AU Washington College of Law, Program on Information Justice & Intellectual Property
Chatham House rules.
Opening discussion: The Role of the Public Interest in
Framing the Copyright Act of 1976
AI is sucking all the air out of © discussion about other
topics. It doesn’t have to be the predominant question of © and the public
interest. 76 Act was a point where libraries and educational institutions
realized they had to advocate for the public interest, but it was basically
them negotiating against publishers; DMCA was enacted w/ a broader coalition of
public interest organizations. Now: parents, consumer groups pushing back
against platforms, and now plaintiffs’ bar finding chinks in armor w/product
liability suits. Question: is social media having an “Unsafe at any speed”
moment? Is this a resurgence of digital consumer rights, or something else? If
it is, can we imagine © as a part of this, e.g. in a digital consumer’s bill of
rights? Should YT have to have due process before a third strike nukes your
account? Doesn’t seem like Congress will change © law comprehensively any time
soon; we’ll be living with the 76 Act for a while.
Enactment of 76 Act: keep in mind the excitement of that
time, not b/c there was necessarily so much love for the new law or experience
of the new law as a loss, but a collective idea that one was about to be
presented with a new tool that was fundamentally different from what had come
before—part of the first generation of lawyers who first applied the law (and
then of course immediately began to complain about it). Present in the
classroom and in practice for decades. Ben Kaplan’s advice to students:
occasionally at least, one working in the field ought to represent clients who aren’t
copyright owners; classes of such people with copyright-relevant interests did
develop (called alleged infringers in the cases) but were not always visible at
the time.
Authorship and Copyrightable Subject Matter - From the Sweat
of the Brow Doctrine to Originality
Sweat of the Brow doctrine wouldn’t die—the directory cases
were where the money was! BigLaw of the time was litigating directory cases b/c
knowledge-based works were valuable. They all involved Lockean labor theory
type claims. Connecting to AI: as people start to make AI output claims, they
feel like twists on sweat of the brow—which was always a reward for capital
investment into someone else’s labor—we shouldn’t reward the labor of operating
capital-intensive machinery any more than we would do that.
We need a term to capture the AI output that would satisfy
the originality standard if created by a human—a work, but not a work of
authorship—to name that aspect of something that might also have human
authorship in it. The human contribution: not really a derivative work; how do
we conceive of the collective expression of the human/AI output? If mixed into
interdependent work that would otherwise count as joint authorship, seems like
we just have a single author—but have we really dealt with that issue (we don’t
even know whether there can be joint authorship with someone who is an
employee/making a WFH)? Will people find state law alternatives for protection
that aren’t preempted? There are already attempts to create state protection.
AI on the Lot conference in LA: working lawyers’
perspectives in Hollywood: lots of lip service to need to center human creativity
in AI generated work and the need to document it. See Jacob
Noti-Victor’s current project on deciding infringement cases after AI. Using
human actors & generated backgrounds: human-generated performances keep
that package protected. There was some doubt about fully AI-generated animation.
Clients want to see prompts before they’re used to see if
they are intentionally or unintentionally going to infringe. They also want
indemnity for the studio.
Registration: how does this scale? How do I have to document
& disclose my AI use? Is the Office going to be able to meaningfully review
it anyway? Is this an empty exercise? Critterz: AI-generated short film from April
2023; used text prompts to create visual characters & background design.
Was registered: Script was human, voices were human, editing was human, and
humans animated the characters once designed, which included human performance
capture. So registration isn’t surprising, but how did it do that?
General advice: infuse as much human involvement & control
if possible at every step of creation and document those touchpoints—scripts;
document how humans influenced, modified, selected/arranged/combined outputs.
Instead of just text prompting, use outputs other than text: have a sketch fed
into a multimodal engine; make text prompts as detailed as possible in case
those will confer protection. Keep the original outputs confidential, including
in your filing with the Office.
3 practical Qs: (1) The Office demands you disclaim © in AI
generated stuff, but they may be wrong that you don’t have a © in that. Waiting
on Theater De Opera Spatial case: fully AI-generated artwork based on a text
prompt; challenging denial of registration. (2) how much specificity does the
Office actually need in registration? Detailed audit trail (keep for litigation
purposes but might not need to expose). (3) how do you keep people from copying
unprotected bits? If you vague it up, people won’t be able to be sure what they
can freely copy.
No pushback from Office for Critterz by saying: (1) “to the
extent not copyrightable” andgeneral description of images, etc. generated by
AI, with same disclaimer; (2) no detailed audit trail, and Office didn’t even
publish their general 2-page description document with the registration, which
doesn’t say much/can scale; (3) if you keep disclosures vague and specific
outputs confidential, people can’t copy any part of the work w/o risking
copying the human element. Great for © owner but calls into question whether
Is this right for registration? Burrow-Giles said: they made
these choices so the whole thing is ©able.
Single Piece of American Cheese: the documentation of the creation
was extensive, which led to a finding that the selection & arrangement was ©able
even though the visuals were AI-generated. It may matter who you are:
represented by BigLaw, perhaps you can vague it up more readily. Or if you
signal that you are following Office policy.
What kind of market do you imagine? If you’re worried about wholesale
copying, all you need is a Burrow-Giles ©. It’s only when you care about
copying bits that this kind of copying really matters.
Sure, the “whole thing” is copyrighted, but not everything
in Burrow-Giles was ©able—Oscar Wilde’s appearance was not copyrightable. A
copy of the photo would be an infringement, but not another photo of Wilde.
If that’s what you believe, what have we accomplished with
the Office’s policies where registrants don’t really tell you enough to decide
what you can use?
Might become even harder to ascertain what’s protected than
it already was. (Relevant: Jeanne Fromer’s
work on scope and claiming by exemplar.) You can look at a photo and decide
what’s ©able more easily (though not always!).
Costs of nailing scope down are generally not worth bearing
at the registration stage, but what if AI changes that? Alternative discussed
by Noti-Victor: Markman hearing analogy, so claim scope costs are not borne at
registration stage but are still required early in infringement claim.
Multiple page-long prompts may be used for a single shot,
including what lens was “used.”
The issue is that the output may not be copyrightable at
all, even though there are always uncopyrightable elements in a work; we used
to do this reasoning only in the context of infringement, not at the front end.
Doing it at the front end is not stable/scalable b/c Feist says the tiniest bit
of creativity is enough to make a work original.
Issue isn’t doctrine but economic power. Human
creators/actors want creativity not b/c of © but b/c of labor power. Restraints
should come from the outside.
Single Piece of American Cheese: wanted to prove concept of
selection, coordination and arrangement resulting in © for people who didn’t
have (other) artistic skills. That sounds nice, but consider the power of the
owners of the tools. Adobe v. open source: nothing in our copyright doctrine
speaks to that.
Fair Use and the Constitutional Purpose of Copyright
Start w/ assignment given to students: is the quote “use the
force, Luke” substantially similar to Star Wars, the movie from which it comes?
Half the students, reading the standard cases, say it is, b/c it’s recognizable
and not de minimis. That is, when they just read the cases, they conclude: Fair
use has to handle everything if there’s a recognizable quote.
This is a bad idea, and contributes to the challenges we’re
facing in a world where a lot of the litigated fair use victories are big data,
not uses of individual works. Sometimes
defending big data means denigrating acts of choosing a specific work to react
to: 9th Circuit said in one of its Perfect 10 cases: “a search engine
may be more transformative than a parody because a search engine provides an
entirely new use for the original work, while a parody typically has the same
entertainment purpose as the original work.” Robust fair use requires multiple
types, not crapping on one type for the benefit of another.
Big data cases also highlight the distinction b/t straight
reproductive copying and alteration: also important to avoid tilt towards the
former on purpose-based grounds, which would mean that fair use was mainly for
quotations in a different context and then once in a while for parody. The kind
of analysis needed to produce ASTM v. Public Resource gave us a good result,
but also heavily favors pure reproduction. One thesis: Warhol’s language has negative
effects on copying that alters.
Relatedly: we face a risk of substituting fair use for
substantial similarity analysis, which did great damage in Warhol. Example:
concurrences in the Sedlik cases suggesting that b/c copying in fact was so
obvious, there was obviously substantial similarity, leaving only fair use as a
limit on use of photographic references. Asking about fair use first without
establishing substantial similarity enlarges the scope of what courts consider
protectable, to the detriment of future creators who might not do so well on
the first and fourth factors.
Also, the way courts use factor three in small-amount
cases—favoring fair use because the defendant took almost nothing
protectable—sits poorly with the rest of the structure of fair use analysis.
Factor three is generally a calibrating factor: it assesses
whether the defendant took more than was reasonable in light of the purpose of
the use. That’s why full copies can be fair use, when a transformative purpose
requires a full copy. Using fair use as a substitute for putting a lower bound
on substantial similarity, by contrast, is purpose-independent, suggesting that
taking only a little bit is a virtue in itself.
Third and relatedly, although there are justified
predictable subclasses of fair use cases, the only justification for a
small-amount subclass is to avoid the need to put a lower bound on substantial
similarity. That has value to judges understandably looking for shortcuts, or
for Justices looking to put together a majority (looking at you, Google v.
Oracle), but it can’t supply overmuch in the way of predictability, and it
has serious costs for the defendants who are forced to litigate the extra
issues required for a fair use analysis, including expensive economic
discovery.
Fourth, and potentially of greatest significance, Warhol may
put defendants who didn’t take much, if anything, that was protectable at a
significant disadvantage going forward. Because the Supreme Court suggested
that competing uses were not transformative even if they had different meanings
or impacts, and because defendants who copy a lot of unprotectable material
tend to be using it in ways that compete with the plaintiff (or at least with
the plaintiff’s licensees), such defendants may be particularly likely to have
the highly important first fair use factor weighed against them. For the same
reasons, the market harm factor will do the same—and courts often say that
those are more important factors than two and three, where what I’m calling small-copying
defendants do well. Rentmeester is a good example of a case where a
similar-in-unprotectable-ways photo would fare badly on factors one and
four—both photos of Michael Jordan depicted him as a star player of almost
unearthly talent, and both would be fine illustrations for articles about him.
This problem is particularly acute in the context of
litigation over training generative AI. Recently, the Copyright Office and one
court have offered a new theory of “copyright dilution”: works produced by
generative AI might cause market harm by flooding the market for certain
genres. But this only looks like cognizable harm by skipping over the question
of substantial similarity: unless the AI-generated works are themselves
infringing because they contain substantial amounts of protectable expression,
they should not (and I think constitutionally cannot) count as inflicting
cognizable harm. Quickly skipping past
substantial similarity reduces infringement to a question of but-for causation.
Even if you don’t like generative AI—and I don’t—that is the wrong standard.
Shows how fair use and scope of protection need to be considered separately to
do their modern jobs.
Is public interest really a fifth factor, rather than part
of the fourth factor? Where does it really fit in?
Another issue: We have to figure out how to deal w/use by
use analysis without chopping everything into too-small bits. Use by use
analysis kind of makes sense, but we’re in early days figuring it out.
We the public can & should fit into all four factors,
& Breyer actually did that in GvO.
Rhetoric of copying and morality that infuses every conversation:
by the time we get to fair use, we’re using words of morality, which can lead
to twisted or troublesome logic. We end up with analysis that starts with
copying in fact & moves directly to fair use, or even skips over copying in
fact. You can’t demand that the public do the inquiries in order as you can
w/students—how do we shape the conversation about the moral valence of copying
and fairness?
One concern is how much weight is being put on fair use—all of
AI, much of computer programming, etc. Part of that is the dysfunctionality of
Congress. And 76 Act doesn’t say what the infringement test is. Fair use is at
least a test! Worries of judges in GvO: protectability would be across the
board so it was more comfortable to use fair use instead which is case by case.
Wild because protectability in GvO was also entirely case by case—nobody else
was going to copy those 11,000 lines of code. Substantial similarity is also
supposed to be case by case!
Structure is just more comfortable to many lawyers/students for
analysis for fair use than seemingly unstructured analysis for substantial
similarity (certainly no structure for infringement present in statute). Questions
of law can be raised—that’s why we have more SCt decisions on fair use after
the 76 Act, b/c now they can say they’re interpreting a statute.
Sedlik is an opportunity to think about clarifying
infringement/substantial similarity. Commit to all the limits on ©! Not just fair
use. Opportunity for a Leval-type theory for theorizing substantial similarity
to reinterpret the test.
People are obsessed with copying in fact. We have a bench of
people who like to think they’re originalists, but as we’re talking about the
public interest being part of the fair use factors, that is a way to put the constitutional
concept of “progress” into the factors. (Unless the public interest is market
aggregation.)
Worst offender: Sotomayor, the most experienced © Justice,
not an originalist, but substantial similarity analysis in 2d Circuit in Warhol
is laughable and she just buys it.
Do you try substantial similarity and fair use together in
one trial? Give the jury a special verdict form? Try substantial similarity
first for PF infringement before getting to fair use, or let jury hear both
together? Leval says: from now on, fair use facts should be decided through
special verdict; jury answers those questions; judge makes ultimate determination.
AI makes some GvO/big data case features look regrettable in
hindsight: (1) ignoring commercial v. noncommercial use. Puts us in position
now where courts have to bless fair use by the most heavily capitalized
industries on the planet who could clearly pay for what they want; if they find
no fair use, that will prejudice the ability of noncommercial, public interest,
open source AI efforts they might want to promote. (2) Google Books turned on
copy protections at end of product, but open source AI means you aren’t the one
deploying it or people can take them off even if you do deploy copy protections.
Model building bears more weight in fair use than in the
real world—people are mad about outputs, but think the only thing they can
actually control is model building, so they focus on that. They don’t think
outputs are controllable. © has statutory damages and enforcement mechanisms so
it seems like the point of control.
Judges resort to fair use to avoid big swings that set
precedent they’re scared of: GvO and ASTM where they don’t want to say the
standards aren’t copyrightable. But fair use can also be used to wave away bad consequences
of protection decisions in Eldred and Golan. Sedlik: J Wardlaw was shaking with
rage at the concept that the jury found the works not substantially similar. Courts
will appeal to fair use if they broaden substantial similarity—allowing them to
answer doctrinal questions w/o attending to public interest which is shoved
into fair use.
First Sale – From Ownership to Licensing
People used to have book & music collections, and now
that’s not how we access pop culture, educational materials, and everything
else. Questions about the common law basis. Single recovery rule has public
interest consequences that have been underexplored. Origins: early 20th
c antitrust/competition cases, but SCt reiterated single recovery rule in
Kirtsaeng, Lexmark—you get to recoup your costs through the first sale. You don’t
get to make a sale over and over again. That’s the single recovery rule. Now
honored in the breach through licensing despite SCt unanimity.
Recommendation: don’t use exhaustion and first sale interchangeably:
whether a country decides to allow national exhaustion is different from first
sale—Kirtsaeng was the first case to confirm that we have exhaustion as well.
Many countries don’t have broad exhaustion even internally—many
countries have lending rights and we (mostly) don’t, as a clear choice. Heavy
users don’t pay more for their copies as a conscious decision. But that’s
destroyed by licensing, which requires libraries to pay substantially more than
everyone else, putting users at mercy of platforms. For most users, that’s not
horrible. For preservation, it’s horrible. At the margin it does allow platform
or © owner to control ongoing uses.
Provocation: First sale is dead; get over it.
It matters that this is a wealth transfer to specific
entities. Amazon is not principally passing on the revenues to authors to
support authorship and creativity. It’s yet another wealth transfer to the
platforms like Elsevier.
Libby/Overdrive is owned by private equity. So much profit
being sucked out of this—makes publishers look good!
Publishers do get a share of a lot of this. Not all
publishers are equal. Overdrive has over 95% of the public library, 774 million/year
before sale to KKR. Publishers have always really disliked secondary markets
like used book stores. Contract preemption: US is unique when it comes to
contract preemption limiting exhaustion, creating a situation in which “first
sale is dead” is a real possibility. But what is the technical solution for
digital copies? Internet Archive case: if you scan something it’s a new copy. Without
contract preemption there’d still be other contract doctrines that caused problems;
this is like printing money for publishers/platforms so they want to defend it even
with the huge problems for the public interest.
Restraints on alienation of chattels: those doctrines should
be resurrected b/c of their public benefits. Lots of problems also stem from
the contracts being really bad. 6-7 streaming platforms distribute 90% of video
content; 5 ebook distributors; etc. Consumer protection and antitrust are important
parts of the solution.
We need to be able to reuse our things: first sale is
essential to conservation.
Digital first sale is going to be tough to achieve, but
right to repair movement is a positive development.
©-specific policy interest relates to the public interest:
access to knowledge. That is a double-edged sword when it comes to exhaustion
& first sale. European reactions to Kirtsaeng/shift from ownership to
access model: concern over disappearance of content. E.g., Infinity Train.
There are advantages to high quantity uses paying more. What happened to prices
in Thailand after Kirtsaeng? (One possible cite: Zhang, Z. and Feng, J, “Price
of Identical Product with Gray Market Sales: An Analytical Model and Empirical
Analysis,” Information Systems Research, 28(2), pp. 397-412, June 2017.) Global
price discrimination might promote cheaper access to knowledge—exhaustion v.
first sale.
109: every cobbled-together exception to allow Blockbuster,
but not renting software. CleanFlicks at the bottom. All Bobbs-Merrill said was
that if the legislature wants to modify the common law it has to do so explicitly.
There’s no constitutional right; the legislature could erase access or © for
any of these things. Book publishers would love to see first sale go away, b/c
that’s the path for every AI company to build its corpus in training its model.
Details matter: do we want compulsory licenses at a set
rate? The publisher decides whether a library gets the book? Does money go to
the author? 109 could be the home of a new rule that says “you can’t scan a
book for AI”—it would fit there. Or it could be the home of a rule that allowed
that.
Consider anti-criticism licenses—you can use it as long as
you don’t criticize it. That’s also a power issue.
Specific Exceptions: Lessons Learned
Questions about preservation, classroom use, other subjects
of best practices—almost always come down to fair use because the specific
exceptions haven’t been powerful for public institutions. Trying to craft
specific exceptions will probably fail: too narrow b/c of negotiations with stakeholders;
technology will change substantially by the time it makes it into actual law.
Copyright law is working great for libraries, but they don’t
get to live in © space any more, they live in contract land. © policy choices
reflect the public interest and we should be able to live in that land! 110(1)
is amazing, 110(2) is not: when it aligns with your incentives and intuitions
as an educator, as 110(1) does, an exception can be amazing. When it doesn’t, as
with 110(2), it is not. We have a model for doing a good job: general, values-driven
as in 110(1); horsetrading means it doesn’t work in 110(2).
Everyone is terrified of asking Congress to fix 110 b/c we
know they won’t. What do you do? The only alternative is “fair use for
everything,” with its own disadvantages. Think very carefully about long term implications
of opportunities to enact change: university counsels can’t rely on 110(2).
108: not litigated; just used to apply savings clause to confirm that fair use
is different. Maybe libraries are using it a lot and doing a bunch of things
that don’t trigger litigation—that’s likely what’s happening.
Maybe fair use helps confidence that 108 means what
libraries want it to mean. Jon Band’s Gravitational
Pull paper—unintended effect could be to shade how we understand fair use.
As time went on and fair use proved expansive, librarians’
interpretations of 108 probably became broader/put a fair use gloss on 108.
Compared to 110(1), which is clear, short, and understandable, and has developed
its own fair use gloss—to allow distance education, doing what 110(2) doesn’t
allow. The question for us: how did 110(1) get so good, even compared to 108?
Patry wrote that 110(1) came from language from an agreement
b/t publishers & teachers, whereas 108 was trench warfare over 10 years.
110(1) was already existing practice of teachers; it was late-breaking insertion
into the statute but not controversial. Tech constraints: they imagined
teachers using mimeo machine in the principal’s office. But also: under the
1909 Act, public performance right was only for profit; when they decided to
lift that restriction, they also decided to grandfather in existing practices—education,
church, agricultural fairs, etc.
Don’t lose track of the damages part. You can think 108
means whatever you want and there’s essentially no litigation b/c you can’t get
statutory damages against a librarian acting in good faith. Institutions need a
substantive hook to point to say they’re good © citizens, and they need
liability protection. Maybe that’s the formula: give an exception plus a
statutory damages safe harbor. Libraries also are comprised of people who want
to do the right thing.
Limitations and exceptions may contribute to courts thinking
“if no exception then infringement” as we’re seeing in TM. Savings clause in 108
is really important. At oral argument in Hachette rights holders argued that 108
meant there was no need for fair use, but the cases have been very hostile.
But there are cases saying that “de minimis isn’t a listed
defense so there’s no de minimis exception”—if it’s written too narrowly to be
effective and has the compounding effect of broadening rights, that’s
worse. [Like the statutory DMCA 1201 exceptions!]
University counsel might not be © experts; they like being
able to look at a provision that says “notwithstanding the above, this isn’t an
infringement.”
Penalties: Remedies, Secondary Liability, and their
Consequences
Today’s Hikma borrows Grokster to really
require active steps/intent to induce—the worry of loosening inducement is
something the Court is signaling against. Gershwin no longer stands for
material contribution + knowledge—it stands for inducement. Not causation!
What about vicarious liability? We’ll see more efforts to
expand that to bypass Cox. But there’s also a case to be made for
implications for vicarious liability in Cox. Basic takeaway from this
and Hikma: they want to see purposeful, culpable conduct. Vicarious liability
has been expanded past employer/employee, where lack of sufficient supervision
was your problem/culpable conduct. So too w/independent contractors. But when
you’re talking about entities that are farther apart, it will/should be harder
to expand vicarious liability.
Also going to be pressure on volitional conduct: something
that used to be considered indirect becomes direct, then you won’t have to
worry about knowledge, intent, or control.
But if the courts take seriously that nothing in the statute
gives rise to that broad liability, then there will be a narrowing. Wish it
would have implications for 1202! Framing it as Twiqbal: when you just
make general assertions in the complaint, that’s not enough.
eBay worked to limit injunctions—fewer sought, fewer
granted. Libraries & educational institutions don’t need to worry about profit
disgorgement; elsewhere, courts really need to think about disgorgement as a
tool of equity, to be adjusted down where appropriate. Likewise, 103(a) should
have equitable limits; otherwise Goldsmith might own the (c) to the Warhol
images and the estate would have no (c) interest.
Good that SCt accepted cert in Cox on
willfulness/statutory damages—they are clearly interested in the issue.
We should also be citing, quoting, relying on the
Restatement of Copyright, which says that statutory damages should approximate
actual damages.
Are class actions appropriate in © cases? Could we make more
progress on that front?
Texaco was the key there: 80 journal publishers as a
plaintiff class.
Build on what broke Google Books class certfication—similarity
of interests or lack thereof.
Is there room for apology as a remedy or other equitable
remedies unrelated to the economic conditions ordinarily associated with ©?
Here’s an argument: In our textualist era: there isn’t any
contributory infringement provision in the © statute. Therefore Congress never
authorized statutory damages for contributory infringement and they don’t
apply.
Secondary liability in the context of TRIPS: Didn’t say
anything about secondary liability, so countries have maximal flexibility on
whether to adopt it.
If “to authorize” is the scope of secondary liability it
will shrink in scope, which would be fine!
Posts
No comments:
Post a Comment