Friday, August 12, 2022

IPSC Breakout Session 4: IP, AI, & Data

Resolving Online Content Infringement Disputes with the Use of AI Technology

Faye Wang

European perspective: Corporations manage notice and takedown, but should a government authority be involved? Large platforms use voluntary automated filtering/content moderation to minimize legal risk. Previous EU legislation was tech-neutral and did not impose a general monitoring obligation. New © Directive does require prevention of future uploads, apparent contradiction of prohibition on general monitoring. Poland challenged; CJEU dismissed challenge; new guidance from 2021: Art. 17 should not be transposed in a way that leads to a general monitoring obligation, but the guidance does not explain how this is supposed to be done if they are also supposed to use automated systems and prevent future uploads. CJEU then held that refraining from putting into place appropriate automate monitoring/filter systems was a factor in whether they communicated to the public. But there is also a proposed AI Act that requires intervention of a human person, requiring assessment of AI before adopted so that they won’t cause harm.

Remaining questions: can AI assisted content recognition/moderation accurately monitor or determine infringement without human intervention? Her conclusion: no. Misrecognition of audio and video may occur. And tech has limited ability to understand context of fair use/fair dealing.

Possible solutions to reduce cost/human hours: AI assistance should be designed to provide better understanding of appropriate context. AI can do preliminary recognition. Then:  Automated computing investigatation process (not AI). Expert manual review for notice and takedown. If failed, online dispute resolution/court litigation.

Good for finding substantial part, but not appropriate for shorter excerpts. In marginal cases, human needs to be involved in reviewing, but they need to be properly trained to identify commentary.

Appropriation of Data-driven Persona

Zahra Takhshid

Should extend privacy to cover data about us. Background in the four torts: use the appropriation tort: one who appropriates to his own use or benefit the name or likeness of another is subject to liability to the other for invasion of his privacy. Expanded to other characteristics such as voice, surroundings/lookalikes (White v. Samsung).

Data privacy as the new frontier. May help address TransUnion case: finding a concrete injury rooted in the common law. What type of data? PII, treated as a standard and not a rule/contextually. Commerciality requirement? Quotes SD Fla: “the mere act of misappropriating the p’s identity may be sufficient evidence of commercial value to survive even a motion for summary judgment.” Consent: we’re transferring our rights all the time by contract. But scope can be challenged.

Q: CCPA—is there a need for this in California with its comprehensive regulation?

A: Yes, because we continue to rely on common law privacy torts.

RT: Thinking about Dobbs and the focus on private benefit in proposals like yours: is this supposed to be a right against the government? If so, the regulatory state is under threat in ways you might like for Dobbs but not so much for taxes. If not, whether categorically or because the government’s interests essentially always count as substantial enough to override the privacy right, that’s a reconfiguration that speaks to a profound change in the relationship of privacy to protecting the individual versus protecting the individual against non-government actors, which concerns me deeply. If you don’t change contract law, you may not have bought much with this reconfiguration either.

A: Should be against anybody including the government. [Then I reiterate my concerns about the regulatory state: how should we handle the gov’t’s interest? Balancing isn’t part of the tort, though it comes in for First Amendment-inflected defenses. Does the gov’t have to show justification? Why won’t it always win?]

Q: we consent all the time. How does that work in your model?

A: we need to help the common law grow and not die.

Jennifer Rothman: May want to spend time defining what you mean by privacy to deal with the gov’t q. If your project is just about the private law appropriation tort, that can be made clear. [Can it be asserted against the gov’t?] The real challenge is the First Amendment. Her own position is that ROP can protect privacy in the way you’re proposing, but the First Amendment has to be considered: how much we can tolerate this protection as against speech interest in public data. Need to engage that. Not convinced that you need a commerciality restriction; the common law tort doesn’t have that. [I think that the gov’t’s interest in collecting taxes is not a First Amendment interest; perhaps one can take the position that you aren’t entitled to not pay taxes so that doesn’t count as an “advantage” appropriated by the government, but that formulation strikes me as extra manipulable—and especially since data such as “what I bought from Amazon” are co-created with Amazon, I wonder if you’re entitled to that data as against Amazon under the same logic.]

