Month: March 2024

Solow-Niederman on Do Cases Generate Bad AI Law (?)

Alicia Solow-Niederman (GW Law) has posted “Do Cases Generate Bad AI Law?” (Columbia Science and Technology Law Review, Forthcoming) on SSRN. Here is the abstract:

There’s an AI governance problem, but it’s not (just) the one you think. The problem is that our judicial system is already regulating the deployment of AI systems—yet we are not coding what is happening in the courts as privately driven AI regulation. That’s a mistake. AI lawsuits here and now are determining who gets to seek redress for AI injuries; when and where emerging claims are resolved; what is understood as a cognizable AI harm and what is not, and why that is so.

This Essay exposes how our judicial system is regulating AI today and critically assesses the governance stakes. When we do not situate the generative AI cases being decided by today’s human judges as a type of regulation, we fail to consider which emerging tendencies of adjudication about AI are likely to make good or bad AI law. For instance, litigation may do good agenda-setting and deliberative work as well as surface important information about the operation of private AI systems. But adjudication of AI issues can be bad, too, given the risk of overgeneralization from particularized facts; the potential for too much homogeneity in the location of lawsuits and the kinds of litigants; and the existence of fundamental tensions between social concerns and current legal precedents.

If we overlook these dynamics, we risk missing a vital lesson: AI governance requires better accounting for the interactive relationship between regulation of AI through the judicial system and more traditional public regulation of AI. Shifting our perspective creates space to consider new AI governance possibilities. For instance, litigation incentives (such as motivations for bringing a lawsuit, or motivations to settle) or the types of remedies available may open up or close down further regulatory development. This shift in perspective also allows us to see how considerations that on their face have nothing to do with AI – such as access to justice measures and the role of judicial minimalism – in fact shape the path of AI regulation through the courts. Today’s AI lawsuits provide an early opportunity to expand AI governance toolkits and to understand AI adjudication and public regulation as complementary regulatory approaches. We should not throw away our shot.

Savelka & Ashley on LLMs in Zero-Shot Semantic Annotation of Legal Texts

Jaromir Savelka (Carnegie Mellon University) and Kevin Ashley (U Pitt Law) have posted “The Unreasonable Effectiveness of Large Language Models in Zero-Shot Semantic Annotation of Legal Texts” (Frontiers in Artificial Intelligence, Vol. 6, p. 1, 2023) on SSRN. Here is the abstract:

The emergence of ChatGPT has sensitized the general public, including the legal profession, to large language models’ (LLMs) potential uses (e.g., document drafting, question answering, and summarization). Although recent studies have shown how well the technology performs in diverse semantic annotation tasks focused on legal texts, an influx of newer, more capable (GPT-4) or cost-effective (GPT-3.5-turbo) models requires another analysis. This paper addresses recent developments in the ability of LLMs to semantically annotate legal texts in zero-shot learning settings. Given the transition to mature generative AI systems, we examine the performance of GPT-4 and GPT-3.5-turbo(-16k), comparing it to the previous generation of GPT models, on three legal text annotation tasks involving diverse documents such as adjudicatory opinions, contractual clauses, or statutory provisions. We also compare the models’ performance and cost to better understand the trade-offs. We found that the GPT-4 model clearly outperforms the GPT-3.5 models on two of the three tasks. The cost-effective GPT-3.5-turbo matches the performance of the 20× more expensive text-davinci-003 model. While one can annotate multiple data points within a single prompt, the performance degrades as the size of the batch increases. This work provides valuable information relevant for many practical applications (e.g., in contract review) and research projects (e.g., in empirical legal studies). Legal scholars and practicing lawyers alike can leverage these findings to guide their decisions in integrating LLMs in a wide range of workflows involving semantic annotation of legal texts.

Sag on Fairness and Fair Use in Generative AI

Matthew Sag (Emory U Law) has posted “Fairness and Fair Use in Generative AI” (Fordham Law Review, Forthcoming) on SSRN. Here is the abstract:

Although we are still a long way from the science fiction version of artificial general intelligence that thinks, feels, and refuses to “open the pod bay doors”, recent advances in machine learning and artificial intelligence (“AI”) have captured the public’s imagination and lawmakers’ interest. We now have large language models (“LLMs”) that can pass the bar exam, carry on (what passes for) a conversation on almost any topic, create new music, and create new visual art. These artifacts are often indistinguishable from their human authored counterparts, and yet can be produced at a speed and scale that transcends human ability.

