Beyond Static Analysis

Complexity Science and Dynamic Competition in Modern Antitrust and Consumer Protection

Update: In the past 14 days I testified in three different congressional hearings on: AI and energy, AI and competition, and AI and government efficiency. My respective written testimonies are here, here, and here.

Preamble: The post below is drawn from remarks and materials I developed for my participation in the Chair’s Showcase at the ABA Antitrust Division’s Spring Meeting, the world’s largest annual gathering of competition, consumer protection, and data privacy experts. The panel, “Have We Been Doing it Wrong?” explored how antitrust needs to evolve to address the increasingly complex economy. Steve Cernak, who is the Division’s Chair and organized and moderated the panel, did a nice roundup. Other panelists were Koren Wong, Nicolas Petit, and Diana Moss. The panel provoked some of the usual neobrandeisians, who are adamant that the only way to change antitrust is to make it more political and less dynamic.

For those of you who read me for AI-related content, you might be most interested in the part where I use the AI ecosystem as an case study of the need for new antitrust approaches.

photos by Nicolas Petit

Markets and societies aren’t machines that can be easily designed, controlled, or predicted. They’re more like crowds at a stadium spontaneously creating “the Wave”—orderly yet emergent and surprising. Antitrust needs to embrace this reality, moving from rigid rules to approaches that better respect complexity and encourage innovation.

From Industrial Revolution to High Modernism

In the late 19th and early 20th centuries, humanity surfed an unprecedented wave of scientific achievement and technological advancement. Man tamed nature through science. We proved theories like Newtonian physics and built practical inventions like the steam engine. The result was massive economic growth during the Industrial Revolution. Science was revealing nature’s every secret and making us better off.

This profound confidence coded thought about society and governance. The successes of practical science—demonstrating our newfound power to predict, control, and engineer reliable outcomes—sparked optimism about social “sciences.” The economy and society were viewed as great machines to be rebuilt and re-tuned. Historian James C. Scott call the emerging idea “High Modernism,” a movement aimed at scientifically redesigning all of society from the top down. U.S. President Woodrow Wilson, among others, became an enthusiastic champion of this idea.

Fueled by a belief in centralized expertise, governments rapidly expanded their bureaucratic institutions. Policymakers imagined society as clear, legible, and moldable—something that could be precisely managed through technocratic solutions. This was the mindset under which the Federal Trade Commission was created and U.S. antitrust law was born.

But the Industrial Age-inspired dream of societal control ran into complex reality.

Classical science (mechanistic, Newtonian, deterministic) excelled at explaining phenomena at very small scales, like atoms and molecules, and very large scales, such as planets and solar systems. It also worked well in situations where countless interactions could be neatly averaged without losing critical details.

Thus, the assumption that classical science — so useful at the macro and the micro — could also work at the intermediate scale was understandable. But it was wrong.

Classical science struggled profoundly with the vast complexity in the middle. This overlooked middle ground turned out to be far messier, far more unpredictable, and fundamentally, irreducibly complex. Traditional math and science modeling tools didn’t work on such complex adaptive systems in the physical world. No surprise that such methods also faltered when applied to human society and the economy.

Emergent Order: Understanding Complex Adaptive Systems

What is Emergent Order?

Emergent order is the complex behavior of a system created by the interactions of many smaller components following simpler rules with no central control.

Unlike designed order or random chaos, emergent order arises organically through the interactions of system participants acting according to their own interests and information.

In economics the paradigmatic example of emergent order is market prices: no authority determines the "correct" price for most goods, yet prices efficiently coordinate the activities of millions of producers and consumers. Almost every social phenomenon has emergent order characteristics. Other examples include natural ecosystems where no one dictates how many wolves, deer, or trees should exist in a forest, yet remarkably stable patterns emerge from countless interactions.

Example: A sports stadium

But my favorite example helps distinguish the three fundamental categories of organization: randomness, designed order, and emergent order.

