Andrew D. Selbst (UCLA Law), Suresh Venkatasubramanian (University of Utah), and I. Elizabeth Kumar (Brown University, Students) have posted “Deconstructing Design Decisions: Why Courts Must Interrogate Machine Learning and Other Technologies” (85 Ohio State Law Journal (forthcoming 2024)) on SSRN. Here is the abstract:

Technologies do not just come about. They are designed, and those design choices affect everything the technology touches. Yet unless a legal question directly implicates the technological design, courts are not likely to interrogate it. In this Article, we use examples from machine learning to demonstrate that the design choices matter even for cases where the legal questions do not involve technology directly. We start by describing formal “abstraction,” a fundamental design technique in computer science that treats systems and subsystems as defined entirely by their inputs, outputs, and the relationship that transforms inputs to outputs. We show how this technique causes the resulting technologies to be effectively making claims about responsibility and knowability that competes with courts’ own determinations. We further show that these claims are rendered invisible over time. Thus, we argue that courts must unearth—or deconstruct—the original design choices in order to understand the legal claims in a given case—even those cases that do not on their face appear to be about technological design.

There is, of course, a reasonable concern that courts are not capable or are not the best venue to make judgments about technological design. While we agree that courts are not the optimal front-line regulators of technology, we argue that they cannot avoid these questions as technologies begin show up in every type of case—a phenomenon that will only grow with time. But besides being forced to consider technology, courts are actually capable of doing so when motivated to. We demonstrate that in certain cases that clearly tee up technological design, such as products liability, copyright retransmission, and the functionality doctrines of intellectual property, courts have no problem diving in and questioning the design choices, asking what could have and should have been. Where courts can perform analysis in one arena, they can do so in another. Finally, through extended hypotheticals in the areas of negligence, discrimination, and criminal justice, we demonstrate how courts can effectively deconstruct technological design.

Herwig C.H. Hofmann (Universite du Luxembourg Law) has posted “Automated Decision-Making (ADM) in EU Public Law” on SSRN. Here is the abstract:

Decision making in EU public law and the implementation of policies is increasingly supported by automation. The understanding of the effects thereof on rights and procedures as well as concepts of how to ensure accountability of such automated decision making (ADM) in EU public law are evolving and are influenced by a developing legislative framework and interpretation by the CJEU as well as developing case law by courts in the EU and the Member States.

Automated decision-making (ADM) is based on software supporting, or replacing, elements of human decision making in implementation of EU law. ADM systems are deployed in an increasing amount of policy areas. Improved availability of information, advanced computation power and advanced forms of programming using fast evolving technologies to process such information produces benefits for decision-making. But integrating technological solutions into decision making procedures also risks introducing potential dysfunctionalities, diminishing individual rights, and reducing accountability.

A key feature of the integration of ADM technologies in various phases of decision-making is that it has a profound effect on procedures leading to the delivery of public policies in the EU. This has the potential to improve decisional quality and efficiency. But equally it can influence the realisation of key procedural values of public law in the EU. Influencing procedures on the basis of technical specifications without clear orientation towards values and rights in EU law risks a growing disconnect between real-life procedures and central values and principles of democratic societies operating under the rule of law, which in turn can result in increasing de-legitimisation of the exercise of public powers.

This chapter looks at central questions which the use of ADM in public decision-making procedures in the scope of EU law raises in terms of public law. It first analyses the role and the origin of information as source of decision making in EU public law and the automated thereof. It secondly looks at the central values and fundamental rights touched by ADM. Third, the chapter asks which requirements of technical design of ADM systems and their relation to the data basis, which are used as sources of information searches and analysis are necessary.

Surong Zhu (Beijing Foreign Studies University) and Guoyang Ma (Beijing Jiaotong University) have posted “The Chinese Path to Generative AI Governance” on SSRN. Here is the abstract:

The emergence of generative AI has brought significant development opportunities for the AI industry, but it has also triggered legal issues such as data leakage and technology abuse. Therefore, how to ensure the upward and positive development of generative AI technology has become a focus of attention for countries worldwide. China has taken the lead in legislative measures by introducing July 2023 the world’s first departmental regulation dedicated to generative AI-Interim Administrative Measures for Generative Artificial Intelligence Services. The Interim Measures is a product of the vertical iterative legislative model, upholding the governance attitude of encouraging development and drawing a bottom line, and adopting the governance strategy of inclusiveness and prudence, categorization and grading, complemented by the regulatory means of segmented governance and clear responsibilities. At the same time, the EU and the US have effectively addressed the risks posed by generative AI by taking existing legal norms as the cornerstone and adopting law enforcement and legislative measures. This article provides a detailed description of China’s regulatory ideas and specific methods. It compares the different views of China with those of the EU and the US, further commenting on the innovations and shortcomings of China’s program.