Generative AI systems like the GPT and LLaMA language models and the Stable Diffusion and Midjourney text-to-image models were built by ingesting massive quantities of text and images from the Internet. This was done with little or no regard to whether those works were subject to copyright and whether the authors would object. The rise of generative AI poses important questions for copyright law. These questions are not entirely new, however. Generative AI gives us yet another context to consider copyright’s most fundamental question; where do the rights of the copyright owner end, and the freedom to use copyrighted works begin?

My aim in this Essay is not establish that generative AI is, or should be, noninfringing; it is to outline an analytical framework for making that assessment in particular cases.

The Essay is based on my keynote address at delivered the 12th Annual Peter A. Jaszi Distinguished Lecture on Copyright Law at American University Washington College of Law on September 28, 2023.

Guggenberger on Moderating Monopolies

Nikolas Guggenberger (U Houston Law Center) has posted “Moderating Monopolies” (Berkeley Technology Law Journal, Vol. 38, No. 1, 2023) on SSRN. Here is the abstract:

Industrial organization predetermines content moderation online. At the core of today’s dysfunctions in the digital public sphere is a market power problem. Meta, Google, Apple, and a few other digital platforms control the infrastructure of the digital public sphere. A tiny group of corporations governs online speech, causing systemic problems to public discourse and individual harm to stakeholders. Current approaches to content moderation build on a deeply flawed market structure, addressing symptoms of systemic failures at best and cementing ailments at worst.

Market concentration creates monocultures for communication susceptible to systemic failures and raises the stakes for individual content moderation decisions, like takedowns of posts or bans of individuals. As these decisions are inherently prone to errors, those errors are magnified by the platforms’ scale and market power. Platform monopolies also harm individual stakeholders: persisting monopolies lead to higher prices, lower quality, or less innovation. As platforms’ services include content moderation, degraded services may increase the error rate of takedown decisions and over-expose users to toxic content, misinformation, or harassment. Platform monopolies can also get away with discriminatory and exclusionary conduct more easily because users lack voice and exit opportunities.

Stricter antitrust enforcement is imperative, but contemporary antitrust doctrine alone cannot hope to provide sufficient relief to the digital public sphere. First, a narrowly understood consumer welfare standard overemphasizes easily quantifiable, short-term price effects. Second, the levels of concentration necessary to trigger antitrust scrutiny far exceed those of a market conducive to pluralistic discourse. Third, requiring specific anticompetitive conduct, the focal point of current antitrust doctrine, ignores structural dysfunction mighty bottlenecks create in public discourse, irrespective of the origins or even benevolent exercise of their power.

In this Article, I suggest three types of remedies to address the market power problem behind the dysfunctions in the digital public sphere. First, mandating active interoperability between platforms would drastically reduce lock-in effects. Second, scaling back quasi-property exclusivity online would spur follow-on innovation. Third, no-fault liability and broader objectives in antitrust doctrine would establish more effective counterweights to concentrating effects in the digital public sphere. While these pro-competitive measures cannot provide a panacea to all online woes, they would lower the stakes of inevitable content moderation decisions, incentivize investments in better decision-making processes, and contribute to healthier pluralistic discourse.

Douek on The Meta Oversight Board and the Empty Promise of Legitimacy

Evelyn Douek (Stanford Law School) has posted “The Meta Oversight Board and the Empty Promise of Legitimacy” (Harvard Journal of Law & Technology, Vol. 37, 2024 Forthcoming) on SSRN. Here is the abstract:

The Meta Oversight Board is an audacious experiment in self-regulation by one of the world’s most powerful corporations, set up to oversee one of the largest systems of speech regulation in history. In the few years since its establishment, the Board has in some ways defied its many skeptics, by becoming a consistent and accepted feature of academic and public discourse about content moderation. It has also achieved meaningful independence from Meta, shed light on the otherwise completely opaque processes within the corporation, instantiated meaningful reforms to Meta’s content moderation systems, and provided an avenue for greater stakeholder engagement in content moderation decision-making. But the Board has also failed to live up to core aspects of its role, in ways that have gone underappreciated. The Board has consistently shied away from answering the hardest and most controversial questions that come before it—that is, the very questions it was set up to tackle—and has not provided meaningful yardsticks for quantifying its actual impact. Understanding why the Board eschews these questions, and why it has nevertheless managed to acquire a significant amount of institutional legitimacy, suggests important lessons about institutional incentives and the revealed preferences of stakeholders in content moderation governance. Ultimately, this Article argues, the current political environment incentivizes a kind of oversight that is formalistic and unmoored from substantive goals. This is a problem that plagues regulatory reform far beyond the Board itself, and shows that generalized calls for “more legitimate” content moderation governance are underspecified and may, as a result, incentivize poor outcomes.

