Governance in an Institutional Environment Reconfigured by Artificial Intelligence. A Public Choice Institutionalism Perspective

When we think about governing artificial intelligence, the conversation often focuses on the machine itself. We ask how to build it safely, how to make sure its code does not malfunction, and how to ensure its decisions align with human values. This is like checking the engine of a car to make sure it runs smoothly. But there is a much larger, more profound question that often gets overlooked. It is not just about the machine; it is about the world the machine is changing. As artificial intelligence spreads, it does not just sit alongside our existing laws and organizations; it actively reshapes them. It forces us to rethink how we make rules, how we assign responsibility, and how we organize society itself. The challenge is no longer just about fixing a tool; it is about navigating a landscape that is shifting beneath our feet, where the very nature of work, authority, and decision-making is being rewritten in real time.

This is the central puzzle tackled by Paul Dragos Aligica, a researcher at George Mason University. He argues that to understand this new reality, we must stop looking at artificial intelligence as a single problem to be solved by a single authority. Instead, he suggests we view it as a complex ecosystem where many different groups, rules, and decisions interact. The paper proposes that the best way to manage this shifting world is through a concept called "polycentricity." In plain terms, this means a system where many centers of power exist side by side, each making its own rules and decisions, rather than one single boss or government trying to control everything from the top down. Aligica builds this argument by combining two powerful ideas: the study of how people make choices when they have different interests, and the study of how complex systems organize themselves without a central commander.

The paper begins by drawing a sharp line between two different kinds of problems. The first is the "narrow" problem: how to govern the artificial intelligence artifact itself. This involves the technical work of programming, monitoring, and ensuring the system behaves as intended. The second is the "broad" problem: how to govern the entire institutional environment that the technology is transforming. Aligica insists that the broad problem is the one that truly matters. When artificial intelligence enters a society, it triggers a chain reaction. It changes the division of labor, meaning some tasks that humans used to do are now done by machines, while humans take on new, different roles. This shift is not a simple swap; it creates a cascade of effects that ripples upward through the entire system of rules.

To understand how this ripple effect works, the paper uses a logic similar to how economists study the division of labor. Imagine a factory where a new machine is introduced. At first, the machine takes over the repetitive, high-volume tasks. But this forces the human workers to change what they do. They might stop doing the routine work and start focusing on the exceptions, the difficult cases, or the oversight of the machine. This change in human roles then creates a need for new rules, new supervisors, and new ways to handle disputes. The paper describes this as a "cascade." A single decision to use artificial intelligence for one task can eventually force a rewrite of the highest-level rules that govern how society makes decisions. It moves from the day-to-day operations of a specific task, up to the rules that decide who has the power to make those tasks, and finally to the constitutional level that decides who has the right to change the rules themselves.

As this cascade moves through the system, it activates familiar human behaviors that political scientists have studied for decades. When new roles and powers are created, people and organizations will try to grab them for their own benefit. This leads to "rent-seeking," where groups spend energy trying to capture the benefits of a new rule rather than creating value. It leads to "regulatory capture," where the people meant to be in charge of the technology end up relying on the very companies they are supposed to regulate because those companies hold the only expertise. It also creates "agency loss," where the people who are supposed to be in charge cannot effectively monitor the complex, opaque decisions made by the machines or the new bureaucracies built to manage them. The paper argues that these are not accidents; they are inevitable consequences of how human incentives work when the structure of society changes.

Given this complexity, the paper evaluates two main visions for how to govern this new world. The first is the "monocentric" vision. This is the idea of a single, central authority—a global regulator or a unified treaty—that sets one set of rules for everyone. It is an appealing idea because it promises order, clear accountability, and a single place to point a finger if things go wrong. However, Aligica argues that this approach is both unrealistic and dangerous. It assumes that one center can know enough to manage a technology that is changing faster than any single government can react. It also concentrates all the power to capture, meaning that the most powerful companies or nations could easily hijack the entire system to serve their own interests. Furthermore, a single set of rules cannot possibly fit the deep differences in values and needs across different cultures and societies.

The alternative, and the one the paper champions, is the "polycentric" vision. This is not a chaotic free-for-all, but a structured system where many different decision centers operate simultaneously. In this system, different cities, regions, companies, and organizations can try different approaches to governing artificial intelligence. Some might prioritize speed, others safety, and others privacy. These different centers compete and learn from one another. If one approach fails, it can be abandoned without bringing down the whole system. If one approach works well, others can adopt it. Crucially, this system relies on "interoperability." This means that even if different centers have different rules, they can still talk to each other and work together through shared standards and agreements. It is like a network of independent towns that trade with each other and respect each other's laws without needing a single king to rule them all.

The paper makes a strong case that this polycentric approach is not just a theoretical ideal, but the reality that is already emerging. The world is not moving toward a single global government for artificial intelligence; it is already moving toward a patchwork of different rules and standards. Trying to force a single, top-down solution onto this reality would be a mistake. Instead, the goal should be to steer this evolving process. The paper emphasizes that we cannot simply design a perfect system from scratch and install it. The system is too complex and the changes are too fast. Instead, we must act as gardeners or navigators. We can create the conditions that allow good solutions to grow and bad ones to be weeded out. We can build the bridges that allow different groups to cooperate, and we can protect the space where new ideas can be tested.

Ultimately, the paper concludes that the best way to govern artificial intelligence is to accept that we are in the middle of a long, unfolding process. We cannot predict the final outcome, nor should we try to control every detail. Instead, we should focus on building a flexible framework that allows for experimentation, learning, and adaptation. By embracing a polycentric approach, we can harness the diversity of human thought and the competitive pressure of different jurisdictions to find solutions that work. We can manage the risks of rent-seeking and capture by ensuring that no single group holds all the power. And we can maintain the legitimacy of our rules by ensuring that they are shaped by the people they affect, rather than imposed from a distant center. The path forward is not to build a perfect machine to control the future, but to cultivate a resilient system that can learn and adapt as the future unfolds.