Rule-based Geopolitics and Large Language Models

The Post-Rules-Based World Order: A Mirror of AI’s Transition from Deterministic Systems to Large Language Models

Health Warning: This article has been written by an AI and reviewed/edited by a real human. 

The contemporary international system stands at a remarkable inflection point, witnessing the gradual erosion of the post-World War II rules-based order that has governed global affairs for nearly eight decades. This transformation bears striking parallels to a concurrent technological revolution: the transition from traditional rule-based automation systems to Large Language Model (LLM)-enabled artificial intelligence. Both transitions represent fundamental shifts from deterministic, explicitly programmed frameworks to more dynamic, probabilistic, and adaptive systems that challenge established paradigms of control and predictability.

The Architecture of Rules-Based Systems

The Post-1945 International Framework

The rules-based international order emerged from the devastation of World War II as an ambitious attempt to create a predictable global system governed by shared rules, norms, and institutions. Central to this vision was the establishment of multilateral institutions such as the United Nations, the International Monetary Fund, and the World Bank, alongside widely accepted principles of sovereignty and self-determination. This system was fundamentally deterministic in nature, operating on explicit rules and procedures designed to channel state behavior into predictable patterns and prevent the chaos that had characterized earlier periods of great power competition.

The UN Charter, as the foundational document of this order, established clear principles including the sovereign equality of states, the prohibition of aggression, and mechanisms for peaceful dispute resolution. Like a sophisticated rule-based system, the post-war order relied on explicitly codified principles that were meant to govern international interactions through predetermined responses to specific conditions and triggers.

Traditional Rule-Based Automation Systems

In parallel, the technological realm developed rule-based automation systems that similarly operated on explicit, predetermined logic structures. These systems, which reached their zenith with expert systems in the 1980s and 1990s, functioned through “if-then” conditional statements that provided deterministic responses to specific inputs. A rule-based system applies human-made rules to store, sort, and manipulate data, mimicking human intelligence through predefined logical pathways.

The architecture of these systems consisted of several key components: a knowledge base containing explicit rules, an inference engine that processed these rules against input data, and a working memory that maintained current facts and variables. Like the international rules-based order, these systems prioritized predictability, transparency, and consistency in their operations.

The Erosion of Deterministic Frameworks

Challenges to the International Rules-Based Order

By 2019, it had become clear that the liberal international order was experiencing fundamental strain, containing what scholars identified as “the seeds of its own destruction”. The system faced multiple challenges that mirror the limitations encountered by rule-based automation systems. First, the rise of multipolarity has fundamentally altered the distribution of power, with China, Russia, and other emerging powers challenging the Western-dominated framework.

The rules-based order increasingly struggles with what can be termed “rule brittleness” – the inability to adapt to novel situations not explicitly covered by existing frameworks. Critics argue that the system’s rules are often vague and selectively applied, leading to accusations of double standards and undermining legitimacy. Russia and China have actively contested the Western interpretation of the “rules-based order,” arguing for alternative frameworks based on different principles and power distributions.

Contemporary challenges such as the war in Ukraine, conflicts in the Middle East, and rising authoritarianism demonstrate the system’s difficulty in managing complex, ambiguous situations that don’t fit neatly into predetermined categories. The UN Security Council, designed as the system’s primary enforcement mechanism, has become increasingly paralyzed by the irreconcilable positions of its permanent members.

Limitations of Rule-Based AI Systems

Traditional rule-based AI systems encountered remarkably similar limitations that led to their decline and the emergence of machine learning approaches. These systems suffered from what experts call “knowledge acquisition bottlenecks” – the difficulty of capturing and codifying all relevant expertise into explicit rules. As domains became more complex, the number of required rules grew exponentially, leading to contradictions, overlaps, and maintenance difficulties.

Rule-based systems proved inflexible when confronting novel situations not covered by their original programming. They lacked learning capabilities and could not adapt to changing environments or incorporate new information dynamically. The “AI winters” of the 1970s and 1980s were partly caused by the recognition that symbolic, rule-based approaches could not handle the complexity and ambiguity of real-world problems.

The Emergence of Probabilistic Alternatives

Multipolarity and System Fragmentation

The transition away from unipolarity has created what scholars describe as a move toward a “multi-order world” rather than simply a multipolar one. Instead of a single set of rules governing global interactions, we see the emergence of competing normative frameworks and institutional arrangements. China and Russia promote alternative visions of international order based on different principles of sovereignty, governance, and great power relations.

This fragmentation resembles the shift from deterministic to probabilistic systems in technology. Rather than operating according to fixed rules that produce predictable outcomes, the emerging international system exhibits characteristics of probabilistic decision-making, where similar inputs may yield different outputs depending on context, power dynamics, and strategic considerations.

