 |
| Pix Credit Image created using DALL-E (visual representation of the relationship between state and individuals through markets |
I was delighted to have had the opportunity to present a series of Lectures hosted by the East China University of Political Science and Law (ECUPL) at the end of May.
The overall theme (and thus the title) of the lectures was AI Governance in Comparative Perspective, Theory and Practice: China, U.S. and E.U, With a Sideways Glance at the U.N. The subject of the lectures requires little by way of introduction: Artificial intelligence is the broad term that has come to represent a growing cluster of non-human and digitalized processes and operations that has as its primary task the constitution of non-human systems capable of performing tasks that were once thought to require human intelligence. And so is the impulse to manage, control, exploit, embed, understand, and regulate these processes, systems, and perhaps eventually non-human consciousness with a huge potential to undertake many of the computational tasks (the mathematical and logical processing of data) that were once the sole domain of and perhaps defined what it meant to be human. That is the point where things get interesting. It is at the point where the development of machines, that is of non-human systems, capable of performing tasks that were once thought to require human intelligence, collide with regulatory structures meant to manage, contain, constrain, liberate, embed, project and exploit such non-human systems, whether they are traditional or emerging, public or private regulatory systems, that human collectives and the machine-systems they have created now find themselves.
The eight lectures progress sequentially from conceptual and theoretical frameworks (lectures 1 and 2, the objects and subjects of AI regulation), through a deeper consideration of regulatory systems in three distinguishable regulatory regimes--the US, EU, and China (Lectures 3, 4.5). The last two lectures consider judicial efforts to embed AI within traditional legal orders (Lecture 6), and the way in which the object of regulation (in the form of the owners of the larger AI enterprises) understand the relationship between AI, the state, and society (Lecture 7) . Lecture 8 summarizes and draws larger themes going forward.
In a previous post introducing Lecture 1 (From Algorithms to Foundation Models: What Contemporary AI is “Made of”) I suggested that perhaps a useful way of approaching the issue of AI regulation is to start by considering the nature and characteristics of the regulatory subject--what we euphemistically refer to as "AI." It then occurred to me that it might be useful as well to see if that regulatory object had views of their own respecting their nature character and, more importantly, the relationship of regulation projects to that (self) perception of their nature and character. So I approached Google's Gemini with a series of questions which I thought, in the process of what might pass for a conversation, might help humans begin to understand how at least one AI program thinks of itself. That conversation was incorporated into Lecture 1A. In Lecture 2 we moved from the object to the subjects of regurgitation. Like its regulatory objects, regulatory subjects are functionally differentiated and can be disaggregated. In either case the connection between object and subject becomes complicated.
This post includes a summary of the Lecture 3 Notes, as well as the Lecture 3 PPT. Those interested may reach out to me to discuss availability of audio of the lecture and the full text of the Lecture 1 notes
Given the nature of the project I thought it might be useful to engage with an commercially available AI service for the production of a summary of the Lecture 1 materials. After some back and forth with Anthropic's Calude (Lecture 2 used Chat GPT; Lecture 1 and 1A used Google's Gemini), we came up with the following abstract of Lecture 3.
 |
| Image generated using Claude |
ABSTRACT:
The United States and the Governance of Artificial Intelligence: Markets, Strategy, and Fragmentation . The United States governs artificial intelligence not through a single comprehensive statute but through a distributed institutional ecosystem in which markets, agencies, courts, procurement contracts, voluntary standards, state legislatures, and national security apparatus each occupy a portion of the regulatory terrain. This arrangement reflects the country's broader political economy and carries distinctive characteristics that merit careful analysis across four principal themes.
The first theme is the primacy of innovation over precautionary authorization. The American regulatory framework proceeds from a foundational premise: technology enters the market unless existing law provides a basis for intervention. No federal agency pre-approves most AI systems before deployment. Enforcement follows from harm rather than preceding it. The Federal Trade Commission addresses deceptive claims after they appear in the market. Employment regulators examine discriminatory hiring tools through existing anti-discrimination frameworks. Financial regulators apply model risk and fair lending requirements to credit systems. The Food and Drug Administration governs AI-enabled medical devices through established product-safety authority. This ex post orientation is not incidental to the American approach but constitutive of it, reflecting a settled preference for private innovation as the primary driver of technological development.
