AI Law & Innovation Institute

Robin’s Rules of Order for AI

This series of reflections by Professor Feldman provides principles for how to think about developments in AI.

Feldman Interviews Feldman: “How I Used AI in Writing a Book about AI”

This video podcast episode with Boris Feldman, described by Chambers USA as “The King of Silicon Valley,” examines Professor Feldman’s experience of using different AI models to learn about the math and science underlying AI. The discussion emphasizes that “AI serves best as human augmentation, not as human replacement.”

Government Activities Related to AI

For almost a decade, Professor Feldman and C4i have provided guidance on AI to government institutions, including:

• Technical advice to congressional committees and state officials on regulation of AI
• The Army Cyber Institute’s threat casting exercise on Weaponization of Data
• The GAO’s Artificial Intelligence Roundtable report to Congress on the future of AI
• The United Nations (address delivered to the 2023 General Assembly Science Summit on the impact of AI, delivered by Chancellor David Faigman)
• The Federal Trade Commission’s hearing on Emerging Competition, Innovation, and Market Structure Questions Around Algorithms, AI, and Predictive Analytics
• The US Patent & Trademark Office’s Listening Session on Patents and AI Inventorship
• The National Academies’ Workshop on AI and Machine Learning to   Accelerate Translational Research, for the Government-University-Industry Research Roundtable
• The National Academies’ Workshop on Robotics and AI

Selected Publications by Professor Robin Feldman

AI Versus IP: Rewriting Creativity

• AI shrinks the value proposition: As AI continues to embed itself throughout society, it shakes loose the foundations of what we choose to protect with IP, forcing us to reconsider how each pillar of IP derives its value—copyright, patent, trademark, and trade secret.
• Intellectual property regimes have largely assumed the centrality of humans to innovation and creativity. But, if AI systems can easily produce much of what humans invent, create, or capture through trade secrets, many human contributions to creativity may no longer satisfy the requirements of protection.

AI Versus IP describes how the legal system can mitigate the problems ahead by 1) casting the net only around the remarkable, to preserve value, and 2) restoring confidence in both AI and IP by establishing a public–private certification and benchmarking body (akin to the “Good Housekeeping Seal of Approval”).

AI and Antitrust: The Algorithm Made Me Do It

• AI can enable price collusion without direct communication among competitors.
• The market structure and private information of PBMs create a heightened risk of collusion.
• With AI algorithms, it is hard to draw the line between collusion and efficiency.

Artificial Intelligence and Cracks in the Foundation of IP

Competition at the Dawn of AI , proposing that for AI-generated works, companies should receive a shorter period of protection, enforced through the context of regulatory approval, in exchange for openness to the regulatory agency. This is modeled partially after FDA data rights for pharmaceuticals.

Highlights three potential issues with patenting AI-generated inventions:

• Timeline: A 20-year patent is an eternity for AI. When it comes to the speed of change, AI travels in an entirely different dimension.
• Transparency: Where the invention calls for a method patent—for example, a method of using an AI to determine when a car hits the brakes, or whether an applicant will receive a loan—the limited disclosure norms in software patent law may not be enough. To gain societal acceptance of AI, policy makers and the public will want someone to look under the hood.
• Collective contribution to creativity: To the extent that AI systems are deriving their creative results, in part, through the collective decisions of numerous people, can the AI’s creativity be attributable solely to the program, or its operator, or its owner?

Argues further that both in terms of the incentives given to people by listing them as the inventor of a patent, and in terms of the susceptibility to deterrence presented by rights controlled by others, it is neither socially desirable nor entirely coherent to list AI on patents.

AI Governance in the Financial Industry
with former SEC Commissioner Kara Stein

Recommends a structure for the regulation of AI built on three pieces of structural scaffolding:

• Touchpoints: where AI most tangibly interacts with the broader financial system.
• Types of evil: dividing harms inflicted by AI into the categories of the evil you planned, the evil you could have predicted, and unpredictable harms.
• Types of players: identifying actors as users, intermediaries, or creators of AI, and acknowledging that different harms may be reasonably predictable to the AI creator, for example, than to the user, or to actors in other fields.

Artificial Intelligence: The Importance of Trust and Distrust

Flowing from an Army Cyber-Institute threat casting exercise and published 18 months before COVID-19 emerged in China:

• Hypothesizes a public health emergency arising out of Asia that disrupts the U.S. health care system and creates distrust of government information. (The hypothesized scenario originates from data corruption, not from a biological virus).
• Predicts that some US sub-populations will look for other sources of information that are not uniformly reliable or of the best intentions.
• Stresses that some US sub-populations will look for other sources of information that are not uniformly reliable or of the best intentions.
• Proposes that “AI systems should be subject to review entirely outside the system itself – either industry bodies or public bodies. As an average citizen, I may never understand how a biologic interchangeable is being produced, at least not enough to trust that the drug is safe. Nevertheless, I might trust the FDA. This form of institutionalized outside review, whether by private or public entities, will be essential for adequate trust and distrust.”

Artificial Intelligence in the Health Care Space: How We Can Trust What We Cannot Know; 30 Stanford L. & Pol’y Rev. 399

Suggests that “the pathways we use to place our trust in medicine provide useful models for learning to trust AI. As we stand on the brink of the Al revolution, our challenge is to create the structures and expertise that give all of society confidence in decision-making and information integrity.”

Proposes that a government body could create and regulate standards to:

• Document a dataset’s purpose, intended use, potential misuse, and areas of ethical and legal concern.
• Provide information integrity requirements for accuracy, completeness, and archival purposes.
• Address when conflicts of interest arise between cost savings from deploying AI and quality of patient/consumer care.