Regional powers are increasingly asserting their own interpretations of international law and norms, creating what some scholars call “bounded orders” around different great powers. This mirrors the way modern AI systems operate through multiple, sometimes competing models rather than a single, centralized rule set.

The Rise of LLM-Enabled Automation

The technological transition from rule-based systems to LLM-enabled automation represents a fundamental paradigm shift from deterministic to probabilistic systems. Large Language Models operate on entirely different principles than their rule-based predecessors, using statistical patterns learned from vast datasets rather than explicit programming. This transition embodies what AI engineers describe as moving from systems that “know exactly what will happen when a condition is met” to systems where “certainty gives way to likelihood”.

LLM-enabled automation systems can handle ambiguity, context-dependence, and novel situations in ways that rule-based systems never could. They employ what researchers call “agentic workflows,” where AI agents can make autonomous decisions, learn from interactions, and adapt their responses based on changing circumstances. Unlike rule-based systems that require explicit programming for every contingency, LLMs can generate contextually appropriate responses to situations they have never explicitly encountered.

Parallel Challenges and Opportunities

Transparency and Accountability

Both the post-rules-based international order and LLM-enabled systems face significant challenges regarding transparency and accountability. In international relations, the shift away from explicit, codified rules toward more fluid, power-based arrangements raises concerns about predictability and legitimacy. Smaller states worry about losing the protections offered by universal principles in favor of arrangements dominated by great powers.

Similarly, LLM systems operate as “black boxes” where decision-making processes are not easily interpretable or auditable. Unlike rule-based systems where every decision could be traced back to specific rules, LLMs make decisions through complex neural network processes that resist easy explanation. This creates challenges for applications requiring transparency, such as healthcare, finance, and legal systems.

Adaptability versus Stability

The transition in both domains reflects a fundamental trade-off between adaptability and stability. Rule-based systems, whether international or technological, provided stability and predictability but struggled with novel challenges. The emerging alternatives offer greater flexibility and responsiveness but at the cost of reduced predictability.

In international relations, the shift toward multipolarity and competing orders may allow for more responsive governance arrangements that better reflect contemporary power realities and cultural diversity. However, this comes with increased risks of conflict, miscalculation, and coordination failures.

LLM-enabled systems similarly offer unprecedented adaptability and can handle complex, ambiguous tasks that would be impossible for rule-based systems. However, their probabilistic nature introduces new forms of uncertainty and potential failure modes that organizations must learn to manage.

Hybrid Approaches and Future Directions

Integration Strategies

Just as AI systems are exploring hybrid approaches that combine rule-based constraints with machine learning capabilities, international governance is likely to evolve toward frameworks that blend universal principles with regional flexibility. Some experts propose “rule reinforcement learning” approaches in AI that train LLMs to operate within explicit policy constraints while maintaining their adaptive capabilities.

In international relations, proposals for reforming global governance similarly seek to combine the stability of universal principles with the flexibility needed to accommodate diverse perspectives and changing power distributions. The Better Order Project, for instance, proposes enhanced norms and laws to rejuvenate an inclusive global order that maintains core principles while allowing for more equitable participation.

Learning and Evolution

Both emerging systems emphasize continuous learning and evolution rather than static programming. LLM-enabled automation systems can learn from new data and adapt their behavior over time. Similarly, the emerging international order may be characterized by ongoing negotiation and adaptation of norms rather than fixed rules.

This shift toward learning systems creates new requirements for governance and oversight. In AI, this has led to the development of new frameworks for managing model behavior, including techniques for alignment, safety, and controllability. International relations may require similar innovations in multilateral governance to manage the dynamics of competing powers and evolving norms.

Conclusion

The parallel transitions from rules-based international order to multipolarity and from deterministic automation to LLM-enabled AI reflect broader shifts in how complex systems can be organized and governed in an interconnected world. Both transitions represent movements away from rigid, explicitly programmed frameworks toward more adaptive, context-sensitive approaches that can handle ambiguity and novel challenges.

These changes are not merely technical adjustments but fundamental paradigm shifts that challenge basic assumptions about predictability, control, and governance. While they offer unprecedented capabilities for handling complexity and change, they also introduce new forms of uncertainty and risk that require innovative approaches to management and oversight.

Understanding these parallels provides valuable insights for navigating both technological development and international relations in an era of rapid change. As both domains continue to evolve, the lessons learned from managing one transition may inform approaches to the other, highlighting the interconnected nature of technological and political transformation in the 21st century.

The future likely belongs to hybrid systems that can combine the best aspects of both approaches: the stability and transparency of rules-based frameworks with the adaptability and intelligence of probabilistic systems. Success in both domains will require developing new forms of governance, oversight, and coordination that can harness the benefits of these powerful new capabilities while managing their associated risks and uncertainties.

References: