IPSC Closing Plenary: Fair Use After Warhol

Christopher Buccafusco, Copyrighting Style

There are plenty of precedents that say style isn’t ©able. Appropriating the “core” of a work is not infringing in the absence of substantial similarity.there are also cases suggesting ©ability of style: Steinberg, saying style is part of expression. Blehm v. Jacobs, Bandana v. TJX, Reece v. Mark Ecko say that style is an ingredient of expression etc.

Unfortunately, this matters because of AI. Greg Rutkowski is a more popular prompt than Picasso and he doesn’t like this. Visual art for examples, but can be extended to music and text.

To some extent, style is ©able and to some extent not. Style sits between idea/content/subject and method/process, things that are excluded from ©ability. © attaches to the how, not the what: the means by which the ideas are given expression. The manner by which the author makes certain choices. These are forms—but that’s one way of defining style, the manner in which we clothe our forms of thought. Synonymy: the idea that you can say the same thing in many different ways. That’s a questionable commitment in aesthetic philosophy these days.

Same problems if we think of style v. method/process. Style feels a lot like method/process—ways in which you learn to engage in 2-point perspective drawing seems like a method for doing things. But not everything with a bunch of steps in it is uncopyrightable. Choreography can be ©able.

We can also make incentive claims about broader protection for style. But © doesn’t (shouldn’t) care about fame or distinctiveness. Copyright attaches to works, not ouvres. The scope of © shouldn’t grow b/c an author has made six works when the six works made by six different authors would have a smaller scope of ©. Style inevitably involves authorial choices about manner and form of expressing ideas; impossible to say with any coherence when these choices become sufficiently abstract to be ideas. Telling the jury “good luck” admits that we don’t know how to do this.

Path forward: copyright consists in the original properties by which a work expresses: by which it represents, denotes, and exemplifies. These include both formal features and the subject or content of the work. A work is an expressive unit of both the formal features and the underlying content; the relationship between those is what constitutes the work. Only when all the relevant features are substantially similar should we find infringement: combining formal features and underlying content should be required. Subject of work acts as a limitation on scope of ©. Monet gets a © in his depiction of waterlilies but not as against depiction of violets. 2 photos of Michael Jordan in grande jete pose but with different formal features.

This will get hard: Steinberg-style drawings of other cities are hard for him.

This is basically what Krofft said in the parts that are good and not terrible: you need substantial similarity in expression of ideas, not merely of ideas. Statue of a horse can’t embody the same idea as statue of a dancer even if formal features are the same, nor would a painting of a dancer.

Cubism makes it hard: could Braque have sued Picasso? Hard to figure out what the subjects are.

Is this relevant to fair use? Suggests that satire should have a greater claim to noninfringement than it historically has. Satire involves using the same style to clothe different ideas; therefore it shouldn’t infringe (lack of substantial similarity as in the Greatest American Hero case; German case law; perhaps the jury’s reasoning in the Kat von D case).

Courts should avoid analyzing “feel” and more focus on formal features. More about glint in eye than “uncomfortable” versus “larger than life” in Prince.

[seems like this is going to have trouble with derivative works]

Amanda Levendowski, Fairer Public Benefit

Bias and harms of works aren’t taken into account in fair use analysis: recruits a legal tool typically aimed at one set of problems for the purpose of cleverly addressing a different set of problems. Here the tool is the “public benefit” subfactor of fair use, aimed at balancing public benefits with market harms. Novel tech can promote sociotech harms like bias, mis/disinformation and social/environmental destruction. Particularly pronounced with algorithms and AI. Defining public benefit could mitigate some of those harms. [I am v skeptical of this given the inability of courts to do full cost benefit analysis; 1201 exemption analysis shows that putting downstream effects into the hopper is generally a way to suppress fair uses, and competition, in the service of fears of lack of control by big companies like John Deere.]

Cases that have discussed public benefit in fair use: 19 of 38 involve novel technologies—VCRs, digital libraries, early algorithms.

Defining public benefit is difficult. But deeply troubled by algorithmic fair use cases that allowed lots of exploitation—iParadigms (plagiarism detection) and Perfect 10 (exposing women to public view when they contracted for more restricted nudity). Amplified misinfo: these women didn’t consent to having their nudity exploited publicly. This is inconsistent w/the promotion of progress, and misinformation doesn’t do that. [How do you know that viewers received that message or perceived any difference between “posing for Perfect 10” and “having images online”?] Not saying Perfect 10 was wrongly decided, but not an uncritical public benefit. Could have named the costs, which influences how lawyers and courts think and behave.

Should consider public benefits of use discounted by public harms. Benefits and harms are both difficult to quantify.

Courts, as well lawyers counseling clients, can do this. Courts can elevate scholarship into doctrine. [Do we think that today’s courts will embrace the broader social benefits of a trans Joker, or instead consider the film to inflict broader social harm?] This wouldn’t be dispositive and anyway the harms will already have been caused in litigation.

What about Congress? Codifying this definition is risky.

Fairer public benefit can be operationalized by client-centered lawyers seeking to help clients develop better technologies—exploring possible consequences of their actions. Consequences are often sociotechnical, not legal. Lawyers should be counseling on them. Formalize the tech; assess public benefits; identify public harms; reconsider public benefits; report to the client.

[why do you need to bring © into this? If ethical lawyering assesses sociotech impacts, then © is just a distraction.]

Michael Mattioli, Facilitative Fair Use

Implications of Warhol on training AI: surprising ways that AI is actually facilitating some good things in © ecosystem. Little AI services may be much more beneficial than Big AI. Amanda Wood-Shapiro, writer with dyslexia, describing use of AI as game-changer for author. Ability/desire to tell stories; difficulty organizing thoughts. More than just spellcheck/grammar checker; AI was a mentor for her (Grammarly). Not the only neurodivergent person using LLMs to get ideas out as authors. Publishers drowning under slush piles may be able to use LLMs to identify the diamond in the rough. Sudowrite and other editors—help speed of production, looking at narrative consistency and pacing. Writing ad copy, which authors need to sell their works. Not just books; ScriptBook for scriptwriting can evaluate potential in script—character likeability, whether there’s a big enough audience, etc.

Word of pessimism: if it’s not run properly, especially if trained on limited/biased dataset, could narrow the range of voices. But these tools are already there. AI models for casting data to determine who to cast—but trained on existing ads and films. Generating storyboards: directors can experiment w/different camera angles and scene composition.

AI is used to produce/disseminate games—porting it from platform to platform.

W/o fair use, these tools are far more limited. Licensing=only really deep pockets can play, and even they will have access to a far more limited set of training data, steering © system towards more bias and fewer voices. [Although the companies didn’t love how much omegaverse showed up in their first few iterations.]

Pamela Samuelson, Justifications for Fair Uses

Warhol: main opinion uses “justified,” “unjustified,” and “justification” 28 times in discussing factor 1; identified “targeting” as a justification, with 10 references to “target” (5 in dissent). Warhol Foundation, majority found, didn’t offer a persuasive justification for offering a license to Conde Nast to use Orange Prince on a magazine cover. Need justification for each use, not just initial creation. Will courts interpret Warhol as limiting justifications to targeting?

Whyte Monkee v. Netflix: court did exactly that—using parts of a funeral video in documentary was not transformative b/c didn’t comment on the authorial choices and intended meaning of the video. Fortunately, 10^th Cir. reheard the case; waiting to see whether they don’t mess it up quite as badly.

Other questions: will/should justification become a new subfactor in factor 1? She hopes not.

What fair use justifications will/should courts find sufficient besides targeting the first work for criticism or comment? (lots of them hopefully)

Are justifications only relevant to factor 1? (no)

Are justifications only needed if the second work has the same purpose as the first work? (maybe)

What kind of evidence is relevant to proving justifications? (??)

Leval’s conception of justifications: also uses “justif” words 33 times, almost all in relation to factor 1. For him, justification turned on transformativeness, adding something new with a further purpose or different character, altering the first with new expression, meaning or message. Historians & journalists need to quote others’ works to support their arguments/theories—a key concern for Leval. Thus transformativeness enables © to fulfill its larger public purposes. A justification may exist under factor 1, and it must be weighed v. other factors, e.g., did D take too much. Transformativeness is for him the justification—you don’t need a justification on top of a justification.

Campbell took a holistic conception of fair use justifications and emphasized the importance of fair use to provide breathing space; courts must consider all factors together and weigh them in light of the ends of copyright law. Transformative purpose may have spillover effects in relation to the amount and market effects factors—the more transformative the second work, the more spillover effects are likely. Every case to be decided on its own facts.

Campbell did involve targeting for criticism, but this isn’t the only kind of targeting in case law. Targeting the author, not the work: New Era—what a cruel, bigoted person L Ron Hubbard was. Targeting the subject of the work: Kienitz, saying the mayor was a jerk. Targeting a third person: Hill v. Public Advocate—photo of gay couple kissing to target legislator who supported gay marriage. Targeting a theory—Time v. Bernard Geis Assoc.

Justifications beyond factor 1: GvO involves factor 2 justifications; programmer investments in learning Java declarations that were inextricably bound up with method calls & implementing code

Factor 3 justifications: Faulkner Literary Rights v. Sony: “miniscule qualitative” part was fair use.

