From: "Thatcher, Sanford Gray" <[log in to unmask]>
Date: Mon, 29 Jan 2024 08:33:38 +0000
I beg to differ with this claim that "transformative use" supports seeing
training AI on material from the New York Times (and the novels of the
seven trade book authors who have also sued OpenAI) as fair use.
The basic argument appears to rest on an unquestioning belief in the social
utility of AI technology. Let me remind people that, in deciding the suit
against Texaco in 1995, Judge Jon Newman admitted the social utility of
photocopying but nevertheless found the use of this mechanical copying to
be copyright infringement. One van also question just how sociallly useful
AI is when it learns how to write scientific papers and then attribute
their authorship to scientists who did not indeed write them, or the "deep
fakes" that AI has been generating to bedevil people ranging from President
Biden to Taylor Swift. Technology is not inherently good. Social media
platforms once seemed very promising but have been creating havoc and
divisiveness in our political world of late. And those of who saw the
movie about Oppenheimer will remember how mixed his views were about the
technology that led to the creation of the atomic and hydrogen bombs. And
think about how "intelligent" AI is really going to be when it trains on
only 3% of the world's stored knowledge, the vast majority of which lies
locked behind firewalls controlled by journal publishers. So, you'll excuse
me if I don't bow down and worship AI technology uncritically.
Transformative use is a concept invented by Judge Pierre Leval in a now
classic Harvard Law Review article published in 1990. It was first applied
in a major case by the US Supreme Court in the 1994 case involving rap
group 2 Live Crew's parody of the song "Pretty Woman." The court ruled
that if the copying done were itself an act of creativity, it could be seen
as fair use because of its transformative nature and Leval's article was
quoted to that effect. But then a series of cases in the Silicon Valley
friendly Ninth CVircuit applied—misapplied, in my view—this concept to
circumstances where socially useful indexes and other products were used
algorithmically, allowing for subsequent creators to make use of them,
though the acts of making these products were not in themselves creative
anymore than the photocopying machines in the Texaco case were. This
mistake was then carried over in the decisions about HathiTrust and Google.
In the latter case Judge Chin explicitly argued that it was the later
creative uses facilitated by the mass copying that justified calling it
fair use, in direct contradiction to the precedent set by the Texas
decision in the same circuit. So much for precedent!
The ARL seems so enamored of technology that in its Code of Best Practices
in Fair Use for Academic and Research Libraries (2012) it put forward the
alarming argument that, because scientific journal articles and novels were
not originally aimed at students in the college classroom for which use
they have been "repurposed," it therefore should be fair use for libraries
to make as many copies as needed to supply these materials to
undergraduates in coursepacks with no permission needed from the publishers
or authors and no compensation paid to them. This to my mind is the
reductio ad absurdum of the whole line of argument libraries have been
using to justify massive amounts of copying with no regard to the fourth
factor's emphasis on the effects of such copying on the market. It was the
preeminence of the fourth factor that we in the Association of American
University Presses (as it was then called) emphasized in my testimony
before a Congressional committee hearing about fair use in 1973 that, had
it been adopted, would have led to a much clearer and cleaner jurisprudence
of fair use than the tangled mess we have on our hands today.
I go into greater detail on all this in an article I wrote for the
50th anniversary
issue of the Journal of Scholarly Publishing, which is available "open
access" at this site:
https://scholarsphere.psu.edu/resources/460d0813-f82b-400a-abc8-137bf9d1f647
Sandy Thatcher
Reflections on Copyright Law and Scholarly Publishing over Fifty Years
<https://scholarsphere.psu.edu/resources/460d0813-f82b-400a-abc8-137bf9d1f647>
For the 50th anniversary issue of the Journal of Scholarly Publishing this
article reviews the past half-century of copyright law as it bears on
scholarly publishing, especially that done by university presses, It pays
special attention to how the jurisprudence concerning "fair use" developed
over this period during which time the author was serving on the copyright
committees of the Association of American University Presses and the
Association of American Publishers and also on the board of directors of
the Copyright Clearance Center.
scholarsphere.psu.edu
------------------------------
From: Ann Shumelda Okerson <[log in to unmask] <[log in to unmask]>>
Date: Fri, 26 Jan 2024 21:35:09 -0500
Training Generative AI Models on Copyrighted Works Is Fair Use
by Katherine Klosek, Director of Information Policy and Federal Relations,
Association of Research Libraries (ARL), and Marjory S. Blumenthal, Senior
Policy Fellow, American Library Association (ALA) Office of Public Policy
and Advocacy | January 23, 2024
. . . But as champions of fair use, free speech, and freedom of
information, libraries have a stake in maintaining the balance of copyright
law so that it is not used to block or restrict access to information. We
[LCA] drafted the principles on AI and copyright in response to efforts to
amend copyright law to require licensing schemes for generative AI that
could stunt the development of this technology, and undermine its utility
to researchers, students, creators, and the public. The LCA principles hold
that copyright law as applied and interpreted by the Copyright Office and
the courts is flexible and robust enough to address issues of copyright and
AI without amendment. The LCA principles also make the careful and critical
distinction between input to train an LLM, and output—which could
potentially be infringing if it is substantially similar to an original
expressive work.
More here:
https://www.arl.org/blog/training-generative-ai-models-on-copyrighted-works-is-fair-use/
|
