The Innovation Paradox: What Andrew Leigh's Framework Reveals About Trump's AI Executive Order
A Tale of Two Philosophies
In his compelling work "The Shortest History of Innovation," Australian economist Andrew Leigh traces humanity's greatest leaps forward through three essential forces: tinkering (local experimentation and diverse approaches), teams (collaboration at scale), and trade (the free flow of ideas across borders and disciplines). These forces, Leigh argues, determine whether innovation flourishes or withers.
Now consider this: In December 2024, President Trump signed an executive order aimed at cementing American AI dominance. The order takes direct aim at state-level AI regulations, establishing a federal litigation task force to challenge "onerous" state laws and making states with stricter AI rules ineligible for critical broadband infrastructure funding. Over 1,000 AI-related bills were introduced across U.S. states in 2025, representing a genuine marketplace of regulatory ideas. The administration argues this diversity creates compliance chaos that stifles innovation, particularly for startups.
But here's the paradox: In trying to win the AI race through centralized control, are we undermining the very innovation dynamics that created American technological dominance in the first place?
Leigh's Framework Meets the AI Revolution
Before examining Trump's order, we need to understand how Leigh's three forces are already shaping the global AI landscape in revealing ways.
The Tinkering Explosion: AI development is producing wildly different innovations across different contexts. DeepSeek's rise through open-source models and free access has gained particular traction in China, Russia, Iran, Cuba, Belarus, and across Africa. This isn't random—it demonstrates how different environments, constraints, and values produce distinct innovations tailored to local needs. The same technology spawns different solutions when different communities tinker with it.
The Teams Divide: Here's where inequality becomes stark. Countries investing early in digital infrastructure, AI skills, and government adoption—UAE, Singapore, Norway, Ireland, France, and Spain—continue to lead, with the UAE reaching 64% AI adoption among working-age adults. Meanwhile, 3.7 billion people in Asia-Pacific remain on the sidelines, with a quarter of the population still offline. AI adoption in the Global North grew nearly twice as fast as in the Global South: 24.7% versus 14.1%. This threatens what researchers are calling "The Next Great Divergence"—potentially reversing a half-century trend of narrowing development gaps between nations.
The Trade Battleground: This is where Leigh's framework reveals the most tension. The US is reportedly planning to restrict exports of advanced AI chips on a country and company basis, citing national security. AI competition between the US and China has become a race to promote adoption of their respective national models. This represents exactly the kind of trade barriers Leigh warns about—forces that can suppress innovation by blocking the flow of ideas.
The question hovering over all of this: Will AI spread inclusively, or will it concentrate power? Will the "trade" of ideas flow freely, or get choked by protectionism and inequality?
Trump's Executive Order: Centralization in Innovation's Clothing
Enter Trump's executive order, which promises to cut through regulatory red tape and position America for global AI leadership. The order establishes an AI Litigation Task Force to challenge state laws and directs the Commerce Department to identify "onerous" state AI regulations. Most significantly, it makes states with such laws ineligible for Broadband Equity Access and Deployment Program funds.
The administration's argument has surface appeal: a patchwork of 50 different regulatory regimes makes compliance more challenging, particularly for startups trying to scale. National uniformity, they argue, will free innovators from navigating contradictory state requirements and allow American AI companies to move faster than international competitors.
But let's examine this through Leigh's framework, because the reality is far more complex.
Tinkering Under Threat: The Laboratory of Democracy
Leigh celebrates tinkering—the messy, diverse experimentation that produces breakthrough innovations. American federalism has historically been innovation's greatest ally. States serve as "laboratories of democracy," testing different approaches to emerging challenges. Some experiments fail. Some succeed spectacularly. The successful ones spread organically as other states observe and adapt them.
In 2025, this laboratory was working overtime. More than 1,000 AI-related bills were introduced across U.S. states and territories. Colorado developed algorithmic discrimination standards. California explored comprehensive AI frameworks. Different states were testing different balances between innovation and safety, between corporate flexibility and consumer protection, between rapid deployment and careful oversight.
This diversity wasn't chaos—it was the innovation process itself.
Federal preemption threatens to flatten 50 experiments into one approach. Instead of learning which regulatory models work best through real-world testing, we'd impose a single untested framework nationwide. If that framework is wrong, everyone loses. If it's right, we'll never know what even better approaches we missed.
History offers a warning here. When has federal uniformity accelerated innovation versus stifled it? The internet flourished under light-touch federal regulation that allowed different business models to compete. Telecommunications innovation stagnated under the AT&T monopoly until forced breakup created competition. Environmental protection improved when California's stricter standards pushed national improvements, not when federal rules prevented state experimentation.
The Startup Paradox: Who Really Benefits?
Here's where the order's justification collapses under scrutiny. The administration claims it's helping startups by removing compliance burdens. But consider what state-level regulation actually provides for new entrants:
State ecosystems offer startups multiple pathways to success. A company might thrive under Colorado's framework but struggle under a different model. State programs provide funding, technical assistance, and supportive infrastructure tailored to local strengths. State procurement creates early customer bases. State universities provide talent pipelines.
Uniformity doesn't help startups—it helps established players. Large tech companies have compliance departments that can navigate any regulatory framework. They have lobbying power to shape federal rules. They have market dominance that insulates them from new entrants.
Startups need the opposite: diverse regulatory environments where their specific innovations might find friendlier conditions. They need states competing to attract the next breakthrough company. They need regulatory diversity that prevents any single gatekeeper from blocking their path.
By eliminating state variation, the order doesn't level the playing field—it tilts it toward whoever controls the federal regulatory process.
Teams and Scale: The Double-Edged Sword
Leigh emphasizes that innovation requires teams operating at scale—collaboration, resource pooling, and institutional support. The administration has a legitimate point here: companies growing from one state to national scale do face real coordination challenges when regulations vary.
But this assumes all companies need the same scale, on the same timeline, under the same model. It ignores that different scales need different regulatory environments. A local AI healthcare startup serving one hospital system has different needs than a national social media platform deploying recommendation algorithms to billions.
The EU offers an alternative model: tiered regulation based on risk and scale. High-risk applications face stricter scrutiny. Low-risk applications get lighter touch. Companies graduate between tiers as they grow. This allows both local innovation and scaled deployment without forcing everyone into the same box.
Federal uniformity treats a seed-stage startup and Google AI identically. That's not removing barriers to teams—it's ignoring that teams operate at different stages with different needs.
Trade as Coercion: Infrastructure Hostage-Taking
This is where Trump's order most dramatically inverts Leigh's vision. Leigh celebrates trade as the free flow of ideas across borders and disciplines. Ideas spread because they work, because people voluntarily adopt them, because they solve problems better than alternatives.
The executive order's BEAD funding conditionality does something entirely different: it uses infrastructure access as leverage to enforce regulatory uniformity. States with "onerous" AI laws lose federal broadband funding. This isn't the trade of ideas—it's coercion.
Think about the implications. Broadband access is critical infrastructure for economic participation, education, and civic engagement. Making it conditional on AI regulatory compliance means states must choose to adopt federal AI preferences or accept that your rural communities, underserved populations, and small businesses won't get connectivity.
This is trade in name only. Real trade means California could try strict AI transparency rules, Texas could try light-touch regulation, and both could learn from each other's results. Successful approaches would spread as other states observed benefits. Failed approaches would be abandoned as costs became clear.
Instead, we get enforced uniformity through financial punishment—exactly what Leigh warns kills innovation.
The Global Irony: Becoming What We Oppose
Here's the deepest irony: The order aims to position the U.S. for "global AI dominance" by preventing China from gaining advantage. But in pursuing this goal through centralized federal control, we're adopting the very model that we claim puts China at a disadvantage.
China's centralized AI governance allows rapid national deployment but suppresses the distributed experimentation that produces breakthrough innovations. The Chinese government can mandate AI adoption across industries, but it struggles to create the messy, bottom-up innovation that characterized American tech success.
American technological dominance didn't come from central planning. It came from garage startups, university spin-offs, state-supported innovation hubs, competing regional ecosystems, and regulatory diversity that allowed different approaches to compete.
Silicon Valley, Boston's biotech corridor, Austin's tech scene, North Carolina's Research Triangle—these emerged through local initiatives, state support, university partnerships, and regulatory environments that varied by region. Federal uniformity would never have produced this diversity.
By centralizing AI regulation, we're creating two monolithic blocs—the U.S. federal approach versus China's state approach—rather than the messy marketplace of ideas that Leigh shows drive progress. We're having a centralization contest when we should be having an innovation contest.
The autocrat's advantage is a mirage. Centralized systems can move fast initially but lack the resilience and adaptability of distributed systems. They're brittle—when the center gets it wrong, everything fails.
What the Data Shows: The Stakes Beyond America
This isn't just about U.S. policy—it's about global inequality and who gets left behind in the AI revolution. The current trajectory is deeply troubling.
The Global North-South AI adoption gap is widening: 24.7% versus 14.1%. That represents billions of people being excluded from the productivity gains, educational opportunities, and economic benefits of AI tools. Meanwhile, 3.7 billion people in Asia-Pacific remain entirely offline.
When the world's two largest economies both pursue centralized AI control—the U.S. through federal preemption, China through state direction—what happens to the Global South? They don't get diverse models to choose from based on their needs. They get two dominant paradigms, both designed for great power competition, neither optimized for inclusive development.
Leigh would recognize this pattern. Throughout history, innovation either spreads broadly and lifts development, or concentrates in elite hands and deepens inequality. The current trajectory points toward concentration.
What Leigh Would Recommend: A Different Path
So what would an innovation-friendly approach to AI governance look like, viewed through Leigh's framework?
Embrace regulatory federalism as a feature, not a bug. Recognize that 50 states testing different approaches provides invaluable information about what works. Some states will get it wrong—that's fine, others will learn from their mistakes. Some will discover breakthrough approaches that can spread nationally.
Create interoperability standards without uniformity mandates. The federal role should be ensuring states can share data, coordinate enforcement, and allow companies to operate across borders—not dictating the substance of regulations. Think internet protocols: common standards that allow infinite diversity in implementation.
Fund state experimentation rather than punish it. Instead of making BEAD funding conditional on AI regulatory compliance, use federal resources to support diverse state approaches. Fund pilot programs. Facilitate knowledge sharing. Help states learn from each other. Let successful models prove themselves and spread organically.
Facilitate horizontal trade of regulatory insights. Create mechanisms for states to share results, compare outcomes, and adapt each other's successful innovations. The National Conference of State Legislatures could coordinate AI policy laboratories. Academic institutions could study comparative results. The federal government could synthesize findings without mandating conclusions.
Let markets and civil society shape adoption. If a state's AI rules genuinely harm innovation, companies will struggle and voters will demand change. If another state's rules successfully balance innovation and protection, others will copy the approach. Trust the process that's worked for American innovation for centuries.
The Core Insight: Innovation Cannot Be Commanded
Leigh's fundamental insight is that innovation emerges from conditions that enable it—diverse experimentation, collaborative teams, and free-flowing ideas—not from top-down direction. You can't centralize your way to distributed innovation. You can't command creativity. You can't mandate breakthroughs.
The history of innovation is littered with central planners who thought they knew the optimal approach and tried to impose it. The Soviet Union invested heavily in science but produced little commercial innovation. France's Minitel was a top-down success that couldn't compete with the bottom-up chaos of the internet. Japan's Fifth Generation Computer Project spent billions on government-directed AI research in the 1980s and failed to match distributed American innovation.
The pattern is consistent: centralized systems can execute known strategies efficiently but struggle to discover new approaches. Distributed systems are messy and inefficient but generate the variation and selection that produces breakthroughs.
Trump's executive order assumes we already know the right approach to AI governance and just need to implement it uniformly. Leigh's framework suggests the opposite: we don't know the right approach yet, which is exactly why we need many different experiments running in parallel.
Stakes Beyond AI: A Precedent for All Emerging Technologies
This matters beyond artificial intelligence. The precedent set here will shape how America governs every emerging technology—biotechnology, quantum computing, nanotechnology, climate technology, synthetic biology.
If federal preemption becomes the model, we'll face the same pattern repeatedly: Washington identifies an important technology, claims national uniformity is essential for competitiveness, overrides state experimentation, and imposes a single approach designed by whoever has the most influence over federal policy.
That's not a recipe for innovative leadership. It's a recipe for regulatory capture, institutional sclerosis, and competitive decline.
Defending messy state experimentation is defending America's competitive advantage. The chaos, the redundancy, the contradictions, the failures—these aren't bugs in the system. They're features that enable innovation.
Conclusion: The Real Race
The real race isn't United States versus China. It's centralization versus the creative chaos that drives human progress.
China is testing whether centralized AI governance can produce both control and innovation. That's their bet, consistent with their political system. The early results are mixed—impressive deployment speed, but breakthrough innovations still emerging primarily from more open systems.
America's bet has always been different: that distributed innovation, messy federalism, competing jurisdictions, and diverse experiments produce better long-term results than centralized planning. Not faster results in the short term. Not tidier results. Not more controlled results. But better results—more resilient, more adaptive, more likely to produce unexpected breakthroughs.
Trump's executive order abandons that bet. It assumes the Chinese model is winning and we need to match it. But Leigh's framework suggests we should be doubling down on what makes America different, not copying what makes China similar.
The question before us isn't whether we'll have AI regulation. We will. The question is whether that regulation will emerge through diverse state experimentation or federal imposition. Whether it will adapt through decentralized learning or calcify through centralized control. Whether it will enable the tinkering, teams, and trade that Leigh shows drive innovation, or suppress them in the name of uniformity.
Andrew Leigh traces how new ideas emerge, why some spread while others stall, and the engines powering progress. His answer is clear: ideas emerge from experimentation, spread through voluntary adoption, and progress through the free flow of knowledge.
Trump's AI executive order does the opposite: it suppresses experimentation, coerces adoption, and restricts the flow of regulatory knowledge.
History suggests we'll regret that choice. The innovation we need won't come from commands. It never does.