Henderson, Li, Jurafsky, Hashimoto, Lemley & Liang on Foundation Models and Fair Use

Peter Henderson (Stanford University), Xuechen Li (same), Dan Jurafsky (same), Tatsunori Hashimoto (same), Mark A. Lemley
(Stanford Law School), and Percy Liang (Stanford Computer Science) have posted “Foundation Models and Fair Use” on SSRN. Here is the abstract:

Existing foundation models are trained on copyrighted material. Deploying these models can pose both legal and ethical risks when data creators fail to receive appropriate attribution or compensation. In the United States and several other countries, copyrighted content may be used to build foundation models without incurring liability due to the fair use doctrine. However, there is a caveat: If the model produces output that is similar to copyrighted data, particularly in scenarios that affect the market of that data, fair use may no longer apply to the output of the model. In this work, we emphasize that fair use is not guaranteed, and additional work may be necessary to keep model development and deployment squarely in the realm of fair use. First, we survey the potential risks of developing and deploying foundation models based on copyrighted content. We review relevant U.S. case law, drawing parallels to existing and potential applications for generating text, source code, and visual art. Experiments confirm that popular foundation models can generate content considerably similar to copyrighted material. Second, we discuss technical mitigations that can help foundation models stay in line with fair use. We argue that more research is needed to align mitigation strategies with the current state of the law. Lastly, we suggest that the law and technical mitigations should co-evolve. For example, coupled with other policy mechanisms, the law could more explicitly consider safe harbors when strong technical tools are used to mitigate infringement harms. This co-evolution may help strike a balance between intellectual property and innovation, which speaks to the original goal of fair use. But we emphasize that the strategies we describe here are not a panacea and more work is needed to develop policies that address the potential harms of foundation models.

Asay on the DMCA’s Anti-Circumvention Provisions

Clark D. Asay (Brigham Young Law) has posted “An Empirical Study of the DMCA’s Anti-Circumvention Provisions” on SSRN. Here is the abstract:

The DMCA has been a flashpoint during most of its twenty-five-year existence. One of the most controversial parts of the DMCA is Section 1201. Among other things, Section 1201 prohibits third parties from circumventing certain controls to copyrighted content or trafficking in tools that enable circumvention of technological controls. However, despite its nearly quarter-of-a-century lifespan, we know very little about Section 1201 empirically. While certain aspects of the broader DMCA have received empirical assessments, Section 1201 has not. Our understanding of Section 1201 is largely based on anecdotal evidence, in the form of leading opinions from historically prominent copyright circuits. But this anecdotal evidence is hardly a solid basis for ongoing discussions about how Section 1201 is performing and whether it needs revising. In this Article, we seek to address these and other issues.

To do so, we conducted a broad-based search of Westlaw to collect every issued opinion, whether reported or not, where a court purported to apply some part of Section 1201. We then reviewed these cases to glean as much useful information about Section 1201 as possible. This review led to a number of important and, in some cases, surprising results. First, Section 1201 opinions are a relative rarity. In the nearly quarter of a century since the DMCA’s enactment, we could find only a little over 200 opinions, with only about sixty of those being published. The average number of opinions during the DMCA’s existence has been around nine annually, which pales in comparison to other types of copyright cases. Second, despite the Second Circuit receiving much attention in anecdotal accountings of Section 1201, courts within it issue Section 1201 opinions infrequently. The Ninth Circuit is the dominant Section 1201 court, both in terms of citations to its opinions and overall number of opinions, and the Sixth and Eleventh Circuits both issue more Section 1201 opinions than the Second Circuit. This result stands in contrast to other types of copyright litigation, where the Second Circuit is a behemoth. Third, the most common subject matter in dispute in Section 1201 cases is computer software, followed distantly by audiovisual material such as movies. Music stands in last place, showing up in only a couple issued opinions. Debates at the time of the DMCA’s enactment were informed by widespread fears of copyright infringement relating to digital music and other types of digital content. Yet Section 1201 has resulted in but few litigations involving those subject matters. Fourth, suits and defaults against individuals happen frequently in the Section 1201 context, with courts often assessing large statutory damages against those individuals. As we discuss in the paper, this result raises important equity issues. Fifth, despite Section 1201 including a number of statutory exceptions, these exceptions basically never make their way into issued opinions. Fair use, too, only infrequently enters courts’ Section 1201 discussions. This means, effectively, that the primary way to escape Section 1201 liability is through administrative exceptions granted by the Library of Congress on a triennial basis. But as we shall see, this process has significant holes. Finally, plaintiffs disproportionately win Section 1201 cases. This result is somewhat bloated because of the frequency of defaults against individuals. Setting these aside, plaintiffs still enjoy tremendous success under Section 1201. However, when looking at opinions only outside of the Ninth Circuit, win rates become mostly even.

I conclude with several calls for DMCA reform. These include bolstering statutory exceptions and more closely tying Section 1201 to copyright infringement. Pursuing these reforms, I argue, will more faithfully align Section 1201 with its purported objectives.

Whalen & Zingg on The Patent-Eligibility of Artificial Intelligence after Alice Corp. v. CLS Bank International

Ryan Whalen (The University of Hong Kong – Faculty of Law) and Raphael Zingg (Waseda University) have posted “Innovating under Uncertainty: The Patent-Eligibility of Artificial Intelligence after Alice Corp. v. CLS Bank International” (Research in Law and Economics, Volume 30 (2022)) on SSRN. Here is the abstract:

Artificial intelligence-related inventions raise complex questions of how to define the boundaries around patentable subject matter. In the United States, many claim that the recent doctrinal developments by the Supreme Court have led to incoherence and excessive uncertainty within the innovation community. In response, policymakers and stakeholders have suggested legislative amendments to address these concerns. We first review these developments, and subsequently use the patent examination record to empirically test the claims of increased uncertainty. We find that, although uncertainty did spike following the Supreme Court’s holding in Alice, it quickly returned to levels comparable to its historic norm. This has implications both for those advocating for legislative changes to the law of eligible subject matter, as well as other jurisdictions considering adopting a test similar to that applied in Alice.

Friedmann on Digital Single Market, First Stop to The Metaverse

Danny Friedmann (Peking University School of Transnational Law) has posted “Digital Single Market, First Stop to The Metaverse: Counterlife of Copyright Protection Wanted” (Law & Econ of the Digital Transformation, Klaus Mathis and Avishalom Tor, eds. (Springer, 2022 Forthcoming)) on SSRN. Here is the abstract:

Building upon the “fair use by design” concept of Niva Elkin-Koren, this chapter is exploring how artificial intelligence can be used to implement exceptions or limitations.

Section 2 will give a brief overview of the evolution of the copyright acquis in the US and EU, in regard to platforms, discuss the implications of strict liability in an era of massive online use and infringements, which already has pushed and, a fortiori, will push platforms in the metaverse in the direction of automatic, scalable solutions, which on their turn, will increase the need for sufficient safeguards of unauthorized but legal use that falls under an exception or limitation.

Section 3 introduces the implications of the metaverse in regard to intermediary liability of copyright infringement and the need for “breathing space” for users and experimenting.

Section 4 explores the safeguards for legitimate use of content, which includes exceptions and limitations.

Building on Elkin-Koren’s “fair use by design” concept, Section 5 provides the prerequisites of designing algorithmic exceptions or limitations, and whether automated content recognition tools should be qualified as “high risk AI” under the proposed Artificial Intelligence Act, and incentives against over-blocking.

Kraizberg on Non-fungible Tokens: A Bubble or the End of an Era of Intellectual Property Rights

Elli Kraizberg (Bar-Ilan University) has posted “Non-fungible Tokens: A Bubble or the End of an Era of Intellectual Property Rights” on SSRN. Here is the abstract:

The viability of the exponentially growing non-fungible token (NFT) markets is evaluated by identifying potential value-generating mechanisms that may be rationalized. NFTs are claimed to securitize “ownership rights short of use”. This paper evaluates the likelihood that NFTs will replace existing mechanisms that protect producers’ rightful claim to use their assets, or replace the need to apply the legal code that governs intellectual property rights (IPR). A condition for this shift is derived for a category of assets whose use or consumption does not reduce their scarce supply.

Tang on Privatizing Copyright

Xiyin Tang (UCLA School of Law; Yale Law School) has posted “Privatizing Copyright” (Michigan Law Review, Forthcoming) on SSRN. Here is the abstract:

Much has been written, and much is understood, about how and why digital platforms regulate free expression on the Internet. Much less has been written— and even much less is understood—about how and why digital platforms regulate creative expression on the Internet—expression that makes use of others’ copyrighted content. While § 512 of the Digital Millennium Copyright Act regulates user-generated content incorporating copyrighted works, just as § 230 of the Communications Decency Act regulates other user speech on the Internet, it is, in fact, rarely used by the largest Internet platforms—Facebook and YouTube. Instead, as this Article details, creative speech on those platforms is instead governed by a series of highly confidential licensing agreements entered into with large copyright holders.

Yet despite the dominance of private contracting in ordering how millions of pieces of digital content are made and distributed on a daily basis, little is known, and far less has been written, on just what the new rules governing create expression are. This is of course, by design: these license agreements contain strict confidentiality clauses that prohibit public disclosure. This Article, however, pieces together clues from publicly-available court filings, news reporting, and leaked documents. The picture it reveals is a world where the substantive law of copyright is being quietly rewritten—by removing the First Amendment safeguard of fair use, by inserting in a new moral right for works that Congress had deemed, in the Copyright Act, ineligible for moral rights protection, and, through other small provisions in the numerous agreements digital platforms negotiate with rightsholders, influencing and reshaping administrative, common, and statutory copyright law. Further still, recent changes or lobbied-for changes to copyright’s statutory law seek to either enshrine the primacy of such private contracting or altogether remove copyright rule-making processes from government oversight, shielding copyright’s public law from independent considerations of public policy and public scrutiny.

Changing copyright’s public law to enshrine the primacy of such private ordering insulates the new rules of copyright from the democratic process, from public participation in, and from public oversight of, the laws that shape our daily lives. Creative expression on the Internet now finds itself at a curious precipice: a seeming glut of low-cost, or free, content, much of which is created directly by, and distributed to, users—yet increasingly regulated by an opaque network of rules created by a select few private parties. An understanding of the Internet’s democratizing potential for creativity is incomplete without a concomitant understanding of how the new private rules of copyright may shape, and harm, that creativity.

Hugendubel on Blockchain Technology and Intellectual Property – A Basic Introduction

Julia Hugendubel (CMS) has posted “Blockchain Technology and Intellectual Property – A Basic Introduction” on SSRN. Here is the abstract:

Blockchain technology is predicted to have a major impact on the intellectual property (IP) ecosystem. More and more projects are being launched, both in the public and private sector. The World Intellectual Property Organization (WIPO) has built up a Blockchain Task Force and is preparing a new WIPO Standard to encompass all types of IP rights and the entire IP lifecycle; the German Government published a strategy paper on blockchain with a chapter on applications in the creative arts sector; a European Blockchain Service Infrastructure is being built up; the European Union Intellectual Property Office (EUIPO) established an Anti-Counterfeiting Forum as part of the broader EU strategy to create a blockchain ecosystem and now has its own blockchain for trademarks and designs in the EU. Furthermore, LVMH, with brands such as Louis Vuitton, developed its own blockchain to track luxury goods; Kodak started a blockchain initiative for image rights management; music and film streaming are offered on blockchain platforms; sports clubs discuss micro-licensing of their IP rights; digital fashion is created for distribution using blockchain.

Moreover, news abounds of blockchain-based non-fungible tokens (NFTs) representing basically everything imaginable – both digital and physical – in the IP ecosystem, to track the origin of art and all manner of products. However, interestingly enough, as early as 2012 a whitepaper was published linking the idea of smart property by Nick Szabo and digital collectibles to blockchain and at the same time introducing the first kind of NFTs, coloured coins.