Beatriz Botero Arcila (Sciences Po Law School; Harvard Berkman Klein) has posted “Is it a Platform? Is it a Search Engine? It’s Chat GPT! The European Liability Regime for Large Language Models” (Journal of Free Speech Law, Vol. 3, Issue 2, 2023) on SSRN. Here is the abstract:

ChatGPT and other AI large language models (LLMs) raise many of the regulatory and ethical challenges familiar to AI and social media scholars: They have been found to confidently invent information and present it as fact. They can be tricked into providing dangerous information even when they have been trained to not answer some of those questions. Their ability to mimic a personalized conversation can be very persuasive, which creates important disinformation and fraud risks. They reproduce various societal biases because they are trained on data from the internet that embodies such biases, for example on issues related to gender and traditional work roles. Thus, like other AI systems, LLMs risk sustaining or enhancing discrimination perpetuating bias, and promoting the growth of corporate surveillance, while being technically and legally opaque. Like social media, LLMs pose risks associated with the production and dissemination of information online that raise the same kind of concern over the quality and content of online conversations and public debate. All these compounded risks threaten to distort political debate, affect democracy, and even endanger public safety. Additionally, OpenAI reported an estimated 100 million active users of ChatGPT in January 2023, which makes the potential for a vast and systemic impact of these risks a considerable one.

LLMs are also expected to have great potential. They will transform a variety of industries, freeing up professionals’ time to focus on different substantive matters. They may also improve access to various services by facilitating the production of personalized content, for example for medical patients or students. Consequently, one of the critical policy questions LLMs pose is how to regulate them so that some of these risks are mitigated while still encouraging innovation and allowing their benefits to be realized.

This Essay examines this question, with a focus on the liability regime for LLMs for speech and informational harms and risks in the European Union. Even though the AI Act has introduced some risk-mitigation obligations for this tool, we are still at least two years away from this becoming mandatory. That’s too long from now. However, because many of the risks these systems raise are risks to the information ecosystem this Essay argues they can and should be addressed, at the outset, with current content moderation law. The Essay thus proposes an interpretation of the newly enacted Digital Services Act that could apply to these tools when they are released in the market in a way that strongly resembles other intermediaries covered by content moderation laws, such as search engines.

Boris Babic (U Toronto) and I. Glenn Cohen (Harvard Law) have posted “The Algorithmic Explainability ‘Bait and Switch'” (Minnesota Law Review, Vol. 108, 2023) on SSRN. Here is the abstract:

Explainability in artificial intelligence and machine learning (“AI/ML”) is emerging as a leading area of academic research and a topic of significant regulatory concern. Indeed, a near-consensus exists in favor of explainable AI/ML among academics, governments, and civil society groups. In this project, we challenge this prevailing trend. We argue that for explainability to be a moral requirement — and even more so for it to be a legal requirement — it should satisfy certain desiderata which it currently does not, and possibly cannot. In particular, we will argue that the currently prevailing approaches to explainable AI/ML are (1) incapable of guiding our action and planning, (2) incapable of making transparent the actual reasons underlying an automated decision, and (3) incapable of underwriting normative (moral/legal) judgments, such as blame and resentment. This stems from the post hoc nature of the explanations offered by prevailing explainability algorithms. As we explain, that these algorithms are “insincere-by-design,” so to speak. And this renders them of very little value to legislators or policymakers who are interested in (the laudable goal of) transparency in automated decision making. There is, however, an alternative — interpretable AI/ML — which we will distinguish from explainable AI/ML. Interpretable AI/ML can be useful where it is appropriate, but represents real trade-offs as to algorithmic performance and in some instances (in medicine and elsewhere) adopting an interpretable AI/ML may mean adopting a less accurate AI/ML. We argue that it is better to face those trade-offs head on, rather than embrace the fool’s gold of explainable AI/ML.