The second theme is the translation of AI-related conduct into established legal categories. Rather than constructing AI as a legally novel object requiring dedicated statutory architecture, the United States routes AI problems through existing normative vocabularies. Algorithmic deception is addressed as consumer fraud. Automated hiring discrimination is addressed as employment discrimination. AI-assisted misrepresentation to investors is addressed as securities fraud. AI components in medical devices are addressed as product safety questions. This translation strategy allows regulators to act under existing authority without awaiting new legislation, and permits sector-specific agencies to apply domain expertise to domain-specific problems. The approach also raises analytical questions about the fit between legal categories developed before modern AI systems and the characteristics of probabilistic, opaque, and continuously updated models operating across complex institutional arrangements.
The third theme is the strategic-industrial dimension of American AI governance. The United States does not govern AI solely through market mechanisms. The Trump Administration's policy documents—including the America's AI Action Plan, Director Kratsios' addresses at the Endless Frontiers Retreat and the APEC Digital and AI Ministerial Meeting, and the March 2026 National AI Legislative Framework—articulate a model in which AI is treated as a foundational technology in geopolitical competition. State activity in this register includes export controls on advanced semiconductors, domestic infrastructure investment, bilateral technology agreements, and active promotion of American AI systems and governance frameworks internationally. The administration's policy language frames AI development as inseparable from national security, economic competitiveness, and what it describes as civilizational contest with strategic rivals. This dimension complicates any characterization of the American model as straightforwardly market-liberal, as it involves substantial state coordination, funding, procurement, and strategic direction.
The fourth theme is federalist fragmentation as a structural feature of American AI governance. The absence of comprehensive federal AI legislation means that governance emerges from the interaction of multiple institutional actors operating under different authorities and at different levels of government. State legislatures have addressed automated decision-making, biometric data, employment AI, and high-risk systems, with Colorado's risk-based statute representing one of the more comprehensive state-level frameworks. Voluntary standards, most notably the NIST AI Risk Management Framework, acquire practical authority through procurement requirements, liability considerations, and industry adoption rather than through direct legislative mandate. Private firms establish their own model policies, access rules, and safety standards. The resulting architecture is one in which regulatory coverage, enforcement capacity, and available remedies vary across sectors, jurisdictions, and affected populations.
Taken together, these four themes describe a governance model that combines monitored markets, existing-law enforcement, voluntary standardization, procurement discipline, state-level experimentation, and national security strategy into an arrangement that is institutionally plural, sectorally differentiated, and geopolitically oriented. The central analytical question the model poses is whether this distributed architecture is adequate to govern AI systems whose characteristics—scalability, opacity, systemic reach, and geopolitical significance—extend beyond the assumptions embedded in the legal and institutional frameworks through which it currently operates.
Links to Lectures:
Lecture 0 -- IntroductionLecture 1—From Algorithms to Foundation Models: What Contemporary AI is “Made of”
Lecture 1A--A Computation/Conversation With Google's "Maschinenmensch" Gemini:
Lecture 2—What Are We Actually Governing When We Govern AI?
Lecture 3—The “Markets State”: U.S. Approach
Lecture 4—The “Rights State”: EU Approach
Lecture 5—The “Guided State”: The Chinese Approach
Lecture 6—Courts, Companies, and the Legal Construction of AI
Lecture 7—AI Narratives: Palantir; Anthropic; Open AI; and Leopold Aschenbrenner
Lecture 8—Putting It All Together: Trends, Trend Lines, and Regulatory Dialectics
Pix credit here
LECTURE · SUMMARY NOTES
The United States Approach to the Governance of Artificial Intelligence: Monitored Markets, Strategic Statecraft, and Institutional Fragmentation
I. The Governing Concept: Monitored Market Governance
The United States does not govern artificial intelligence through a single comprehensive national statute. It governs AI through a distributed and institutionally plural ecosystem composed of markets, voluntary standards, federal procurement, executive policy, agency enforcement, sector-specific regulation, state law, litigation, national security controls, and industrial strategy. The lecture's foundational claim is that this arrangement is not the absence of governance but a particular form of it, one that reflects the country's broader political economy and produces characteristic capabilities and limitations.
