you problems in a trenchcoat? What if I told you that taking this out on well-written free culture licenses is going to help the forces you wish to oppose? What if we banned photocopiers or VCRs?

If we forget history to try to 0wn the tech bros, we can cut off our noses to spite their faces. If we remember it, we can make this a non-issue. So how do we make it a non-issue?

Open weights.

“Open” is an easily critiqued term here, but let’s take it at face value in order to make it real.

LLMs can and must be part of the commons.

Can we use copyright licenses on training data to effect that? Well, no, doing so requires imposing restrictions that are probably unenforceable (still not a lawyer).

Closed weight companies are trying to create “copying our weights” as a capital crime. This looks like a choke point, but beware “rightswashing”; neutralizing a non-existent legal right in such a way that gives it mindshare and effect. Fuck sweat-of-the-brow-copyright in particular.

So LLMs must be part of the commons. And culture must be part of the commons. CC licenses solve the latter. If you’re using them for promotion then that’s fine but you’re still part of the commons. It’s a bargain you knew you were making.

What about the former?

I don’t know.

Nobody does.

But LLMs are this generation’s VCR, or photocopier, or printing press, or sampler, or BitTorrent. And they must be a tool for the public, they must be part of the commons. That is the compromise position. And it is best for everyone.

Cytolysis is the strategy here, more-open-than-thou, not trying to close off the commons to sic it to last generation’s corporations as this generation’s corporations get to write the laws that will lock you out of your own culture.

Not free culture or open weights, but free culture and open weights.

LLMs must be of the commons.

(And, finally, still not a lawyer. Any mention of CC is personal opinion based on public knowledge. I am not currently employed by a tech company and have never worked in AI.)

Adapted from: https://bsky.app/profile/rhea.art/post/3muag34w3zs2q