The developments in the blockchain space, including in the area of IP, are progressing at a rapid pace, both from a technological and a value perspective. For example, the value of crypto art traded on blockchain from 2018 to 2020 was estimated at about 15 million US Dollar (with about 8.2 million US Dollar worth of crypto artwork in December 2020 ). In March 2021, the most expensive piece of artwork linked to an NFT in history was sold by Mike Winkelmann (Beeple) for 42,329.453 Ether, at the time worth 69,346,250 million US Dollar. The NFT marketplace OpenSea has set and then beat daily records several times, reaching a new peak of 322,982,301 million US Dollar of trading volume on 29 August 2021.

Nevertheless, many applications for IP matters are still in their infancy. The reason for this might lay, aside from the rather complex technological details, in uncertainties about their regulation and legal standing in court, such as the recognition of a legal binding smart contract.

Therefore, the article introduces the very basic features of blockchain technology and blockchain-based IP applications. The article then dives into concrete IP use cases that are currently offered and developed on the market. It also gives a general overview of opportunities and existing challenges for the IP ecosystem. The legal perspective is mainly a German and European one.

Moerland & Kafrouni on Online Shopping with Artificial Intelligence: What Role to play for Trade Marks?

Anke Moerland (Maastricht University – Department of International and European Law) and Christie Kafrouni have posted “Online Shopping with Artificial Intelligence: What Role to play for Trade Marks?” on SSRN. Here is the abstract:

The debate on how artificial intelligence (AI) influences intellectual property protection has so far mainly focussed on its effects for patent and copyright protection. Not much attention has been paid to the effects of artificial intelligence technology for trade mark law. In particular, what has not yet been sufficiently investigated is the question as to whether trade marks still fulfil their role in a world in which consumers are assisted by AI technology when purchasing in the online market place. In how far do we still need trade marks to avoid consumer confusion? Or do the other functions of trade marks justify their continuous protection? In view of the fact that intellectual property rights have a market-distorting effect, it is in society’s interest to question whether trade mark protection is still justified.

Goodyear on Embedding Permission Culture: A New Approach to the Server Test Quandary

Michael Goodyear has posted “Embedding Permission Culture: A New Approach to the Server Test Quandary” (75 Okla. L. Rev. __ (Forthcoming 2022)) on SSRN. Here is the abstract:

The practice of embedding—inserting code that displays content located elsewhere on the Internet—is ubiquitous online. Millions of users insert or encounter embeds daily. As a core type of link, embedding has helped disseminate information far and wide, furthering the goals of both copyright law and the Internet. For over a decade, embedding has been considered lawful under copyright law, guaranteed by a Ninth Circuit doctrine known as the server test, which holds that a person only displays a work when he or she hosts and serves it. This rule, which has greatly influenced the growth of the modern Internet, has recently come under siege, with two decisions from the U.S. District Court for the Southern District of New York threatening the harmony of the server test and the future of the practice of embedding.

This article enters the growing debate about the server test for two reasons. First, it offers legal and policy justifications for the server test, demonstrating that the preservation of the server test is desirable. Second, it considers alternative theories for permitting embedding. Finding that the alternative theories that have generally been proposed would be poor defenses for embedding, this article instead identifies a private ordering theory of permission-driven embedding that grants greater choices to content creators while preserving copyright and the Internet’s balance between exclusive rights and the spread of knowledge. As this article explains, permission-driven embedding is already here, with major online platforms already adopting it in part. Therefore, notwithstanding the server test’s prognosis, permission-driven embedding is part of the future of online content distribution.

Gervais on AI Derivatives: the Application to the Derivative Work Right to Literary and Artistic Productions of AI Machines

Daniel J. Gervais (Vanderbilt Law) has posted “AI Derivatives: the Application to the Derivative Work Right to Literary and Artistic Productions of AI Machines” (Seton Hall Law Review, Vol. 53, 2022) on SSRN. Here is the abstract:

This Article predicts that there will be attempts to use courts to try to broaden the derivative work right in litigation either to prevent the use of, or claim protection for, literary and artistic productions made by Artificial Intelligence (AI) machines. The Article considers the normative valence and the (significant) doctrinal pitfalls associated with such attempts. It also considers a possible legislative alternative, namely attempts to introduce a new sui generis right in AI productions. Finally, the Article explains how, whether such attempts succeed or not, the debate on rights (if any) in productions made by AI machines is distinct from the debate on text and data mining exceptions.