AI Governance in Comparative Perspective, Theory and Practice: China, U.S. and E.U.-- Lectures at the East China University of Political Science and Law (May 2026)

 

Image generated with Google Gemini

Introduction:

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. My thanks to my hosts and especially to Sun Yuhua for organizing these opportunities to engage with colleagues. 

 

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. AI has come to dominate , or infect, depending on one’s point of view, virtually every aspect of the organization and operation of collective human systems—as well as shaping the lives of individuals who are plugged into virtual systems of decision making or automated processes that can provide everything from entertainment, to advice, to interaction with other humans and organizational structures.  This is well known as is its capacity for literal impact. One is reminded of Norbert Wiener’s reference to the classic horror tale, the Monkey’s Paw, in his book God & Golem (MIT Press, 1964, pp. 58-60). The tale describes the risk of wishing for or demanding something from a source that is entirely literal minded, which in the case of the Monkey’s Paw was meant to grant three wishes, the means to those ends were undefined. When a person wished for  a sum of money, the wish was granted, in the form of a settlement by their son’s employer as a consequence of an accident that killed the son.  In Wiener’s words: “The magic of automation, and in particular the magic of automatization in which devices learn, may be expected to be similarly literal. If you are playing  a game according to certain rules  and set the playing-machine to play for victory, you will get victory if you get anything at all, and the machine will not pay the slightest attention to any consideration except victory according to the rules.” (Ibid.); unless, of course the rules are constantly changing to embed more and more complex considerations. When the machine becomes self-learning, it can develop its own mechanisms for adding or subtracting considerations as well (rules, norms, values, etc.).   

Pix credit here

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. It is the point, as well, where such collectives are floundering even as these machine-systems become ever more sophisticated and ever more embedded in the life of human collectives and the engagement of individual humans with the world.   One moves here from "The Monkey's Paw" to "Three Thousand Years of Longing" where AI is the genie and his container is regulation, and the lonely professor given three wishes the consumers and producers of AI, that is the genie. 
 

The point of this collision, especially among human sub-systems, the cognitive orientation of which are becoming increasingly better defined and distinct as against the others, that serves as the point of departure for the series of lectures that together constitute AI Governance in Comparative Perspective, Theory and Practice: China, U.S. and E.U, With a Sideways Glance at the U.N. There are eight lectures: 

Lecture 1—From Algorithms to Foundation Models: What Contemporary AI is “Made of”
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

 The Lectures start from the premise that before it is even possible to speak about regulation and regulatory systems, and even more so to speak of such systems in or as some sort of comparison (against what standard is yet another problem of course), one must first have firmly in mind two of the critical elements on which any such discussion might be organized. The first requires  an effort to grasp, or perhaps better to approach an understanding of, the object of regulation. The second is to have some better sense of the regulatory enterprise, especially its forms, placement, approaches, characteristics and logic; and noy just its forms but its sources and the character of its authority within regulatory collectives including but not limited to the governmental apparatus.

Once one understands the complexity and interconnectivity of regulatory subjects (actors within a legal system, that might now also include AI) and objects (the things that can be acted upon, including, perhaps the regulators themselves), one can then approach the enterprise of regulation of AI in context. This starting point is not merely   an effort in taxonomy of regulatory subject and object, which then, as is traditionally undertaken, serves as a basis for the sort of matrix within which it is possible to consider those permutations and combinations of connections between subject and object, thus disaggregated, that serves some purpose or other. Each reflects the other in a certain sense.  AI is a creature made in the image of its creator; the Biblical reference as well as reference to Mary Shelly’s Frankenstein (creating something in the image of the creator) hides the computational element in both, as well as their inter-subjectivity (here in the sense of constant and mutual inter-penetration, sometimes as a form of dialectic). But all of this interactivity, and its consequential impulse to build AI in particular ways and to regulate it in equally peculiar ways, suggests an underlying fundamental set of cognitive patterns that may be reflected in both AI object and regulatory subject.  

 

Neural networks (pattern finding) and large language models (predicting sequences) may provide a useful (or at least interesting) basis for giving some form to that inter-subjectivity between regulatory subject and object—that are also object and subject. Neural networks might provide a useful conceptual basis for understanding , or at least framing, the iterative and dialectical structures within which human collectives as well as computing systems, are organized to process data (inputs, irritants, requests, etc.) through layers of weights, values, biases and activation functions in order not just to identify patterns but to use those patterns to rationalize data and respond to stimulus. In the case of political economic systems, the neural network analogy makes it easier to understand such systems as computing systems that identify and reinforce patterns, and that rationalize the environment in which they operate, through the identification and application of a series of foundational values, biases, and recognized units for receiving, processing and passing along data that together constitute the parameters within which it is possible to receive process and identify objects received . For the purposes of these lectures extracting the fundamental elements of the “hidden layers” of the political-economic “neural network” of a state system is essential for understanding the way in which the regulating body both understands the object of regulation and processes that object in terms of the biases and values of the system to determine the form and direction of regulation.