Zahr Said: Have you developed a theory of abuse of the right? Copyright trolling, TM trolling, what would systematic abuse of this right look like? Sometimes that can help you build in limitations/appropriate scope of the rights.

The Right of Publicity: A New Framework for Regulating Facial Recognition

Jason Schultz

Sociotechnical change and the ROP: moving away from privacy framing, even though FR does have real problems of surveillance and discrimination. Think about when new tech comes along with ability to mass appropriate identities: law evolves, especially w/r/t visual image and identity, going up to videogames and films where people’s identities are reconstituted after death.

Visual identity is central to the existing case law, so there’s no innovation required in subject matter. Setting gov’t and university research aside, there is also clear commercial benefit, e.g., Clearview. Clearview also scraped and didn’t get consent; consent can also be very contextual, as the No Doubt/Activision case indicates.

What are the damages from being used in AI? Commercial value = damage; control; dignity.

Questions about copyright preemption: the distinction is balancing with innovation policy; you don’t have that in ROP cases because the ROP doesn’t care about innovation. Zacchini might be wrong, but if you take all of someone’s identity that’s too much for First Amendment purposes. Baseball stats cases=you aren’t taking everything about the person; taking the image goes too far.

Q: Universities have ethical rules about human subjects; would you change the framing under which they’d go about research?

A: could distinguish university research b/c it’s not products and services. Under Common Rule, publicly available info is ok, so ethical rules don’t restrain anything there. But the consent questions are real.

RT: (1) Universities benefit from research. If you stick with the common law, universities are 100% covered. You can say either universities shouldn’t be able to do this or that there should be a limit for this type of claim, but you need to say something. (2) It’s not persuasive to say that the ROP is better because it doesn’t engage in balancing with innovation policy. [I.e., that it ignores important social concerns.] Copyright didn’t always balance with tech innovation policy; it did so when it started to have big conflicts with it. If this is a common law right, you need to explain why it shouldn’t adapt to new conflicts by balancing.

Rothman: need to develop a bit further the “why.” You’re trying to avoid that by saying you’re just dealing w/commercial appropriation, but the traditional tort is broader; even if it was limited to commerciality you still need a normative justification for these claims, not least b/c you need to do so to figure out the First Amendment analysis. You can likely make out a prima facie case; the harder part is the 1A. Is the advantage/benefit coming from an individual’s identity or from the corpus? That’s a different type of use in the prototypical celebrity or even ordinary person in an ad. That gets to the why of the claim.

A: focusing on dignity and control in the paper. Argument is that the individuals matter because each individual makes the dataset better.

Q: I’m irritated about lots of things corporations do: a traditional phone book can expose information. Why this?

A: Individual rights focus: some people can object even if others don’t. How do you distinguish between “too much” appropriation and fragmented unprotectable data, which is why he’s thinking about the sports cases. There’s something about the image that makes the courts confident there’s a tort there, making it too much. [That ignores a bunch of the news reporting cases, but newsworthiness might be the distinction there.]

AI as Inventor in the Cambridge Handbook of Artificial Intelligence

Chris Mammen

Fed Cir unambiguously concluded that inventors must be human; UKIPO, EIPO have agreed, but the conversation is not done globally/at the marginal cases. One success in South Africa for an AI inventor, but there’s no statutory definition of inventor there and there is no substantive examination of patents, so TBD if/when litigated.

Humanists on one side, industrial policy advocates on the other: both can claim some justification in past statements about patent policy. Practical reality: AI is lab equipment; assisting a human researcher. Sometimes lab assistants get named on a patent but sometimes not. Not just a formal Q of whether the statute requires a human, but the act of invention itself: requires conception, which requires a theory of mind; diligence, which requires a theory of work; and reduction to practice, which requires ability to build the invention or write the patent application. We wave our hands about all of those but we need to address each one for the “AI inventor.”

UK decision is under review: primary holding says an inventor has to be a human. But UK application has some formalities that narrowed the Q. (1) requirement to list inventor to best of applicant’s knowledge. There may be room to say “I don’t know, but I have a right to apply for this patent.” Consider accession doctrines as well.

In response to Q: If we’re going to have a rights regime for AI, we also need a responsibility regime. We’re a long way from the point where I can depose an AI about its claim of inventorship.

No comments: