Prohibition, Monopoly, and the AI Wild West

Artificial intelligence is often described as a Wild West.

Usually, this is meant as shorthand for lawlessness: rules lag behind reality, somebody is shooting somebody at any given street corner, fortunes appear overnight, speculators arrive alongside serious pioneers, and institutions struggle to keep up.

Fair enough. But the Wild West was not only lawlessness. It was also a frontier.

And before congratulating ourselves for ending the Wild West, it is worth remembering what frontiers actually do.

They are messy precisely because no single institution controls the process of discovery. Thousands of people are experimenting at once, producing everything from spectacular failures and outright exploitation to unexpected solutions, new settlements, new connections and entire industries nobody had planned in advance.

The American West was violent, extractive and profoundly unjust in ways that should not be romanticized. It also unleashed enormous parallel experimentation.

Those two facts can coexist; the disorder was part of the danger, but it was also part of what made the frontier generative.

That distinction matters when we use the Wild West as a metaphor for artificial intelligence.

Today’s AI ecosystem is undeniably chaotic. Companies are racing to establish themselves. Capital is pouring into the sector at something like Gold Rush 2.0 speed. The result is not an orderly process of development, but a volatile mix of genuine research, speculative investment, strategic competition and hurried commercialization.

Researchers jump between laboratories. Open-source models compete with proprietary systems. Startups appear and disappear almost overnight. A new “LLM mine” is discovered 50 miles up the hill. Governments are trying to write rules for technologies that change before the rules can make it through a legislature.

There are cowboys, yes, but also snake-oil salesmen. There are probably a few people shooting at the saloon. But there are also thousands of independent experiments happening simultaneously.

A small research group can pursue an idea a large corporation has dismissed, while competing laboratories can challenge one another’s assumptions and expose weaknesses that might otherwise go unnoticed. Researchers can carry expertise from one institution to another, open systems can surface knowledge that closed ones would prefer to keep proprietary, and entirely new applications can emerge far from the organizations that developed the underlying technology.

The Seductive Simplicity of “Just Halt It”

Faced with the possibility that increasingly capable AI could eventually become dangerous at a civilizational scale, there is an understandable response: Stop! Pause development. Halt the race. Do not build systems more powerful than the ones we already have until we understand how to control them.

As an instinct, this makes sense. As a description of what would actually happen, I think it is naive.

You cannot put the genie back in the bottle.

Once knowledge exists, it cannot meaningfully be made unknown again.

The mathematics exists. The algorithms exist. The papers exist. Thousands of engineers understand techniques that were barely conceivable a generation ago.

Those people live in different countries and work for different companies. They publish, copy ideas, combine them, improve them and sometimes leak them.

Artificial intelligence is no longer a secret locked inside one laboratory. Put a roadblock in front of a tank, and someone will mount the technology on a bike and ride away with it. And its potential economic, scientific and military value is far too great for everyone, everywhere, to agree indefinitely not to pursue it.

You can prohibit public development—although governments themselves may have strong incentives to carve out exceptions—or regulate access to computing power, shut companies down, and make open development slower, more expensive and more dangerous.

What you cannot do is rebottle the genie. Someone, somewhere, will continue. And once we accept that, the policy question changes dramatically.

The question is no longer simply how to stop AI development; it becomes: who will still be developing it after everyone else has been persuaded, regulated or frightened into stopping?

We Have Tried Prohibition Before

This is not the first time societies have mistaken prohibition for disappearance.

The United States tried to prohibit alcohol, but alcohol did not vanish. People did not forget how to make it, and demand did not evaporate because Congress said it should. Instead, the structure of the market changed.

Legal production contracted as supply chains moved underground and enforcement became uneven, creating an extraordinarily valuable market for organizations willing to violate the law after legitimate businesses had been pushed out.

Prohibition did not merely suppress an activity; it selected for the actors willing and able to continue that activity illegally.

That is the relevant lesson for AI. Not that nothing should ever be regulated, or that dangerous technologies should be allowed to develop without limits.

The lesson is that prohibiting something while leaving the knowledge, demand and incentives intact can reorganize a problem into a form that is harder to see and potentially harder to control. It is ludicrous to think that solutions that do not work with products like alcohol will work for something as fundamental as knowledge.

Imagine that tomorrow the major American AI companies agree to stop frontier development and Europe follows suit. Open-source development above a certain capability level is prohibited, universities lose access to massive training infrastructure, smaller companies struggle to meet new security and compliance requirements, and public research slows sharply.

For a while, it might look as though the halt worked. But no one could know whether rival states, military or intelligence programs, wealthy private actors, or competing corporations were still pushing forward in secret.

That uncertainty is not incidental. It changes the game.

We could congratulate ourselves for closing the saloons while the distilleries move into the basement. And AI presents an even harder problem than Prohibition did.

Bootleggers were merely pursuing profit, but the actors developing advanced artificial intelligence may be pursuing economic dominance, military advantage, scientific leadership or national security. The incentives to defect are not incidental to the problem; they are built into it.

Which is why even a widely announced halt immediately raises the question that matters most: who believes everyone else has actually stopped?

The Game Theory of a Halt

The problem becomes clearer when nobody involved has to be evil.

Suppose every major AI laboratory sincerely believes that developing these systems too quickly is dangerous. Each laboratory still has to ask what happens if it stops and another does not.

The same applies to governments. The United States might genuinely prefer an enforceable international agreement restricting frontier AI development. But if American officials believe China might secretly continue, stopping American development becomes strategically dangerous.

China can make exactly the same calculation about the United States. Neither side has to want an AI arms race. Each only has to fear losing one. The better everyone else behaves, the greater the potential advantage for whoever defects.

That is the trap.

The stronger the prohibition, the more valuable successful evasion becomes. And if verification is imperfect, suspicion becomes rational.

A government wonders whether another government has hidden a program. A corporation wonders whether a competitor has obtained an exemption. A security agency wonders whether a foreign laboratory has crossed a threshold nobody else knows about. One secret breakthrough could change the balance of economic, military or political power.

Under those conditions, continued development is not some remote possibility that might occur despite a halt. It is what the incentive structure rewards.

The race does not disappear; it changes location, becoming quieter, less transparent and, crucially, accessible to fewer people.

From the Frontier to the Company Town

If the Wild West gives us a historical image of too little control, another part of American history offers a glimpse of the opposite problem: the company town.

Company towns grew around mines, mills, railroads and factories. In some of them, the corporation did not merely employ the worker. It also owned the worker’s housing, the local store and much of the infrastructure necessary for everyday life.

This could be efficient, even comfortable. The danger was not necessarily misery, but dependency.

The same institution paid your wages, rented you your home and sold you what you needed to live. The company was no longer one participant in your economic life; it had become the environment in which your economic life occurred.

That is worth thinking about before we decide that the obvious solution to the AI frontier is to make frontier development so expensive, restricted and regulated that only three or four corporations can participate.

At first, this could look wonderfully responsible: those companies would have sophisticated security departments, dedicated safety researchers, teams of lawyers, government relationships, enormous compliance budgets and secure computing facilities, while regulators would know exactly whom to audit.

The frontier would finally have sheriffs. But imagine what else those same companies might control…

Businesses, researchers, professionals and governments increasingly build their work around these models, while search, analysis and administrative systems become ever more dependent on the same underlying infrastructure. Eventually, the issue is no longer that three companies make the best AI. It is that an increasing portion of civilization thinks through three companies.

That is the AI company town.

Except the company does not own the mine, the workers’ houses and the general store. It owns the compute, the models and the intelligence layer connecting everything else.

The Cure Can Create Its Own Disease

This is what makes the AI problem much harder than the familiar argument between acceleration and restraint.

Competition creates risk by rewarding speed and first-mover advantage, sometimes pushing actors to release systems before their implications are fully understood. Yet competition also distributes power.

Regulation can reduce some of those risks by imposing safety requirements, external testing and liability for reckless behavior. But those protections come with costs, and once those costs become high enough, regulation can stop merely governing a market and start deciding who is allowed to exist within it.

The largest corporations can absorb billion-dollar compliance costs; universities, independent researchers, startups and open-source communities often cannot.

Soon the regulation that was designed to protect society from AI has also protected a handful of AI companies from competition.

A moratorium presents the same paradox. It can slow visible development, but the more strategically valuable the prohibited research becomes, the greater the incentive to conduct it somewhere nobody can see.

Open development makes powerful capabilities available to more people. Closed development concentrates them among the institutions powerful enough to close everyone else out.

There is no clean side of this equation. That is precisely why “just halt it” is not enough.

What If the Mess Is Part of the Safety System?

There is an uncomfortable possibility hidden inside all of this: the chaotic AI ecosystem we currently dislike may contain one of its own safeguards. Not because chaos is safe. It is not. But because distributed knowledge makes complete control difficult.

A plural ecosystem creates its own checks: companies can challenge one another’s claims, researchers can move between institutions, journalists can investigate, competitors can reproduce discoveries, and open-source communities can keep certain techniques from becoming the permanent intellectual property of a tiny number of corporations.

Competition creates dangerous incentives, but it also creates counterweights: the same fragmentation that makes artificial intelligence harder to govern may make it harder for anyone to govern society through artificial intelligence.

This does not mean the answer is laissez-faire.

The Wild West was not some libertarian paradise to which we should aspire. Frontiers eventually need laws, courts, standards and institutions capable of punishing fraud, protecting people from reckless behavior and deciding which activities impose unacceptable risks on everyone else.

The point is not to preserve lawlessness; it is to preserve plurality.

Safety rules are not the same thing as permanent concentration. Independent testing is not the same thing as limiting development to four approved corporations. Liability is not the same thing as creating a regulatory moat that only trillion-dollar companies can cross. International monitoring is not the same thing as pretending an international declaration has caused strategically valuable knowledge to cease existing.

We should regulate the frontier, but we should be very careful about accidentally transferring ownership of it.

The Real Choice

The debate over artificial intelligence is often presented as a choice between acceleration and restraint.

That may already be the wrong question. Once knowledge exists, it is extraordinarily difficult to contain, and AI research will continue somewhere. Someone will push the next experiment forward, accumulate more powerful capabilities, or decide that the strategic reward is worth violating whatever agreement everyone else has signed.

The meaningful question is therefore not whether artificial intelligence continues to develop; it is under what conditions, and in whose hands.

One possibility is an unruly frontier: competitive, dangerous, innovative, difficult to regulate and populated by many actors capable of challenging one another.

Another is a far more orderly world in which advanced artificial intelligence becomes the province of a tiny number of corporations and governments, operating systems so expensive, restricted and strategically valuable that meaningful competition becomes impossible.

The first looks frightening. The second might look reassuringly civilized. But history should make us suspicious of that reassurance.

We have spent centuries developing antitrust law, constitutional checks and balances, competitive markets, free inquiry and divided political authority for a reason. We learned that concentrated power does not become harmless simply because the people exercising it are competent.

Artificial intelligence should not cause us to forget that lesson precisely when the stakes become enormous.

The challenge is not to preserve the Wild West forever; it is to civilize the frontier without turning it into a company town. Because if the knowledge cannot be stopped, then a successful prohibition may not prevent the future we fear. It may simply determine who gets to own it.

Technology & National Boundaries: A Civilization Mismatch

 

One of the stranger realizations that emerges from studying Big History and complexity theory is that technological progress and social maturity do not necessarily move at the same speed.

In fact, they often appear to move at dramatically different speeds.

Humanity can map distant galaxies, sequence genomes, and train large language models on significant portions of civilization’s accumulated knowledge. At the same time, it remains perfectly capable of organizing itself around tribal loyalties, centuries-old grievances, status competitions, and disputes whose origins predate the printing press.

This creates a peculiar form of cognitive whiplash.

On one scale, we inhabit a civilization of astonishing sophistication. On another, we remain a species of highly social primates navigating incentives, identities, and narratives that would have been recognizable to our ancestors thousands of years ago.

The contradiction is only apparent. Both realities are true simultaneously.

Scott Page would likely describe this as a consequence of complex adaptive systems operating on multiple timescales. Technologies can evolve rapidly while institutions, cultures, and governance structures adapt much more slowly. New layers of complexity emerge long before older layers disappear.

The result is a civilization where the props often feel futuristic but the setting still looks archaeological.

Bronze Age instincts coexist with medieval identities, industrial institutions, global communication networks, and frontier artificial intelligence. The layers accumulate faster than they are replaced.

This observation becomes especially relevant when discussing AI.

Many current debates assume that the primary challenge is technical: building capable systems, ensuring safety, increasing performance, and managing deployment. Those are important concerns. Yet an equally important question sits beneath them:

What happens when technologies begin operating at a civilizational scale while governance remains organized around nations?

The mismatch is difficult to ignore.

The training data used by advanced AI systems is not American knowledge, Chinese knowledge, or Argentine knowledge. It is the accumulated symbolic residue of civilization itself: languages, books, scientific papers, software repositories, journalism, philosophy, art, documentation, and billions of human interactions flowing across borders.

The resource is transnational.

The disruption is transnational.

The governance remains national.

Which is a bit like discovering a new continent and then insisting the most important question is which municipal office should process the paperwork.

And that would be manageable if nations themselves behaved like mature participants in a coordinated planetary project. Unfortunately, we often seem determined to prove otherwise.

We can build systems that synthesize the knowledge of billions of people, yet we still struggle to cooperate across borders, parties, regions, and identities. Not because the problems are always impossibly complex, but because incentives, prestige, short-term interests, and the occasional outbreak of political chiquitaje remain remarkably durable features of human affairs.

There is something profoundly puzzling about it.

A species capable of contemplating the origins of the universe can still become hopelessly divided over symbolic disputes, procedural squabbles, and status contests that, viewed from sufficient distance, look suspiciously small. We no longer argue about the exact same goats that wandered into the neighboring field centuries ago, but we continue to manufacture functional equivalents with impressive creativity and enthusiasm.