Factor 4 justifications: harm speculative/work not competing—Sony v. Universal; AG v. HathiTrust.

Other types of justifications? Public interest in access to info: Warhol gives this as example, making information available, citing Authors Guild v. Google; search engine cases (Kelly, Perfect 10, Authors Guild v. Google); online posting of gov’t edicts (ASTM); print disabled access to books (HathiTrust). Private interest in access to info: reverse engineering cases (Sega v. Accolade).

Freedom of expression: protests (Kienitz); anti-censorship (SUntrust)

Evidentiary: documentary film clips used to prove filmmakers point; evidence in court (Bond v. Blum); court reproducing images in opinion (Warhol)

Economic justifications can also come in different flavors; her project is to evaluate different types of justifications.

Post-Warhol caselaw has gotten more boring; courts don’t seem to understand that works could have more than 1 purpose. A “further” purpose is what Campbell talked about. Depends on level of generality of “purpose”—watch out for manipulation. DC Circuit upheld ASTM’s posting as transformative, Warhol notwithstanding. Cramer v. Netflix—tattoos shown briefly in Tiger King show fair b/c of transformatively different purpose. Thomson Reuters v. Ross: jury to decide if training data transformative; Warhol didn’t require TR win. Sedlik v. Kat von D appeal pending—Sedlik argues that he has to win regardless of jury verdict b/c of Warhol. Still early to judge impact, but justification not yet playing major role; most cases involve more than targeting original for criticism. Bringing that to court’s attention is our job.

Chris Sprigman, Copyright, Meet Antitrust: The Warhol Court and the Rise of Competition Analysis in Fair Use

Warhol has two innovations. First, the analysis has to be use by use, not work by work. Imminent in prior jurisprudence, but comes out roaring here. The use in Warhol is the licensing to Conde Nast for use on an issue commemorating Prince’s life. Second, a change in the definition of what a different purpose or character is for transformativeness purposes. Many uses with new meaning/message will be transformative, but needs to be sufficient to make the use different in purpose or character from a use the P might want to make of their work. “[T]he first factor relates to the problem of substitution—©’s bete noire.” That is, the thing copyright hates most is not copying itself, but copying that substitutes. Using a work to achieve a purpose the same as or similar to the original is more likely to supplant the original.

That makes competition key to transformativeness. It was intuitive to the Court that the images looked alike and could be used on magazines, and therefore competed. This is wrong. Milk and Coca-Cola can both be drunk; someone dying of thirst would grab at both; but they are not competitors in any ordinary sense.

Antitrust has a thicker definition. Start with a candidate market (e.g. Goldsmith’s photos). If a single firm owned all products in candidate product market, would it be profit-maximizing to impose a small, significant nontransitory price increase of 5-10%? If not, add the next closest product and try again. Does Peets compete with Starbucks? If they both increased prices, would people defect to Dunkin and defeat the price increase? If they would, you haven’t defined your market.

This methodology is both empirical and normative. Empirical: uses data from historical price changes/econometric analysis; customer/user interviews; objective information re product characteristics. Normative: why 5-10%? This is the competitive margin on which antitrust should focus. This isn’t always the lower threshold; sometimes 1% is enormously consequential for an industry where the margins are very slim. The normativity of the test is connected to its empirics.

Methodology actually applied to fair use cases: a D who creates an artwork for exhibition or sale will almost never compete. When you narrow it to particular uses, as Warhol did, substitutability becomes more tractable. What about widely distributed works? Empirics of antitrust will work for you—there will be historical price data that help you assess cross-elasticity of demand b/t works.

What about Warhol itself, with episodically licensed works on both sides? Hardest case. Prince is one of most photographed men on planet. That means that the market for works to illustrate magazine articles is crowded. Many works available for licensing. Unilateral effects theory: competitive “nodes.” If you have a crowded market, there can still be a harm if there’s a competitive node—if competition b/t the Goldsmith photo and the other photos was much less close than competition b/t Goldsmith photo and Warhol image. That’s what the Court should have asked. Asking about licensing to magazines generally is like saying that milk and Coke compete. Photo is a classical portrait; the Warhol work licensed to Vanity Fair was about iconization—a visual representation of “Purple Fame.” If they had licensing people, people would have observed that the Warhol was compatible with the article and the photo was less so. After Prince’s death, Orange Prince was an icon painting.

Moderator: Pamela Samuelson, Berkeley Law

Lemley for Levendowski: Negative as well as positive public benefits for the market factor too—why not? Level of generality at which you assess benefits v. harms: Perfect 10 was not a tech for depicting naked people, but rather for depicting images. Assess only uses made in particular case, or all uses and future uses that come from having/not having image search?