The organizing concept is monitored market governance. Innovation is assigned primarily to private actors: firms, capital markets, universities, platforms, laboratories, and defense-linked research institutions. The federal government encourages innovation, funds research, procures AI systems, issues guidance, develops voluntary standards, applies existing law, and intervenes through agencies when harms emerge. States fill gaps through their own legislative and regulatory activity. National security institutions govern chips, exports, cloud infrastructure, and strategic competition with foreign powers. The result is a system in which AI is constructed primarily as an innovation market and strategic asset, monitored through existing legal tools rather than enclosed within a unified regulatory code.
This arrangement produces speed, flexibility, and private-sector dynamism. It also produces fragmentation, uncertainty, reactive enforcement, and variation in the protection available to persons affected by AI systems. Both dimensions are treated in the lecture as constitutive features of the model rather than incidental characteristics to be remedied.
II. Innovation First, Enforcement After: The Ex Post Orientation
The dominant regulatory question in the United States is not how to authorize AI systems before they enter the market but how to preserve innovation while managing harms through existing legal tools. This ex post orientation is foundational. With narrow exceptions in regulated sectors such as medical devices, financial services, aviation, and defense, a company may release an AI product without prior federal approval. Legal response follows harm rather than preceding it.
The principal enforcement actors operate under existing statutory authority. The Federal Trade Commission addresses deceptive AI claims and unfair practices. The Equal Employment Opportunity Commission addresses AI hiring tools that produce unlawful discrimination. Banking regulators address model risk, fair lending, and adverse action requirements. The Food and Drug Administration regulates AI-enabled medical devices as product-safety matters. The Securities and Exchange Commission addresses AI-washing, the practice of misleading investors about AI capabilities or strategy. State attorneys general enforce consumer protection statutes and applicable state AI laws. Private plaintiffs bring claims under anti-discrimination law, product liability, negligence, privacy statutes, copyright law, and contract law.
The translation of AI-related conduct into existing legal categories is one of the lecture's central analytical points. The United States declines to treat AI as a legally novel object exempt from established normative frameworks. If AI is used to defraud, it is fraud. If AI is used to discriminate, it is discrimination. If AI is used to mislead investors, it is securities fraud. If AI is incorporated into an unsafe medical device, it is a medical device problem. The legal system routes AI problems through existing categories rather than constructing a new categorical architecture for AI as such.
This approach carries recognized advantages: regulators can act without awaiting new legislation, established legal principles are preserved, sector-specific expertise is brought to bear on sector-specific problems, and firms cannot claim exemption from law simply because their product uses AI. The approach also carries recognized limitations: legal categories developed before modern AI systems may not map cleanly onto probabilistic, opaque, continuously updated, multi-actor systems; affected persons may not know AI was involved in a decision; proof of causation and discrimination may be difficult; agency technical capacity may be constrained; enforcement may be uneven; and sectoral gaps may leave certain systemic harms unaddressed.
III. Voluntary Standards and the NIST Framework
The National Institute of Standards and Technology occupies a significant position in the U.S. approach through its AI Risk Management Framework. The framework is voluntary. It does not impose legal obligations on AI providers or deployers in the manner of the EU AI Act. It offers instead a vocabulary and process for identifying, measuring, managing, and governing AI risks, organized around trustworthy AI characteristics including validity, reliability, safety, security, resilience, accountability, transparency, explainability, interpretability, privacy, and fairness.
Voluntary standards acquire practical authority through indirect mechanisms. Federal procurement may require adherence to the framework. Courts may treat compliance or noncompliance as evidence bearing on reasonable care. Insurers, auditors, customers, and industry groups may incorporate the framework into their expectations and requirements. State laws may reference it. Over time, a voluntary framework can function as a practical compliance baseline even without direct legal mandate.