• Randomness: Fans arriving at a stadium before a game behave randomly. You can't predict when any particular person will arrive and sit down. While there's a general trend toward filling up as game time approaches, there's no discernible pattern to individual arrivals. This is uncoordinated activity without complex structure—true randomness.

• Designed Order: In contrast to randomness, designed order appears when fans hold up colored placards to spell "GO TEAM" across the stadium. This pattern doesn't happen by chance—it's centrally planned. Someone mapped out each placard's position and instructed fans exactly what to do and when. The resulting pattern is intentional and controlled, created through top-down directives rather than spontaneous coordination.

• Emergent Order: Between pure randomness and designed order lies a third possibility. "The Wave" perfectly illustrates emergent order. Fans follow a simple rule: "stand when those nearby stand, then sit back down after a moment." No one controls the overall pattern, yet a coordinated wave travels around the stadium. Each person makes independent decisions based on local information (their neighbors' actions) and personal preferences, yet a complex, stadium-wide pattern emerges.

What makes The Wave particularly fascinating is how it begins and ends. A small group might initiate it, but they quickly become regular participants following the same rules as everyone else. The wave's conclusion is equally revealing—it simply dissolves when enough individuals decide to stop participating, with no central command needed. This self-organizing behavior, where complex coordination emerges without anyone being in control, is the essence of emergent order.

Key Characteristics of Complex Adaptive Systems

Complex systems that exhibit emergent order share several pivotal features:

1. Anti-reductionism: The whole is greater than the sum of its parts. Unlike merely complicated systems (such as an automobile engine), full understanding of every individual component of a complex system is not enough to predict system-level behavior.

2. Emergence from Chaos: Emergent order requires some level of initial disorder or freedom for patterns to form. In fact, designed order can prevent more complex emergent order.

3. Feedback Loops: Both positive feedback (amplifying changes) and negative feedback (dampening changes) shape system behavior.

4. Sensitivity to Initial Conditions: Small differences in starting conditions can lead to dramatically different outcomes over time (the "butterfly effect"). This makes accurate modeling of complex systems very difficult.

5. Adaptation: Components learn and evolve in response to their environment and the actions of other components.

6. Non-linearity: Cause and effect relationships are rarely proportional; small inputs can create large outputs and vice versa.

The Elusiveness of Control

These characteristics make complex systems profoundly difficult to control through centralized action. First, no central authority can gather or process all relevant information, much of which is distributed, tacit, or embedded in the connections between the system parts. Economist F.A. Hayek called this the “knowledge problem.” Second, long-term forecasting of the system is fundamentally limited by sensitivity to initial conditions – even small inaccuracies in the information one does gather can spiral into large effects. Third, complex systems are tangled with feedback loops, meaning that system components react and adapt to interventions, often circumventing their intended effects.

The difficulty of gathering the necessary information about the system and predicting its future path, including how it might adapt to interventions, together mean that interventions into complex systems frequently produce unexpected second and third-order effects, making control elusive.

To the extent that control is achieved, it is typically accomplished by eliminating the system’s complexity.

But while control is futile, influence is always possible. The challenge facing those who would shape the system is identifying what changes will make a positive difference without collapsing the complexity of the system. As complexity scientist John Holland put it in his book Complexity: A Very Short Introduction:

“[A]ll well-studied [complex adaptive systems] exhibit lever points, points where a small directed action causes large predictable changes in aggregate behaviour, as when a vaccine produces long-term changes in an immune system. At present, lever points are almost always located by trial and error.”

Antitrust as a Search for Lever Points

One might fairly describe U.S. antitrust law as a long series of experiments to identify such lever points through trial and error. How can we continue this evolution in the light of increasingly complex economies and industries? Specifically, how can regulators and enforcers discover lever points to influence the system while minimizing damage to its productive capability?

In antitrust, traditional approaches attempt to analyze markets as if they were simple, mechanistic systems that can be reduced to a few key variables like price and output. Through a common law process, U.S. antitrust law has identified areas where these simplifying approaches produce generally acceptable results. We’ve found some lever points. However, understanding markets as complex systems characterized by feedback loops, adaptation, and sensitivity to initial conditions may help us find other lever points.