A: agree, include market benefits as well. Level of generality is a tricky question: privacy law can be imported to provide some frameworks about ©. For Perfect 10, Nissenbaum’s theory of contextual integrity explains why it’s invasive to take images behind a paywall and reproduce them outside the paywall. That aligns with courts’ unwillingness to endorse disclosure of unpublished works.

McKenna for Buccafusco: Formal characteristics and subject of work—there are levels of abstraction problems in identifying subjects—why is Monet’s subject lilies and not “flowers”? Steinberg: you said the formal characteristics were the same but not the subject—do you mean “thing depicted” or “meaning”?

A: definitely creates another hierarchy problem, but may be able to get a better handle on it by separating them. Substance hurdle + formal features hurdle.

Saurabh Vishnubhakat for Sprigman: isn’t your distinction b/t Warhol and photo snobbish and aesthetic?

A: don’t overread snobbishness—my analysis fits any crowded market. It’s about consumer preferences, which we take as they are. If they’re coffee snobs, we let them be; if magazine people think that the Warhol picture fits a theme, we don’t judge that.

Felix Wu: if the lookalikes generated by AI are substitutionary relative to original artists, should we worry about that?

Buccafusco: If we think about competition, Moscow on the Hudson doesn’t compete plausibly with Steinberg; you could increase the price of the latter a lot and Robin Williams’ presence would still make the former nonsubstitutionary. This is also probably true of a lot of the derivative work stuff.

Sprigman: Be disciplined about what part of the FU analysis we’re doing. Factor 1 involves competition. Steinberg is definitely not competitive. But it might not add new meaning or message and therefore might not be transformative. [I thought he was going to pivot to Factor 4 being a distinct analysis.] The court announces a result in Warhol telling us that competition is central for transformativeness, but doesn’t explore the dark corners. Let’s just say we apply substitution analysis to that situation—we would probably find the same purpose. Is purpose defined by competition? Or is purpose guided by competition? He thinks the latter is more useful. If they compete b/c of similarities in nonprotectable expression, that shouldn’t count.

Katrina Geddes for Levendowski: worry about courts’ ability to handle more capacious understanding of public benefit. Would that devolve into a battle of experts where whoever hires the best expert wins? Wouldn’t that harm smaller communities?

A: good question, and part of why I think best way to implement this is pre-litigation stage. Client may just say “thanks, it’s profitable to spread disinformation.” That’s the dangerous part of client centered lawyering. But less a battle of experts and more a battle of amici. [This is also about resources—they need to know that they need amici, and they often don’t until it’s too late.] We saw this in the pending Internet Archive case; they made a robust public benefit argument that the court didn’t like but did grapple with.

Q: Look at CFR 21 in medical device world—you’re always talking about reasonable assurance of safety versus benefit/effectiveness. That’s a way to balance interests. For AI: controls for randomness/going off the rails need to be taken into account. For Sprigman: is this a noun, verb, or adjective?

Q: Antitrust: plaintiffs hate market definition; it’s really hard. Courts hated having patent case inside an antitrust case in pay for delay and might hate this too.

A: Doesn’t have to be fullscale analysis; Court should have gotten an explanation that not all magazine licensing is the same/competitive. Maybe 5-10% isn’t right margin: should look at incentives to create in the first place.

Stefan Bechtold: Price is probably not the only normative goal in ©, so SSNIP might not be the right tool.

A: if we’re honest about ©, it’s trying to remove discipline from prices, so is that right? But what is our specific goal in factor 1 analysis? Court just told us a mission for that, and we can use these tools, albeit not unthinkingly. Who are the buyers? They may be very different in © markets versus commodity market. © has problems; so does antitrust. Every regulatory system has problems. But we learn something from asking the Qs that antitrust wants us to ask.

RT: 1201 exemptions suggest to me it’s a bad idea to ask any authority to do global-cost benefit analysis as part of a fair use inquiry. The Copyright Office thinks that it should only find circumvention noninfringing if it won’t risk people crashing Teslas, and it’s not good at that assessment.

A: Clients come in the door thinking that © is the problem; you need to identify problems earlier, sooner, and louder. It would be better as a general part of good transactional lawyering, but no one is there yet. If clients have © Qs, that gives them an opportunity to engage with public law problems. [I don’t get why that means it’s copyright’s problem. Clients often confuse TM, ©, ROP, employment law, and lots of other stuff. The lawyer’s job is to identify their actual problems.]

Rothman: Prince was terrified of being photographed; even with lots of photos of him, there may be fewer carefully posed photos than you think. Thinks Warhol is more about misappropriation.

A: Prince contained multitudes; he was photographed a lot even if he hated it. The Court may be motivated by something other than substitution, but what it said was substitution.