This dynamic illustrates a broader feature of U.S. governance: law does not always operate through direct command. It operates through standards, incentives, contracts, procurement conditions, liability exposure, and reputational pressure. The institutional ecosystem outside formal legislation carries substantial governance weight. The adaptability of this approach allows standards to evolve more rapidly than statutes. The corresponding limitation is uneven adoption and the absence of automatic rights or guaranteed transparency for persons affected by AI systems.
IV. Federal Executive Policy and the Strategic-Industrial Frame
U.S. AI governance is shaped substantially by executive policy, and the lecture treats the strategic-industrial dimension of that policy as a theme of independent analytical importance. The 2025 America's AI Action Plan articulated more than ninety federal policy actions organized around accelerating innovation, building AI infrastructure, and leading in international diplomacy and security. It framed AI as central to economic competitiveness and national security. Accompanying policy direction emphasized U.S. leadership, domestic infrastructure, export of the American AI technology stack, and international adoption of American AI technologies and governance models.
This strategic frame is elaborated through two addresses by Michael Kratsios, Director of the White House Office of Science and Technology Policy. The first, delivered at the Endless Frontiers Retreat in April 2025 under the title "The Golden Age of American Innovation," is analyzed as articulating the Administration's theory of governance, modernization, and geopolitical competition. The speech invokes President Roosevelt's wartime directive to science advisor Vannevar Bush as a historical model, positioning the current moment as one of national renewal after a period of deviation. The Administration's model is described as one in which markets remain operationally central but the state strategically directs technological development, allocates research and development funding, constructs a pro-innovation regulatory environment, and promotes the adoption and export of American technologies. The lecture notes that this structure—state direction of national development through mobilization of private actors toward defined political objectives—bears structural resemblance to state-guided modernization approaches in other major powers, while differing in its ideological content and institutional form.
The second address, delivered at the APEC Digital and AI Ministerial Meeting in August 2025, concerns the export of American AI systems. Kratsios frames AI exports not merely as commercial transactions but as instruments for building durable strategic relationships between states. The proposed American AI Exports Program would package a full U.S. AI stack—infrastructure, software, financing, and support services—for export to partner countries, using federal financial instruments including loans, guarantees, and equity investments. The lecture's analysis of this address emphasizes that AI systems carry embedded normative assumptions about privacy, property rights, governance, and social organization, and that the export of AI infrastructure is simultaneously the projection of a particular normative order. Global AI competition is understood as simultaneously technological, economic, and civilizational, with the United States, the European Union, and China offering rival techno-political ecosystems with different embedded assumptions and dependencies.
President Trump's March 2026 National AI Legislative Framework extended the executive policy framework into a set of objectives directed at Congress, organized around six themes: protecting children and empowering parents in digital environments; strengthening communities and small businesses through AI-enabled economic growth; respecting intellectual property rights of creators and innovators; preventing censorship and protecting free speech against AI-enabled suppression of lawful expression; enabling innovation and ensuring American AI dominance through removal of barriers and expansion of testing environments; and developing an AI-ready workforce through training and education programs.
V. Procurement, Litigation, and Private Governance
Three additional institutional mechanisms contribute to the governance architecture in ways the lecture identifies as underappreciated or structurally significant.
Procurement converts the state from regulator into buyer. Government agencies at federal, state, and local levels purchase AI systems, and procurement contracts can require vendors to meet standards, disclose information, document risks, allow audits, protect data, ensure cybersecurity, and provide for human oversight. In contexts where government uses AI to allocate benefits, detect fraud, prioritize inspections, assist policing, or manage immigration, the procurement contract may be the primary instrument determining transparency, auditability, appeal mechanisms, and vendor accountability. The limitations of procurement governance include the opacity of contractual arrangements, agency technical constraints, trade secret claims by vendors, and the difficulty of removing embedded systems once deployed.
