IPSC Opening Plenary Session

IP Outside the (Doctrinal) Box

Andrew Gilden: Love & Hate in IP

IP’s struggles with joy, pleasure, and fun—happens with sex, but also with enjoyment. Courts are pretty openly hostile to statements by Ds that they enjoyed engaging in the disputed conduct. Fandom—Polychron (LoTR), Axanar (Star Trek), Harry Potter Lexicon. Fandom is considered a reason to rule against the D. “Perhaps because Vander Ark is such a HP enthusiast, the Lexicon often lacks restraint in using Rowling’s original expression.” Also, efforts to entertain fans are seen as a reason to impose liability: fans’ pleasure as what the IP owner owns. Making people think differently about a work: fair. Fans are framed as hungry, passionate, horny. Anti-fandom or alienation from the work is more likely to be found fair. Lombardo v. Dr. Seuss, Adjmi v. DLT Entertainment (Three’s Company, Adjmi talks about finding pop culture baffling and alienating); Lyons v. Giannoulas (D really hated Barney the dinosaur). Appropriation art: see a distinction between celebration and detachment. Cariou v. Prince finds fairness in many images when Prince didn’t care about other artists; Graham v. Prince rejects fair use when his stated intent was to have fun. Critical distance from work matters.

Sexual pleasure: when mark is used to “titillate” or convey a message that sex is good, that doesn’t establish parody or commentary. Copyrightability of sex toys: court says that providing sexual pleasure is a purely utilitarian function, not creative. Stimulating the mind is not separable from stimulating the body. But insulting someone, including by degrading them sexually, is more likely to be found fair/parodic. Disdain as paradigmatic fair use. Mattel v. Pitt: Dungeon Doll is fair use in the context of sexual slavery and torture. Catcalling in Campbell.

IP cases privilege violent sexuality over affirming, light-hearted sexuality.

By contrast: having fun is not privileged. Buckyballs case; Jack Daniels v. VIP—amusing dog owners is not a relevant interest. But outright mocking does better. Thus, butt-related parodies are less likely to be confusing—Buttwiper, Lardashe. Insult as litigation strategy: Jim Henson made sure to mock Spam in the Muppets case.

Harms of the pleasure taboo: (1) rejects lived experience of creativity: creativity feels good; (2) if you admit to having fun, it weakens your case, so that discourages you from telling the truth/encourages you to tap into your childhood trauma; intellectualize motives; (3) turns fun into something owned by the IP owner. Most fandom involves a combination of love and hate; but IP encourages downplaying the celebration in mix of celebration and critique. Punishing things that are incentives for creativity is bad, actually. Market failure explanations exist, but the flip side is that joy and pleasure end up seen as commercial property.

Recommendations: think critically about love and hate; hate and pain can be silencing; calling a woman ugly is not the proper paradigmatic example of fair use. Happy-talk rules are bad and viewpoint discriminatory according to the Court in Elster; but so are awful-talk rules.

Peter Lee, Synthetic Data and the Future of AI

ML models and training data

Real world training data can be difficult and expensive to gather; needs to be cleaned and label; high cost of acquisition contributes to concentration in ML field. Threats to privacy: personal info used as training data. Real world data may have biases—African Americans have lower historical health care expenses b/c they have lower access, but ML may interpret that as lower needs. Copyright infringement claims: uncertainty creates enormous risk for ML developers.

Synthetic data is artificially created; has a recursive character. AI models create synthetic data to train other models. Estimate: by end of this year, 60% of training data will be synthetic. Varies in how synthetic it is—could rotate an image 15%; augmentation/modify an existing dataset. Model-based synthetic data (deepfakes used to train); synthetic environments like simulated driving space. Training on fictitious records can minimize privacy concerns, fill in gaps of RW datasets for underrepresented populations; avoid © infringement, in theory, though may simply shift issues upstream. But it can be low quality; leak information, too similar to reality, amplify real-world bias. It can also be too different from the real world; “model collapse.”

Innovation mechanisms to ensure robust, responsible development of synthetic data: Law & policy are inputs into the creation of synthetic data. Three objectives: provisioning high quality data; disclosure of data and methods to promote validation; democratization of access and multiple sources of synthetic data generation.

Jacob Noti-Victor, Regulating Hidden AI Authorship

Marvel got in trouble/claimed that AI was only used to supplement humans; it’s getting hard to tell in many cases what’s AI generated. Creates incentive to hide use of AI. Also problems with copyrightability create incentive to hide AI origin.

Are consumers harmed by failure to disclose? Consumers have preferences for process—like ethically sourced coffee. Authenticity concerns; concerns for fandom/connection to the artist. Artists have recourse for appropriation of voice or likeness, but maybe artists license their voice or likeness and so they are ok and the consumers are the ones who lose out. Of course, AI is supplanting a lot of authorial agency—that creates a contrast between assistive uses (editing) versus generating main plot points plus a teeny bit of human tweaking. Copyright Office guidance is actually similar to what he thinks is a good distinction. Prompting an AI system isn’t enough; something more is needed. [Editing doesn’t make someone an author in our system of joint authorship!]

How do we incentivize disclosure? Industry might self-regulate. Watermarking; certification. Mandatory disclosure legislation; FTC enforcement; IP could be used as information forcing if we used copyright misuse to raise the financial stakes of nondisclosure. Fraud on the CO is not serious, but misuse could be.

Jennifer Rothman, The IP Exception to the First Amendment

The SCt is intentionally not giving more robust 1A review and giving IP claims a pass or differential/bespoke/lesser scrutiny—lower-value speech when you use someone else’s IP. The idea of this project is to take that seriously: what if they’re right, by their own lights? If we take it seriously, what are the opportunities to increase predictable outcomes or develop more speech-protective outcomes?

In 2003, Eugene Volokh said it was “unfortunate” the Court hadn’t heard more 1A/IP cases, but 20 years later, they have. (The monkey’s paw!) He hoped for more clarity and also more rules favoring speech like putting the burden on showing lack of fair use on the plaintiff. These decisions are not, as Lemley/Tushnet say, at odds with the Court’s other 1A jurisprudence, but consistent with a history of treating IP differently. (I think those are … the same thing, described differently?) A descriptive, not a normative project: the IP exception to the First Amendment. She thinks Elster shows she’s right. [OK, since I got called out by name, note that Elster represents—and Barrett’s concurrence shows that she understands it to represent—a very different approach to the 1A than other cases, including other IP cases, although there’s a surface resemblance to Eldred. Elster and Rahimi are clearly about the same questions about history and tradition, even if they’re about different amendments. (Even JDI doesn’t give us 1A theory/a theory of 1A coverage and protection; it just announces a conclusion.) The history and tradition view of the 1A would—as Thomas very much wants—rewrite the rest of 1A doctrine. And that might happen! But this is not a continuity story; rather, it’s part of the Trump Justice revolution.]

Idealist vision of IP and 1A: (1) IP laws are content-based speech restrictions; (2) no exception applies; (3) strict scrutiny applies. (1) was controversial 20 years ago, but now isn’t. (2) is where the fight is. (2) seemed to die with Eldred because of the definitional balancing in ©. But now that seems to be a broader story: you can’t use other people’s property to make your speech. It’s not just a © exception, but an IP or property exception to free speech. We see this in claims not justified through the Progress Clause. INS v. AP; Zacchini; SFAA (Gay Olympics); Elster. You can also limit speech to get more and better speech. INS; Zacchini. [But see NetChoice v. Moody, this Term, the same people say: “But this Court has many times held, in many contexts, that it is no job for government to decide what counts as the right balance of private expression—to ‘un-bias’ what it thinks biased, rather than to leave such judgments to speakers and their audiences.”] Her list is a pantheon of hated IP cases, sure, but they are there. The Court loves them and cites them a lot! [I’m not sure this is numerically true, but it’s certainly a fair characterization of Elster. That’s why the speech-promotion rationale for an IP exception contradicts the rest of 1A law.]

Court sees this as competition policy, not speech, which is why there’s a lot of focus on fairness. Market annihilation cases are key: Harper & Row; Zacchini; Warhol—worried about destroying the market for photographers and their work. INS also fits.

Limiting speech to get more speech is a broader notion than just IP. [I wrote something about this.] Not anomalous in IP: heckler’s veto cases limit speech to get more speech; TPM cases are like that. [This, by the way, is why content-based distinctions have been very important in doctrine: TPM restrictions and heckler’s veto cases are not content-based.] Compelled speech concerns loom large—the Court doesn’t like that and disfavors compelled speech. [This is one thing Lemley & I talk about when we talk about the sudden shift in the treatment of corporate speech, most on display in the 5^th Circuit but also showing up in the disclosure parts of the NetChoice cases.]

She would bet a lot of money that dilution is constitutional absent viewpoint discrimination. Commercial use is not a likely dividing line given the property framing. People aren’t facts: they’re protectable property. [This is a version of Thatcher’s line that there is no such thing as society, as if you could carve off a person from the environment in which she exists. The Court majority is not coconut-pilled.] Therefore, countervailing stories about competition and property would be good—the public domain as property. [This was the 10^th Circuit approach in Golan that the Court didn’t deign to mention.] Implications beyond IP: access to property, as in NetChoice and physical spaces; access to data. Propertizing privacy might succeed (although alienability is a big issue). Limiting speech to improve speech can be harnessed to limit deceptive political ads and other deceptive uses of AI

If we don’t like it, we should focus on shifting the property-speech balance more broadly. Overpreferring property—Justice Marshall’s dissent in Lloyd v. Tanner.

SCt has long preferred property to speech. [This is why I have trouble thinking that there’s much good room to work in this framework. The “property is different” justification is different from the “suppress speech to get more and better speech” justification,” and doesn’t authorize the same things. Compare: give the US a property right in the US flag and ban flagburning; give people a property right in their reputations and ban true but damaging statements. Only if “suppress speech to get more and better speech” is why IP is different do you get potential links to misinformation, campaign finance, etc.—and linking tangible property to IP works against, not for, that without a lot of additional work and a progressive view of the affordances of tangible property that, it seems fair to say, will not get much traction with this Court.]

Lemley: if it can be characterized in some way as IP, it is therefore property—that’s a crucial step in this. People are not property! Maybe one way to rethink this is not to rethink land, but about how to challenge the link between IP and land.

Rothman: agrees that IP is different from tangible property. Not suggesting we should treat IP more like property in land, but SCt has, and has for a long time, put IP in the “property” basket. We can argue that they shouldn’t do that, but if that’s how it’s being conceptualized, then is there a way to navigate that to get some results that are more speech-protective? [Note that much of this doesn’t seem internal to IP; it’s more “use the rule that it’s ok to suppress some speech to get more speech outside of IP to get other non-IP regulations we like” which has different implications.] [This treatment—which I think is real descriptively—is also frustrating for a 1L property teacher who spends an entire semester on the concept that the label “property” is the beginning of the inquiry rather than the end—with what boundaries? With what ability to interfere with others’ freedom of decision/movement?]

For Gilden: people hate criticism and love praise; no one will license criticism, so we need a free speech regime for that. But you can turn that around by showing all the situations in which people will not license praise. And refusals to license could even be turned around to show that we should treat things that don’t seem critical should also be favorable.

Gilden: people are not often suing over lost market—they have privacy or dignity concerns. Parody/loss of market isn’t really what’s going on, on either side.

Q for Gilden: people in this room may really enjoy criticism!

A: yes, there is not a bright line; we should think critically about both sides of the equation. Being involved in a hateful political movement can feel really good; that doesn’t make it good. Love v. hate is not a good divider; what are we trying to do and what is the impact of expressions of love and hate on the relevant interest?

Betsy Rosenblatt for Lee: why wouldn’t synthetic data exacerbate these problems by being generated from existing garbage? [E.g., how do you generate missing data for groups you haven’t actually gathered data from?]

Lee: it’s definitely true that there is high and low quality synthetic data. This is one reason to use open source: communal efforts to understand how people use these processes and engender trust. Requires judgment about how the synthetic data should differ from real world data; otherwise can make the bias much worse.