Cory DoctorowToday's threads (a thread)
Inside: Supreme Court saves artists from AI; and more!
Archived at: https://pluralistic.net/2026/03/03/its-a-trap/
1/
Arvīds Kokins> [..] all seek to establish that training an AI model is a copyright infringement. This is wrong [..]
This is such a strange hill to die on. Of course "training" infringes - it's compression.
Framing it as "mathematical analysis" or collecting "facts" is disingenuous.
Fourier transform could be easily described the exact same way, except it's fully reversible.
"Facts" wrt copyright are not meant to be the building blocks needed to reconstruct a work to some degree.
> "Facts" wrt copyright are not meant to be the building blocks needed to reconstruct a work to some degree.
Literally you could write a book describing each brushstroke needed to reconstruct a copyrighted painting without violating that painting's copyright.
@archo Further: the cases against models are only tangentially related to memorization. It's undeniable that these suits would exist even if models had "guardrails" that *perfectly* prevented them from reproducing their training data verbatim. Memorization is *not* the crux of any of these complaints - *training* is.
@pluralistic Perfect memorization is also irrelevant - lossy compression is also a thing, doesn't stop copyright from applying (otherwise reencoding videos would remove their copyright).
Ultimately though what matters for infringement is whether the original work was substantially used to produce the new work (regardless of the exact process) - and clearly that's the case.
Not even the AI companies dispute infringement, they argue that it's "fair use" instead.
> Ultimately though what matters for infringement is whether the original work was substantially used to produce the new work (regardless of the exact process) - and clearly that's the case.
This is completely incorrect.
You could cut a picture into a billion pixels and reconstitute them as a completely different picture without infringing anyone's copyright.
Copyright has nothing to do with whether the constituent components came from a copyrighted work. That's just completely wrong
@archo Internationally, the limitations and exceptions to copyright are cabined by the Berne "three step" test which does not mention whether the constituent parts are drawn from a copyrighted work.
In the US, there's Fair Use's "four factors," one of which does interrogate the amount of the taking, but is not in itself sufficient for a finding of infringement.
@pluralistic In that case you can argue that the particular work was not substantially used in the end result, if all you needed was a pile of pixels (or a palette) without their positional association to each other.
To me, cases like this sit far outside the "substantially used" circle.
AI "models" on the other hand, explicitly consider the positional associations of the pixels of the original images at various scales, not just their color.
@archo You could produce a painting that consisted of the "average" face by measuring and averaging out the features of every face in every painting extant and it would not infringe copyright.
@pluralistic If it's just measurement and not color, features or anything else, then probably yes.
Aggregate functions are naturally destructive - the source data doesn't come through almost at all. Very much unlike autoencoders.
@pluralistic That doesn't make sense. That's what most computer data effectively is - instructions to reconstruct data in a more accessible format. IDK if you're familiar with art tools but some literally encode the brush strokes (like Substance Painter or Clip Studio Paint vector layers) needed to reconstruct the picture.
@archo Yes, there's some interesting edge-cases implicit in this new, more explicit rule that copyright inheres only for human (and not machine) work.
While it's unlikely a case regarding the copyrightability of a vector layer (or even a clone brush) would reach a judgment (such cases being likely to founder on the *de minimis* standard that excludes trivialities from legal action), such a judgment might assign a "thin" copyright comparable to a "selection" right.
@pluralistic Hypothetically, if someone were to extract all the brush strokes from a file, wrote them into a book, retyped that book back into a file and re-rendered them to claim that it's original work, to me that's still a clear-cut infringement, not even de minimis (which could apply if only e.g. less than 0.1% of the brush strokes were used).
At the end of the day, copyright is an economical tool to ensure the author is fairly rewarded for their work, not an obfuscation challenge.
> retyped that book back into a file and re-rendered them to claim that it's original work
You are describing an infringing *use*, which does not make the machine itself infringing. This is a bedrock of copyright law, established in 1984 with the Supreme Court's ruling in Sony v Universal (the Betamax decision), which I referenced in my thread: "A device capable of sustaining a substantial noninfringing use is not infringing."
> At the end of the day, copyright is an economical tool to ensure the author is fairly rewarded for their work
Categorically untrue. In the US, copyright is one of two elements of the Constitution that contain a justification. The first is the Second Amendment ("a well-regulated militia").
The second is copyright ("To promote the progress of the useful arts and sciences"). What's more, the "progress clause" is a "may," not a "shall," meaning Congress need not provide *any* copyright
@archo If Copyright is intended to ensure "fair reward," it has failed on its own terms, repeatedly. As I wrote in the thread:
@archo I co-wrote an entire bestselling book of policies that will increase the share of income that goes to creators. None of them involves more alienable exclusive rights:
@pluralistic I don't want to argue in favor of concentration of wealth, power or copyright. :)
Just saying that was the intent of the law. There's definitely plenty of room for improvement and possibly alternatives.
@pluralistic What's untrue?
https://constitution.congress.gov/browse/essay/artI-S8-C8-2-2/ALDE_00013062/
it says:
> "encourage, by *proper premiums* and provisions, the advancement of useful knowledge and discoveries";
@archo "to promote the useful arts and sciences" - not to "ensure fair reward."
@archo If the framers had intended this as a compensation scheme, it would have read "shall," not "may" (the majority of the constitution's prescriptions are mandatory, this is one of the few that is optional - that's not an oversight. The framers did not extend *any* copyright to foreign works, for example, because they judged that padding the margins of US printers with the works of foreign authors would promote the useful arts)
@pluralistic Obviously they phrased it differently but it's also clear from the associated text that the "promotion" they had in mind was an incentive using some forms of rewards.
The "proper" to me reads as "fair" but I'm sure some may disagree. Fairness isn't entirely objective anyway.
@pluralistic I don't think anyone's arguing in court that a genAI model training script is infringement by itself?
The issue is with the data set the companies collected and fed into the script.
If they had the rights to every single image they used, there would be no lawsuit.
Sorry if I was initially unclear, that was what I was referring to, not the general act of "training".
No, you're still wrong.
The act of making a transient copy of a work in order to examine it and publish facts derived from it does not infringe copyright.
You don't need the rights to a work to do these things.
@pluralistic This is the same kind of reductive phrasing I already addressed in my initial post.
Not to mention that the copy on their hard drives isn't even especially "transient" considering that they want to reuse the data for subsequent "training" runs.