Litigation functions as a governance mechanism through discovery, precedent, settlement pressure, damages, and deterrence. It provides affected persons a potential path to remedy under anti-discrimination law, consumer protection law, securities law, product liability, negligence, privacy statutes, copyright, contract, and constitutional claims when public actors are involved. The recognized limitations include cost, delay, evidentiary barriers, arbitration clauses that may foreclose class litigation, and the difficulty courts face in analyzing complex probabilistic systems.
Private governance constitutes a distinctive and substantial feature of the U.S. model. Major AI firms set model policies, access rules, content filters, safety standards, API restrictions, and release practices. Platforms govern recommender systems, content moderation, and synthetic media policies. Cloud providers determine infrastructure access. Enterprise customers impose contractual protections. The significance of private governance follows directly from the fact that private firms build and operate the predominant share of AI infrastructure. The lecture notes the legitimacy questions this raises: corporate policies are not democratically enacted, may be opaque, may change rapidly, and reflect business interests that may or may not align with broader public interests. The concentration of control over foundation models, cloud infrastructure, and distribution platforms in a small number of firms raises market-structure questions that governance analysis cannot set aside. The April-May 2026 episode in which industry leaders successfully lobbied against a proposed Executive Order that would have required developers to provide government access to advanced AI models up to ninety days before release illustrates the practical significance of state-enterprise interactions in shaping both policy and the direction of AI development.
VI. State Law, Federalist Fragmentation, and the Patchwork Dynamic
The absence of comprehensive federal AI legislation has made state-level activity increasingly significant. State legislatures and regulators have addressed automated decision-making, biometric privacy, deepfakes, employment AI, consumer protection, insurance algorithms, healthcare AI, children's online safety, political synthetic media, and high-risk AI systems. Colorado's AI statute, which applies to high-risk AI systems and requires developers and deployers to use reasonable care to protect consumers from algorithmic discrimination, including obligations related to documentation, impact assessments, notices, risk management, and correction opportunities, represents one of the more architecturally developed state-level frameworks. California's Governor issued a May 2026 Executive Order directing state agencies to study severance standards, employment insurance, worker ownership rules, and updates to workplace layoff notice requirements in response to AI-driven labor market disruption, illustrating the use of executive guidance at the state level to address AI's economic effects.
State-level activity creates opportunities for jurisdictional experimentation and may provide protections when federal action is slow. It also creates compliance complexity for national AI providers subject to varying rules across jurisdictions. The resulting dynamic is federalist, fragmented, and contested, with AI governance emerging through ongoing interactions among states, federal agencies, Congress, courts, firms, and advocacy organizations.
VII. Strengths, Limitations, and the Comparative Position
The lecture's assessment of the U.S. model identifies its principal strengths as the capacity for rapid innovation, the availability of world-leading research and capital infrastructure, the adaptability of the institutional ecosystem, the ability to deploy sector-specific expertise, and the capacity to act through existing law without waiting for a comprehensive AI definition. Its principal limitations are fragmentation across agencies and jurisdictions, the reactive character of enforcement, the potential for harms to occur before legal response, variation in protections across affected populations, and the capacity of voluntary standards to be ignored by actors who choose not to adopt them. The system may be effective against clearly defined individual harms such as deception or identifiable discrimination while being less well-suited to address diffuse systemic harms involving platform manipulation, labor surveillance, or concentration of computational resources.
The U.S. model is positioned in the lecture series as a contrast to the European Union approach examined in the following lecture. Where the EU attempts to classify AI systems in advance through a comprehensive risk-based statute and impose duties across the deployment lifecycle connected to safety and fundamental rights, the United States routes AI governance through existing legal channels and relies on markets, agencies, courts, standards, and procurement to discipline AI conduct after it occurs. The central analytical question the lecture poses is whether this distributed architecture is adequate to govern AI systems whose characteristics—scalability, opacity, systemic reach, and geopolitical significance—extend beyond the assumptions embedded in the legal and institutional frameworks through which governance currently operates.
No comments:
Post a Comment