Surden on Computable Law and Artificial Intelligence

Harry Surden (U Colorado Law) has posted “Computable Law and Artificial Intelligence” (Cambridge Handbook of Private Law and Artificial Intelligence (forthcoming 2024)) on SSRN. Here is the abstract:

This article explores the theory and application of “Computable Law”.

‘Computable Law’ is a research area focused on the creation and use of computer models of laws.

What does it mean to model a law computationally? There are a few broad approaches. In one method, researchers begin with traditional, written legal sources of law – such as statutes, contracts, administrative regulations, and court opinions – and identify legal rules that they wish to model. They then aim to ‘translate’ aspects of these legal obligations into comparable sets of organised data, programming instructions, and other forms of expression that computers can easily process. In that approach, one begins with a familiar legal text written in a ‘natural language’ such as English, and then aims to represent qualities of the legal obligations described – such as their structure, meaning, or application – in terms of data, programming rules and other highly organised forms of expression that are easier for computers to handle.

The other approach allows us to express legal obligations as data from the outset. There, one begins with laws expressed as computer data in their initial form – a departure from the written-language through which laws have traditionally been conveyed. An example of this approach can be found in the so-called data-oriented,‘computable contracts’.

These are legal agreements created electronically, whose core terms are expressed largely as data rather than as written paragraphs, and which are frequently used in finance, electronic commerce, cryptocurrency, and other areas.

Through this ‘data-oriented’ method we are still ultimately able to display legal obligations in forms that people can understand, such as in language or visually on a computer screen. However, what is interesting is that the human-understandable versions are typically derived upwards from the underlying data. In other words, one can present to users what appear to be ordinary written legal documents on a screen or on paper, but the contents of those documents are actually generated by processing lower-level computer data. In those cases, it is sometimes best to think of the law’s native data-oriented representation as the authoritative version (or source of ‘ground-truth’) for information about the legal obligations.

Bliss on Teaching Law in the Age of Generative AI

John Bliss (U Denver Law) has posted “Teaching Law in the Age of Generative AI” (Jurimetrics, forthcoming) on SSRN. Here is the abstract:

With the rise of large language models capable of passing law school exams and the Unified Bar Exam, how should legal educators prepare their students for an age of transformative AI advances? Text-generating AI is poised to become a standard tool of legal research and writing, as it is being integrated in legal research and word processing applications (such as LexisNexis and Microsoft Word) that automate the drafting of legal documents based on human prompts. This Article explores the implications of these developments for legal education, focusing on pedagogy, curriculum, and assessment.

The Article draws from four key perspectives relevant to the use of generative AI in law teaching: a survey of law students who participated in an AI-integrated course; a national survey of law faculty; an overview of the current state and projected futures of AI in the legal profession; and a summary of findings from the remarkably extensive educational literature that has arisen around the globe exploring the use of ChatGPT in different teaching contexts. These perspectives tend to support the development of an AI-integrated legal education. Yet, most of the surveyed law faculty, even those who strongly agreed that students should be prepared to use and critically evaluate generative AI, emphasized that they were uninformed about this technology and unsure how to proceed.

This Article provides guidance, recommending that legal educators begin teaching with emerging AI tools, while exploring how implementation might vary across the legal curriculum. These recommendations are based on a number of factors, including consideration of how AI-integrated teaching may affect emerging professional competencies, traditional learning goals, academic integrity, and equity among students. The Article concludes by offering practical suggestions for incorporating generative AI in law teaching, including examples of exercises where students collaborate with generative AI in their writing, evaluate AI outputs, create their own AI tutors and debate partners, role-play with chatbots in classroom simulations, and reflect on the responsible use of generative AI in the legal profession.