Below, I identify some of the challenges of using existing antitrust approaches in an increasingly complex world and then explore dynamic competition frameworks as a way to bring complex system analysis to antitrust law. I also explore parallel implications for consumer protection law.

The Limits of Current Approaches

Today's antitrust and consumer protection frameworks rely heavily on static economic tools that often fail to capture the full dynamics of modern markets—particularly those driven by rapid innovation and technological change. These traditional approaches:

• Focus primarily on short-term price effects rather than long-term innovation benefits

• Rely on narrowly defined relevant markets that may not reflect economic reality

• Overemphasize market concentration measures as primary indicators of potential harm

• Apply simplified assumptions about market equilibrium that rarely exist in practice

• Underestimate the role of potential and nascent competition

• Struggle to account for multi-sided platforms and network or aggregation effects

These traditional economic approaches remain valuable in many contexts. Static analysis provides a clear, predictable, and judicially manageable framework for evaluating competition in stable, mature markets with well-defined products and services. These models excel when examining horizontal mergers in concentrated industries, evaluating the effects of explicit collusion among competitors, or analyzing straightforward price effects in markets with homogeneous products. The clarity and relative simplicity of traditional approaches also offer practical advantages for enforcement agencies and courts, providing a structured methodology with established precedent and defined standards of proof. In these contexts, traditional tools like market definition, concentration metrics, and price-based analyses can efficiently identify potential competitive harms without requiring more complex and resource-intensive investigation methods.

However, the traditional framework is less well-equipped to address complex, rapidly evolving ecosystems like artificial intelligence, digital platforms, and other innovation-driven markets. As these technologies become increasingly pervasive and meaningful in the economy, a more nuanced approach is needed—one that incorporates insights from complexity science and emergent order.

Dynamic Competition: A Framework for Complex Markets

Effective competition analysis must understand how firms compete. Traditional antitrust analysis focuses on how firms compete through cost-saving efficiencies to deliver the same products at lower prices. Yet the forms of competition that most enhance consumer welfare often involve creating entirely new products, services, and business models. Traditional antitrust analysis has struggled with analyzing these forms of competition.

The dynamic competition framework developed by economist David Teece and others better reflect the complex systems nature of real markets. Such frameworks acknowledge that innovation, capabilities, and adaptation are fundamental aspects of competition—not mere afterthoughts or side effects. They recognize that competition is an evolutionary process unfolding through time rather than a static state that can be captured in a snapshot.

Key elements of the dynamic competition framework include:

1. Innovation-Driven Competition: Antitrust typically considers innovation to be a result of effective competition. But innovation drives competition at least as much as competition drives innovation. ChatGPT's disruption of the search market shows how innovation can fundamentally reshape competitive dynamics.

2. Capabilities-Based Analysis: A firm's competitive position depends on its technological and organizational capabilities, not just its market share. Understanding how a firm’s capabilities complement, substitute for, or conflict with other firms’ capabilities provides a more nuanced analytical framework.

3. Forward-Looking Perspective: Analysis must consider potential future markets and competitive threats. This includes a greater emphasis on supply-side factors, including technological capabilities and potential market entrants. It also includes an enhanced focus on emerging competitive threats.

4. Ecosystem Approach: Markets should be viewed as broader ecosystems rather than narrowly defined product categories. Such analysis can help identify entry threats who can maintain competitive pressures.

5. Long-Term Consumer Welfare: Benefits to consumers include innovation, product quality, and availability—not just price effects. In markets with highly differentiated, non-substitutable products or services, finding ways to evaluate these other benefits becomes more important.

6. Mergers and Acquisitions: Transactions should be evaluated based on their potential to enhance innovation and capabilities, not just their effect on market concentration.

Dynamic competition approaches are especially important in rapidly evolving sectors where innovation drives both disruption and consumer benefit. This perspective allows us to better assess competitive effects not just on current market structures but on the ongoing process of creative destruction that fuels long-term economic growth and consumer welfare.

Here are some of the implications of a dynamic competition framework for a variety of antitrust doctrines.

Market Definition and Power Assessment

Traditional market definition exercises often produce artificial boundaries that fail to capture competitive realities in dynamic markets. Dynamic competition analysis suggests a more nuanced approach that extends beyond conventional methods. Rather than relying on narrowly defined product markets, this framework encourages consideration of broader ecosystems where competition may manifest in less obvious ways. It emphasizes the importance of examining potential competition across different ecosystem or supply-chain layers, recognizing that threats to incumbent market power may emerge from adjacent markets or from firms operating at different stages of production. Furthermore, dynamic analysis prioritizes assessment of firms' capabilities—their technological assets, organizational know-how, and innovation potential—rather than focusing exclusively on market shares as indicators of competitive significance. This approach also calls for a more sophisticated evaluation of barriers to entry that accounts for dynamic factors like access to complementary assets, the rate of technological change, and learning effects. Perhaps most importantly, dynamic competition frameworks acknowledge that market power is often more transient in innovation-driven industries, where technological disruption can rapidly undermine seemingly entrenched positions.

Merger and Investment Reviews

The dynamic competition framework offers the most value in assessing acquisitions of nascent competitors. Teece criticizes using traditional tools to effectively evaluate such situations, calling it a “fool’s errand” unless enforcers “look under the hood of the target and acquirer in ways agencies have not yet done.”

Specifically, when evaluating mergers involving nascent competitors, Teece proposes seven threshold factors. His analysis first examines whether the acquiring firm holds monopoly power, establishing a baseline for potential competitive harm.

Attention then turns to evaluating whether the nascent firm is sustainable absent acquisition. Has the nascent firm’s technology moved beyond theory to demonstrate viability and practical proof of concept? Has the emerging company established a successful business model that can effectively monetize its innovations? The strength of the firm's leadership also matters—specifically, does the management possess the talent needed to execute its vision independently?

The analysis then pivots to the technological relationship between the firms, focusing particularly on whether the nascent firm's innovations could disrupt the acquiring firm's core revenue streams and capabilities or merely complement them.

Finally, Teece’s framework evaluates the wider competitive environment, considering whether other similarly situated nascent firms might offer competitive pressures even if this particular firm is acquired.

This nuanced approach guards against both excessive enforcement, which could block acquisitions that promote innovation, and inadequate enforcement, which could allow genuinely anticompetitive mergers.

Exclusionary Conduct Analysis

Dynamic competition frameworks offer a more sophisticated approach to analyzing potentially exclusionary conduct. At their core, these frameworks emphasize the critical distinction between conduct that genuinely harms the competitive process and actions that merely disadvantage individual competitors—recognizing that vigorous competition often produces losers even as it benefits consumers and the market overall. Traditional analysis gives lip service to this concept, but dynamic competition seeks to give it teeth. This approach considers whether the conduct at issue enhances innovation capacity even if it temporarily disrupts current market structures, acknowledging that creative destruction frequently requires displacing established arrangements. The analysis extends beyond immediate price effects to evaluate how conduct influences long-term consumer welfare through improved products, services, and business models. Dynamic frameworks also recognize that some degree of temporary market power may be inevitable to incentivize significant innovation investments, particularly in industries with high development costs or substantial spillover effects. Perhaps most importantly, this approach assesses whether the conduct in question enables or impedes the market's natural adaptive processes—allowing inefficient structures to evolve into more productive arrangements or artificially preserving outdated business models that might otherwise yield to superior alternatives.

Complex Systems Implications for Consumer Protection

While dynamic competition frameworks address antitrust analysis, the underlying principles of complexity science and emergent order also have profound implications for consumer protection policy. This should not be surprising, as both domains deal with the same underlying complex systems in the economy, and both ultimately seek to enhance consumer welfare. Like antitrust, consumer protection law can also benefit from better reflecting the realities of complex systems.

Ex Post Enforcement Over Ex Ante Regulation: Case-by-case adjudication based on actual market outcomes often proves superior to prescriptive regulations that attempt to anticipate all potential harms. Complex systems resist top-down control and are better influenced through incremental interventions that can adapt to feedback. The FTC's consumer protection model demonstrates how enforcement focused on provable consumer harm can effectively protect consumers while preserving space for innovation and adaptation.

Flexible Standards Rather Than Rigid Rules: Consumer protection frameworks should establish broad principles that can be applied across evolving technologies and business models, rather than specific requirements that quickly become obsolete. This approach recognizes that market participants will continually adapt to regulatory constraints in ways that regulators cannot fully predict, making flexibility essential. Case-by-case adjudication allows agencies to apply general principles to specific situations as they arise, preventing laws from becoming too quickly outdated while enabling more nuanced application of the law to complex or novel situations.

Focus on Outcomes, Not Processes: Regulations that dictate specific processes or business practices often inadvertently lock in obsolete models and create barriers to innovative solutions. Instead, consumer protection should focus on preventing harmful outcomes while remaining agnostic about the means used to achieve beneficial results. This approach acknowledges that in complex systems, the same outcome can emerge from different processes, and innovation often takes unexpected forms.

Harness Market-Based Mechanisms: Complex systems frequently develop self-correcting mechanisms that can address potential harms more effectively than external regulation. Consumer protection policy should recognize and potentially amplify these emergent solutions—such as reputation systems, third-party certifications, and industry standards—rather than displacing them with regulatory substitutes.

Consider Information Asymmetries: Rather than dictating specific disclosures that may become irrelevant as markets evolve, consumer protection should focus on addressing fundamental information asymmetries that prevent markets from functioning efficiently. This approach recognizes that in complex adaptive systems, information flows are crucial to enabling beneficial emergent outcomes.

Assess Full Regulatory Impact: When evaluating potential interventions, agencies should consider not only the direct harms prevented but also the innovation suppressed and benefits foregone. Complex systems analysis reminds us that interventions have ripple effects throughout interconnected markets, often in ways that are difficult to predict but nonetheless significant.

These principles can help consumer protection actions become more consistent with the complex systems involved, making it more adaptive and effective at promoting consumer welfare in rapidly evolving markets and industries.

Case Study: Artificial Intelligence Governance

The AI ecosystem provides an ideal case study for applying complexity science and dynamic competition frameworks to both antitrust and consumer protection challenges. The multi-layered, rapidly evolving nature of AI markets exemplifies the limitations of static analysis and the need for more sophisticated approaches.

Complexity of the AI Ecosystem

AI operates across multiple interconnected layers, each with distinct economic characteristics. These layers include hardware (chips, computing infrastructure), foundation models, application layers, and specific use cases. Each layer exhibits different market dynamics, with varying barriers to entry, network effects, and scale economies. Many companies operate within multiple layers of the AI stack. This complexity makes traditional market definition exercises particularly problematic, as static snapshots fail to capture the dynamic interactions between layers.

The AI landscape also demonstrates how innovation drives competition at least as much as competition drives innovation. The release of ChatGPT, for example, has created intense competitive pressure on online general search engines. A static focus on existing market shares in online search would entirely miss this competitive disruption.

Dynamic Analysis for AI Antitrust

Analyzing competition in AI markets requires moving beyond concentration metrics to examine capabilities, potential competition, and cross-layer dynamics. Traditional market definitions quickly become obsolete as boundaries blur between AI providers, applications, and adjacent services. For instance, cloud providers developing their own AI chips to reduce dependence on other suppliers demonstrates how competition at one layer affects competitive conditions at other layers.

The dynamic competition framework offers value for evaluating partnerships and investments in the AI sector. Competition authorities in the U.S. and elsewhere have scrutinized relationships between established technology companies and emerging AI firms. Rather than focusing narrowly on horizontal market overlap or vertical foreclosure theories, authorities should consider how these arrangements might enhance innovation capacity and capabilities development.

Open-source AI models further complicate traditional antitrust analysis. Early commenters expressed concern that only a few large companies would develop advanced AI models. Yet the barriers to creating highly capable models have fallen rapidly. Open models have proven very popular, fostering collaboration and interoperability within the AI community and potentially driving competition at the application and fine-tuning layers rather than in foundation model development itself. This demonstrates how competitive dynamics can shift across different layers of the stack, requiring a more nuanced competitive analysis than traditional market concentration metrics.

Emergent Order in AI Consumer Protection

The consumer protection challenges posed by AI—including privacy concerns, algorithmic bias, and potential deception—similarly benefit from an emergent order perspective. Traditional regulatory approaches that attempt to anticipate and prevent all potential harms through detailed rules are likely to become quickly outdated in this rapidly evolving field.

The FTC's consumer protection model offers valuable lessons for AI governance. Its case-by-case adjudication approach, focused on addressing and preventing substantial consumer injury, provides flexibility to address novel harms while avoiding prescriptive rules that might impede innovation.

The complexity of AI systems also highlights the importance of information flows and feedback mechanisms. Rather than attempting to directly control AI development in the abstract through rigid regulations, policymakers might focus on identifying potential regulatory gaps in specific high-risk use cases, ensuring commercial promises to users are kept, and monitoring the effectiveness of private sector transparency efforts and accountability mechanisms such as third-party auditing or certification. These approaches recognize that complex AI systems will continually adapt to changing conditions in ways that no one, let alone regulators, can fully predict.

The AI ecosystem thus demonstrates both the limitations of traditional static analysis and the potential of dynamic, complexity-informed approaches to promote innovation while protecting consumers in rapidly evolving technology markets.

Key Takeaways for Practitioners and Regulators

1. Adopt Dynamic Analysis: Move beyond static price effects to consider innovation, capabilities, and long-term consumer welfare.

2. Embrace Humility: Recognize the inherent limitations in predicting and controlling complex systems. Consider what Hayek called "the knowledge problem"—the impossibility of centralizing all relevant information.

3. Favor Incrementalism: Small, reversible interventions allow for learning and adaptation. Avoid comprehensive regulation that may lock in suboptimal arrangements.

4. Harness Local Knowledge: Solutions that leverage context-specific information generally outperform one-size-fits-all approaches.

5. Protect the Innovation Process: Focus on maintaining conditions that enable innovation and competition rather than on specific market outcomes.

6. Consider Unintended Consequences: Evaluate how market participants might adapt to interventions in ways that undermine their intended effects.

7. Balance Present and Future: Weigh immediate consumer benefits against long-term innovation incentives.

8. Monitor Ecosystem Health: Look for signs of adaptation, entry, and innovation rather than just concentration measures.

9. Develop Better Tools: Invest in methods that can better capture the realities of dynamic competition and complex systems.

10. Remember the Goal: The aim of competition policy is to enhance consumer welfare through the competitive process, not to protect competitors or preserve specific market structures.

Conclusion

Complexity science and dynamic competition frameworks offer a path forward for antitrust and consumer protection analysis in rapidly evolving markets. By recognizing the emergent nature of market order and the central role of innovation in driving competition, authorities can develop more effective approaches that protect consumers while fostering the conditions for continued innovation and prosperity.

Rather than abandoning economic analysis, as some critics suggest, we should advance our economic understanding to better reflect the complex reality of modern markets. The characteristics of artificial intelligence and other innovative industries make them the perfect testing ground for this more sophisticated approach to competition policy.

Additional Reading:

Neil Chilson, Getting Out of Control: Emergent Leadership in a Complex World, New Degree Press (2021).

Comments of the Abundance Institute to Department of Justice on workshop, “Promoting Competition in Artificial Intelligence,” https://www.justice.gov/atr/media/1361126/dl?inline.

David J. Teece, The Dynamic Competition Paradigm: Insights and Implications, 2023 Colum. Bus. L. Rev. 373, 451-52 (Aug. 2023). DOI:https://doi.org/10.52214/cblr.v2023i1.11895.

Melanie Mitchell, Complexity: A Guided Tour, Oxford University Press (2011).