Meanwhile, greed has not exactly retired from public life. New technologies arrive, new fortunes emerge, and many leaders discover once again that thinking in terms of the next election cycle, the next quarterly report, or the next personal advantage feels more natural than thinking at the scale of civilization. Not always. But often enough to matter.

The challenge is not that humanity lacks intelligence.

The challenge is that intelligence scales faster than wisdom, and capability scales faster than coordination.

Politicians naturally propose national solutions because nations are where political power resides. Taxation, regulation, ownership structures, and redistribution mechanisms all operate through existing states. Senator Bernie Sanders’ proposal to tax extraordinary AI-driven gains and return a portion of the benefits to the public deserves to be taken seriously in this context. It recognizes something many observers across the political spectrum are beginning to notice: AI systems derive value not only from private investment but also from a vast reservoir of collective human knowledge.

That insight is laudable.

It may even point toward a reasonable path for ensuring that the benefits of increasingly capable systems are shared more broadly rather than concentrated narrowly.

But here comes my “but.”

Even if Sanders’ proposal were implemented perfectly, it would still confront the deeper challenge that the systems themselves operate across borders while the mechanisms for redistribution remain tied to individual nations. A national dividend may help address national consequences. It does not fully answer the civilizational question.

This creates a peculiar asymmetry.

A sufficiently powerful AI system may affect labor markets in dozens of countries simultaneously. It may be trained on knowledge generated by people across the globe. The servers may sit in one jurisdiction, the investors in another, the users in hundreds more. The benefits and disruptions spread through a planetary informational network largely indifferent to political borders.

A similar mismatch appears in public health. We often discuss outbreaks in distant countries as though Marco Polo had just arrived in Venice with alarming tales from a land beyond the edge of the known world. The fact that a pathogen can now cross continents faster than Marco Polo crossed a village somehow does little to diminish that feeling. We continue to treat many global health threats as though they were unfolding on Uranus rather than within the same densely connected civilization we inhabit.

The atmosphere does not care where a molecule originated. Viruses do not carry passports. Increasingly, informational systems appear equally indifferent to national borders.

This does not mean nation-states become irrelevant. Governments still regulate, tax, negotiate, and enforce. Companies remain subject to laws. Infrastructure exists in physical places. Reality eventually cashes out into jurisdictions.

But the scale mismatch remains.

The problem is civilizational.

The available tools are largely national.

Even if every country implemented excellent policies tomorrow, the deeper question would remain unresolved.

Who owns the products of collective learning?

That question is far stranger than it first appears.

AI systems are built using private capital, private engineering, and private risk-taking. Yet they are also built upon public research, open-source software, scientific knowledge, language itself, and centuries of accumulated human culture.

The training corpus looks suspiciously like a civilization-scale commons.

This is why arguments about ownership feel different in the AI era than they did in previous technological revolutions. The debate is no longer only economic. It is epistemic.

Who owns the systems that increasingly mediate knowledge, interpretation, memory, explanation, and attention?

That question begins to sound less like a debate about factories and more like a debate about libraries, universities, communication networks, and the informational infrastructure through which societies think.

Unfortunately, history offers little reassurance that extraordinary capability automatically produces wise outcomes.

A civilization can become extraordinarily capable while using both humans and machines in surprisingly stupid ways.

The Roman world produced remarkable engineering while remaining trapped in recurring political dysfunction. The Industrial Revolution transformed productivity while tolerating extraordinary human misery. The internet connected billions of people and then devoted a meaningful portion of its capacity to outrage optimization.

There is no law stating that intelligence, capability, and wisdom must increase together.

Indeed, they often do not.

The future may not resemble the clean technological trajectories imagined by either utopians or doomers. It may instead resemble a civilization becoming progressively more capable while struggling to coordinate around the consequences of its own success.

A civilization that can train frontier AI systems while remaining politically fragmented.

A civilization that can model climate systems while arguing about basic facts.

A civilization capable of mapping exoplanets while still becoming trapped inside local incentive structures.

And perhaps, if we are being honest, a civilization capable of generating endless new disagreements even after solving some of the old ones. If ancient cities could spend generations arguing over whose goat wandered into whose field, modern societies can certainly invent equally passionate disputes over algorithms, data rights, and digital borders. The names change. The coordination challenge remains.

This is not necessarily a sign of failure.

It may simply be the normal condition of complex adaptive systems.

The truly remarkable fact is not that humans remain tribal, emotional, and imperfect. The remarkable fact is that they have managed to build global systems of cooperation despite those limitations.

Perhaps that is the real lesson of collective learning.

Humanity was never required to become wise before becoming powerful.

It only had to become coordinated enough.

Whether wisdom eventually catches up remains an open question.