For purposes of these lectures the focus is on three of the more influential subjects of regulation: the United States (Lecture 3), the European Union (Lecture 4), and China (Lecture 6). Mere description of regulatory approaches is no longer particularly useful, since AI programs can now produce descriptive summaries that may be better than anything a human can do (controlling for the hallucinations of either). Each discussion of national legislation starts with an effort to grasp the characteristics of each state’s neural network embedded in and as its political economic model.  Those characteristics then shape the approach of each state toward the regulatory object and the character and objectives of regulatory responses.  In that context, the United States might be as the “markets state,” the Europea Union as the “rights state,” and China as the “guided state.” These labels suggest the baseline weights and values, the scope of what each system identifies and processes, how it perceives value and threat and accordingly regulates. The US is markets driven and suspicious of state interference (for the most part though reshaped by reconstituting efforts after the 1920s and its openness to techno bureaucratic leadership). The European Union may be rights driven and trusting of state leadership the purpose of which might be understood to protect and enhance rights and prevent negative impacts. China is development driven with a specific objective (forward movement along a Socialist Path) guided and led by a vanguard of social forces in the context of which all productive forces of the state might be embedded. These “hidden layers” of the national neural networks then shape the way each formulates the “issue” of AI and fashions a response that both reaffirms the patterning of its foundational structures and in the process reinforces them. 

 

Lectures 6 and 7 then consider  two critical areas that have a significant effect on any regulatory project. The first focuses on courts, enterprises, and the legal construction of AI in and as law. Its principle focus is on the way that AI related litigation is both changing traditional areas of jurisprudence (tort, contract, property and the like) and is being changed by it. Lecture 8 then considers the way in which leaders of AI production are seeking to influence approaches to the conceptualization of the relationship between AI and the state, and in the process offer a window into the way that AI leaders reenvision the relationship between AI, the State and the regulation of both.

Lecture 8 then attempts to draw the bigger picture. It suggests the way that a shared vocabulary can cover quite substantial differences in approach, values, and objectives in the context of AI and AI regulation. And it suggests the way that weighting the objectives of innovation, risk, and the role of the state affects  the way in which  common regulatory focus of safety, security, transparency, accountability, innovation, and infrastructure, each with sometimes substantially different regulatory characteristics as a function of the core differences in national “neural networks.”  One comes at last to the understanding of the fundamental contradiction of AI and its regulation: the development of a common language  across political systems the neural networks of which overlap but are fundamentally incompatible producing conceptions of AI and law that can be coordinated but which cannot converge. That seems to be the template for coordinated governance for the 2nd quarter of the 21st century. I noted the effect in the context of the convergence project of the  International Financial Reporting Standards (IFRS) Foundation has released its IFRS Foundation 2025 Annual Report—Fit for the Future (31 March 2026):.

By the start of the second quarter of the 21st century, however, the common language, like written Chinese, covered a wide variation in how it would be "spoken" and the signification with which common use terms would be infused. take transparency, for example. Within the cognitive cages of early 21st century convergence globalization transparency could be read broadly to require disclosure of virtually everything but trade secrets and protected know how--and even that might require some disclosure as a function, for example of its connection with adverse human rights or sustainability impacts (as these terms were to be developed by international institutions through hard and soft measures, norms, declarations, finds, etc.)--subject to increasingly narrowed exceptions for national security, development and the like. A quarter century later, transparency remains a presumption but now much more tightly constrained and perhaps sieved through superior obligations (many generated through domestic legal orders of States) respecting, national security, sanctions, data protection and data sovereignty, anti-espionage statutes, and blocking legislation to counter extraterritorial projections of regulatory systems. Transparency remains the same; its signification is now fractured and defined as a function of higher ranking signified principles, rules, and expectations. The same applies, with substantially more bite, to the construction and deployment of the concept of "risk."
One can then, and with enthusiasm, embrace a common language, while at the same time investing that language with regional or other contextual meaning, and constrain it through higher order local values, especially those encased in domestic (or regional) rules and legislation. The imaginaries of unity are preserved, and, indeed, a platform within which such values and contestations of values, may be produced and consumed, without enforcing a centralizing fundamental political or normative line. Yet that also is a step forward. A single language encasing difference may not have been the goal at the start of the 21st century, but language binds all the same--it binds enough that difference may be constrained by the outer boundaries of rational learning inside a cognitive cage that through its language objects effectively defines both that rationality and its limits. And it is that common language that provides all that is necessary to achieve the fundamentally critical element of this project: comparability and communication across and despite of difference. The rest might come later; or it might not. That convergence is no longer necessary where a common language has been substituted for common norms, beliefs, practices, and expectations. (A Common Language Containing Differentiating Meanings Within Evolving International Standards for Sustainability Disclosure in Financial Statements).

In the posts that follow I will post summaries of the lecture notes and the PowerPoint of each of the lectures which can be accessed by following these links:

 Lecture 1—From Algorithms to Foundation Models: What Contemporary AI is “Made of”
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