Re: "Creators call for action on AI copyright exceptions" - log of use and geographical scope

On 3 Aug 2023 at 12:15, "Laurent Le Meur" <Laurent Le Meur 
<laurent@edrlab.org>> wrote:

> > AI developers and providers of generative AI services could easily provide 
> > a link in each derivative work generated by such a service to an online 
> > searchable index of the works in the corpus from which it was generated, 
> > and require in their terms of service that users include such a link.
> 
> Are you really sure this is true, technically speaking? Sure that a
> system that ingests millions of resources can list the exact set of
> resources used to generate a specific new resource? 

I did not suggest that it would be possible to indicate some specific 
*subset* of the corpus had been used to generate each specific derivative 
work. Rather, I suggested that, given that these works are typically 
derived from the corpus as a whole, appropriate attribution would consist 
of a link to an index of the corpus as a whole. I believe that it would be 
possible to keep a log and create an index of the works ingested, to make 
a searchable copy of that index available online, and to include a link to 
that index in the credits to the sources of each derivative work.

Attribution of creatorship is a moral rights, and pursuant to the Berne 
Convention is is a human right recognized and protected by treaty. It is 
the obligation of those who want to use creative work to respect this 
right, and not to use such works in ways that don't respect this right.

If you can't find a way to do what you want to do legally, don't do it.

If you have other suggestions for how providers of generative AI services 
could provide attribution of authorship, feel free to make them. But 
unfortunately, to date, generative AI services seem not to have provided 
attribution to the authors of the ingested corpus in *any* manner.

For example, Leonard Rosenthol points to this page as indicative of 
"responsible" implementation and use of generative AI:

https://blog.adobe.com/en/publish/2023/03/21/responsible-innovation-age-of-generative-ai

This page includes an image captioned, "Image generated using Adobe 
Firefly." But I can find no attribution anywhere on this page to the 
authors of the works in the corpus from which this image was derived, or 
how I can tell whether or which of my works may have been so used.

Adobe says elsewhere that Adobe Firefy has been trained on, inter alia, 
"openly licensed content". But almost all "open" licenses, including for 
example all of those Creative Commons licenses that allow use to create 
derivative works, require attribution in all such derivative works.

This is a TDMrep working group, not a working group on attribution. I 
raised the issue of attribution only in response to a request that I 
specify which works had been copied and used in the manner complained of 
in the statements from US and EU creators about the TDMrep protocol which 
I brought to the attention of the group. Lack of attribution makes it 
impossible for creators to know with certainty which of their works have 
been copied and used used in which generative AI systems. (Those 
statements were concerned specifically with *generative* AI, and not with 
other types of AI or uses other than generation of derivative works.)

As for the geographical scope of EU and national law: For better or worse, 
both EU and US law, and practices carried out in the USA and EU, have 
global implications. This is why US, EU, and worldwide organizations of 
creators have all expressed the concerns about the EU policies and their 
implications for creators worldwide in the statements I shared.

Sincerely,

Edward Hasbrouck





----------------
Edward Hasbrouck
<ehasbrouck@nwu.org>
+1-415-824-0214 (San Francisco)

National Writers Union
https://nwu.org
+1-212-254-0279 (New York)

Received on Thursday, 3 August 2023 16:03:16 UTC