HKU Law and Technology Centre

Wrapping up this year's workshop in law and technology is Professor Alexander Stremitzer from ETH Zürich who examined the idea of automated law enforcement through an empirical study of the Google Fonts case. Those who missed Alex's talk the first time can watch it here: https://lnkd.in/gxHs2tNe.

👨‍💻 Can big data help predict and prevent professional misconduct by lawyers? 👨‍💼
Professor Albert Yoon from the University of Toronto Faculty of Law shared insights into the determinants of attorney misconduct based on quantitative analyses of a non-public dataset of all lawyers 👨‍💼 admitted to practice in the state of California 🇺🇸 from 1990 to 2023. Professor Yoon found among other things that lawyers with the highest rates of investigation and discipline are drawn disproportionately from graduates from less selective law schools and those receiving low passing scores on the state bar examination ✍ . Gender and ethnicity 👩👦🏿 are also strongly associated with investigation and discipline.

While emphasizing that correlation does not equate to causation, Professor Yoon called for an evidence-based approach towards regulating lawyers. For example, professional bodies can use statistical predictors to identify attorneys 👨‍⚖️ at high risk of discipline and provide them with training and resources to avoid the most common forms of misconduct.

Professor Simon N.M. Young from the HKU Faculty of Law provided thought-provoking commentary and raised the question of whether bar examination scores which are not disclosed to test takers themselves can legitimately be employed for the purposes of monitoring and intervention.

What do you think of Professor Yoon’s proposal? 🤔 Will the public benefit from the application of predictive analytics to the regulation of the legal profession? 👨‍👩‍👧‍👦👨‍💼⚖️

✨ How will artificial intelligence transform legal practice? 👨‍⚖️👩‍⚖️⚖️ In a wide-ranging reflection on the future of the legal profession, Professor Jonathan H. Choi from the University of Southern California Gould School of Law argues that for the foreseeable future, AI will complement and not substitute for human labor. Cindy Li from Futu Holdings Limited was the panelist of the evening talk.

Based on an empirical study of how AI assistance impacts the performance of law students on a variety of legal tasks, Professor Choi suggests that
• AI will have an equalizing effect ⚖️ : it will enhance the work product of the less skilled 👨‍🏭 while increasing the productivity 📈 — though not necessarily the quality — of those who are more skilled. 👨‍💼
• AI may increase the demand for legal professionals by making their services more cheaply and readily available. 👨‍👩‍👧‍👦

What do you think? Do you envision a day when AI systems will replace lawyers and judges? 🖥️🆚👨‍💼👨‍⚖️

🎨 𝗥𝗲𝗰𝗼𝗴𝗻𝗶𝘇𝗮𝗯𝗹𝗲 𝗛𝘂𝗺𝗮𝗻 𝗜𝗻𝗽𝘂𝘁, 𝗖𝗼𝗽𝘆𝗿𝗶𝗴𝗵𝘁𝗮𝗯𝗹𝗲 𝗔𝗜𝗚𝗖 🤖⚖️
On April 11, we were pleased to welcome Professor Renjun Bian (Peking University Law School) to the Advanced Seminar on Law and Technology at the University of Hong Kong, where she shared her latest research on the evolving threshold for copyright protection in the age of generative AI.

🔍 𝗟𝗲𝗰𝘁𝘂𝗿𝗲 𝗛𝗶𝗴𝗵𝗹𝗶𝗴𝗵𝘁𝘀:
Professor Bian examined how courts in the U.S. and China diverge in their approaches to the copyrightability of AI-generated content (AIGC):
🔹 In the U.S., courts apply a “𝗰𝗿𝗲𝗮𝘁𝗶𝘃𝗲 𝗰𝗼𝗻𝘁𝗿𝗼𝗹” standard, requiring evidence of human authorship over the final output (e.g., 𝘛𝘩𝘦𝘢𝘵𝘳𝘦 𝘋’𝘰𝘱𝘦𝘳𝘢 𝘚𝘱𝘢𝘵𝘪𝘢𝘭, 𝘙𝘰𝘴𝘦 𝘌𝘯𝘪𝘨𝘮𝘢).
🔹 In contrast, Chinese courts adopt a more flexible “𝗽𝗲𝗿𝘀𝗼𝗻𝗮𝗹 𝗰𝗵𝗼𝗶𝗰𝗲 𝗮𝗻𝗱 𝗮𝗿𝗿𝗮𝗻𝗴𝗲𝗺𝗲𝗻𝘁” standard, granting protection based on prompting techniques and aesthetic judgment (e.g., 𝘚𝘱𝘳𝘪𝘯𝘨 𝘉𝘳𝘦𝘦𝘻𝘦 𝘉𝘳𝘪𝘯𝘨𝘴 𝘛𝘦𝘯𝘥𝘦𝘳𝘯𝘦𝘴𝘴).

But the central question remains: 𝗛𝗼𝘄 𝗺𝘂𝗰𝗵 𝗵𝘂𝗺𝗮𝗻 𝗶𝗻𝗽𝘂𝘁 𝗶𝘀 𝗲𝗻𝗼𝘂𝗴𝗵—𝗮𝗻𝗱 𝗳𝗼𝗿 𝘄𝗵𝗼𝗺?

She proposed a novel reframing: shifting the focus from protecting authors to considering 𝘁𝗵𝗲 𝘃𝗮𝗹𝘂𝗲 𝗼𝗳 𝗵𝘂𝗺𝗮𝗻 𝗶𝗻𝗽𝘂𝘁 𝗳𝗼𝗿 𝗲𝗻𝗱 𝘂𝘀𝗲𝗿𝘀. If human contributions make AIGC more valuable or recognizable to users, perhaps that should define the threshold for copyright protection.

🧪 𝗘𝗺𝗽𝗶𝗿𝗶𝗰𝗮𝗹 𝗦𝘁𝘂𝗱𝘆 (𝗡 = 𝟵𝟯𝟵):
To explore this, her team conducted a two-part experiment:
1️⃣ 𝗥𝗲𝗰𝗼𝗴𝗻𝗶𝘁𝗶𝗼𝗻 𝗧𝗲𝘀𝘁 – Can users distinguish AIGC with vs. without human input?
2️⃣ 𝗩𝗮𝗹𝘂𝗮𝘁𝗶𝗼𝗻 𝗧𝗲𝘀𝘁 (𝗼𝗻𝗴𝗼𝗶𝗻𝗴) – Do users perceive AIGC with human input as more valuable?

Preliminary results suggest that 𝗲𝘅𝗽𝗿𝗲𝘀𝘀𝗶𝘃𝗲 𝗶𝗻𝗽𝘂𝘁 (e.g., hand-drawn sketches transformed by AI) is the most recognizable and valued form of human contribution. These findings point toward a more nuanced copyright framework—one that considers not just the presence of human input, but its perceptibility and value to users.

The session was moderated by Associate Professor Benjamin Minhao Chen, Director of the Law and Technology Centre, The University of Hong Kong, and joined by students and colleagues interested in copyright, technology, and the future of creative authorship.

💡 Astounding progress in the development of Large Language Models has inspired proposals for using them to discern the ordinary meaning of contractual and statutory language. Consider for example Snell v. United Specialty Insurance Co. where a federal appellate judge in the United States👨‍⚖️ 🇺🇸 consulted a LLM as to whether the installation of an in-ground trampoline qualified as “landscaping”. In this exciting presentation, Professor Jonathan H. Choi from the University of Southern California Gould School of Law demonstrates that LLMs are unreliable oracles. 🤨 Despite being trained on incredibly large datasets:
👉 A LLM often gives substantively different answers to semantically identical prompts
👉 LLMs frequently differ between themselves as to the answer to a given prompt

In short, those who hope that LLMs can bring predictability and certainty in the law are going to be disappointed--for now. 🫤

👩‍⚖️ Justice Amy Barrett of the United States Supreme Court urged all Americans to read the Court’s opinions. 🏛️ “When Congress enacts the law, something driven by policy, you just have the bottom line…there is no explanation of reasoning behind it because it is just the result that matters. But that’s not how the Court works.” Can AI 👩‍💻 make Supreme Court decisions more accessible—and the Supreme Court 🏛️ itself, more legitimate—to the public? 🧐 Aniket Kesari from Fordham University School of Law (Fordham Law) presented an experiment on U.S. citizens that shows that exposure to legal reasoning
- can trigger a negative reaction among those opposed to the Court’s decisions in politically charged cases
- can increase approval of the Court’s decisions in less salient cases
- has no significant effect on the Court’s institutional legitimacy.

Conducted with collaborators Elliott Ash from ETH Zürich, Suresh Naidu from Columbia University, Lena Song from University of Illinois Urbana-Champaign and Dominik Stammbach from Princeton University, Aniket’s study suggests that better legal understanding among the public will not help the Court in navigating these divisive times. 🧐

Recordings of the talk is available at https://lnkd.in/gy-h4Uw7.

Suppose a consumer is harmed by a good or service that is supplied by a provider through a platform. Should the provider or the platform be liable for the harm? 🤔 Under partial strict liability, damages are shared between the provider and the platform irrespective of fault. Gerd Muehlheusser from University of Hamburg applies microeconomic modelling to think about how partial strict liability affects the provider’s investment in the safety of its product.

⭐Here’s the key takeaway:
If the platform is a monopoly🕴‍, having it bear a larger share of liability could result in a safer product being offered to consumers. This is because the platform has enough skin in the game to encourage the producer to spend money on developing a safer product.😊The platform will do this by setting prices below what would be expected in a monopoly setting, thereby increasing the quantity demanded by the public👨‍👩‍👧‍👦and supplied by the provider 🏭.

In short: legal rules ⚖️ can enhance product safety even when these rules do not always assign full liability for harm to the producer!🧐

Recording of the talk is available at https://lnkd.in/g4P9ci_2.

On March 18, Arna Woemmel from the of Hamburg presented her latest research: “Algorithmic Fairness: The Role of Beliefs”.

Arna shared findings from her economic experiment showing that algorithms explicitly designed to be non-discriminatory toward protected groups can backfire when used by discriminatory human decision-makers. 🧑‍

𝗪𝗵𝘆? Because these decision-makers are less likely to accept such 'fair' algorithms — due to their own biased beliefs about protected groups —these fairness interventions fail to reduce discrimination. In fact, they can lead to even higher levels of discrimination overall. 🤨

🧠 𝗞𝗲𝘆 𝘁𝗮𝗸𝗲𝗮𝘄𝗮𝘆: Technical fairness is not enough. To reduce discrimination in practice, we must understand and address how human biases shape the use of algorithmic tools — not just how bias enters through training data. 👩‍💻

Thank you Arna for an insightful talk that challenges assumptions about fairness-by-design and highlights the need to integrate behavioral insights into algorithm governance. 👏

Want to learn more about AI 🤖, law and tech? 👩‍💻 Sign up for the talk series by the HKU Law and Technology Centre at http://lawtech.hk/events! 🚀