Every transformative technology arrives with a familiar question attached: who gets to control it? The printing press loosened the Church's grip on knowledge. The telegraph concentrated power in the hands of those who owned the wires. Now artificial intelligence presents us with the same ancient dilemma, dressed in new clothing.

Should the power of AI be concentrated in a handful of capable hands, or distributed across many?

The answer matters more than it might first appear. How we govern AI will shape not only the pace of innovation, but the freedom of thought itself.If we were to concentrate power in the hands of few bureaucrats, we have seen the abuse of power in the pursuit of control for a specific narrative and agenda.One can witness today the crushing of Free Speech in the UK and other countries.

Below, we explore the two competing philosophies—centralization and decentralization—and weigh what each offers, and what each costs.

1. The Case for (and Against) Centralized AI Control

Centralization is the instinct of order. It concentrates decision rights, standards, tooling, and oversight within a single authority—whether that's a corporate "AI Center of Excellence" or the dominance of a few large cloud providers like Microsoft, OpenAI, Google, and Amazon. There's a certain elegance to this approach, and history is full of moments where centralized stewardship brought stability to chaos.

The advantages are real and worth respecting:

• Consistency and control. A central authority ensures that updates, policies, and security measures roll out uniformly, so everyone benefits from the latest advancements at once.

• Concentration of expertise. Specialized talent and resources are pooled rather than scattered, which often delivers faster, more reliable results in the early stages.

• Lower fixed costs and quicker deployment. Shared tooling and approvals reduce duplication, particularly valuable in regulated, compliance-heavy environments.

• Accountability and auditability. When something goes wrong, there's a clear place to look and a clear party responsible.

Yet concentration of power has always carried a shadow, a very dark shadow and AI is no exception.History shows that the concentration of power in the Union of Soviet Socialist Republics, Chairman Mao’s China or government agencies of the West which were weaponized for their own aims.

Human beings want power, believe that their own self-righteous ethics should be imposed on others and use frameworks to institute their own agendas and narratives.

The disadvantages deserve sober reflection:

• The bottleneck problem. Central teams buckle under their own weight as demand multiplies. Massive bureaucracies create complexity not ease of use.Just think of the IRS on steroids.

• Loss of local context. Airbnb began with a fully centralized data science model, only to abandon it when the central team "lacked enough local context and became reactive"—a service desk rather than a strategic partner. Think of local matters more than national in medicine, education, and more.

• Bias and the narrowing of perspective. AI shaped by a small group of developers inevitably absorbs their blind spots, encoding the assumptions of the few into systems used by the many.We have seen plenty of examples of bias built into AI systems with their engineers hatred or prejudices magnified with their power.Wikipedia’s control of knowledge with its biases and unresponsive editorial structure for changes is a prime example of centralized control.One of the Wikipedia founders, Larry Sanger, has exposed the platforms bias and ideological capture. He wrote an article on how to return the platform to its neutral basis and fix Wikipedia.

• A single point of failure. Concentration creates fragility. One breach, one outage, one corrupted dataset can ripple across everything downstream.

And here lies the deepest concern: censorship and control. When a few entities decide what information an AI may surface, what questions it may answer, and what views it may express, they hold extraordinary influence over public thought. The monopolization of AI doesn't merely stifle innovation—it risks placing the boundaries of human inquiry in the hands of a small, unelected few. It's a quiet kind of power, and history teaches us to be wary of power that operates quietly.

One might recollect Operation Mockingbird. A large-scale program of the United States Central Intelligence Agency (CIA) that began in the early years of the Cold War and manipulated domestic American news media organizations for propaganda purposes.

Carl Bernstein who broke the Watergate story told in his article, “The CIA and the Media” (1977 Rolling Stone article) wrote that over 400 US press members secretly carried out assignments for the CIA. That’s right, the press manipulated the news for the CIA.And not just small time papers.The so-called ‘journalists’ included New York Times publisher Arthur Hays Sulzberger, columnist and political analyst Stewart Alsop and Time magazine as well as Newsweek, CBS and others.According to CIA operative and founder, Frank Wisner, he had recruited Phil Graham, publisher and editor of the Washington Post, to run the operation from inside the news industry.

Agenda, anyone?

Citizen journalism has defeated the ‘official’ narrative and bias by commission or omission many times.Even pre-Musk examples abound such as the Arab Spring uprisings which challenged authoritarian government narratives, the 2009 Iranian election protests, the Obama era NSA surveillance program overreach exposed by Edward Snowden and more.Indeed, Elon Musk’s purchase of Twitter showed internal document exposing the supression and censorship of stories like the Hunter Biden laptop story in 2020 which derailed an election topic which might have changed an American election.

Citizen journalism in the US, Canada, Mexico and Europe has tackled numerous problems including the Minnesota daycare fraud, Hurricane Helene as well as rising immigrant related crime against women and girls throughout Europe. Its use during Covid highlighted abuses, enlivened discussion about lab-leak origins and shared vaccine information.

2. The Case for (and Against) Decentralized AI Control

Decentralization is the instinct of freedom. Rather than concentrating authority, it distributes ownership to those closest to the work—individual teams, independent developers, open-source communities. It trusts that wisdom is dispersed, not hoarded, and that the people feeling a problem most acutely are often best equipped to solve it.

The advantages speak to our better aspirations:

• Innovation through diversity. A distributed system welcomes a broader range of ideas and approaches, resisting the homogeneity that creeps in when one group decides everything. “X” is a prime example of decentralized knowledge and Free Speech.Since Elon’s rescuing of X and promotion of Free Speech, the platform has highlighted Keir Starmer’s complicit inaction in the UK Rape Gang Scandal, the immigration crises and crimes across Europe, other counter narratives in many countries based upon citizen and participant curiosity.

• Resilience. With no single point of failure, decentralized networks are harder to disrupt and harder to capture.

• The democratization of AI. New economic models can put powerful tools into the hands of many, rather than reserving them for the privileged few. The primacy of immediate news countering the official narratives is off the charts with X.

• Better adoption and faster iteration. When people build the tools they themselves use, those tools actually fit. Booking.com, for instance, scaled to 150 successful customer-facing machine learning applications—built and validated by dozens of teams working independently.

But freedom, as any student of history knows, is rarely free of complication.

The disadvantages temper our enthusiasm:

• Fragmentation. Without shared standards, you end up with a patchwork of disconnected experiments, conflicting policies, and no reliable record of what's actually in production.

• Quality and security challenges. Coordinating compatibility, safety, and compliance across many dispersed systems is genuinely difficult, opening the door to "shadow AI" and undocumented data flows.

• The loss of institutional memory. When knowledge lives only with individuals, it walks out the door whenever they do.Perhaps a validation mechanism might warrant knowledge growing like Encyclopedia Galactica on X.

Decentralization, then, offers liberty at the cost of cohesion—a tradeoff humanity has wrestled with in every domain from politics to commerce.

Finding Wisdom in the Middle

A growing number of organizations are discovering that the answer is "a little of both”—sometimes called the federated model.

The principle is simple yet profound: centralize the platform, governance, and standards; decentralize the ideation, experimentation, and the building itself.

Shared foundations, distributed creativity.

Much like “X” itself with a standard structure and reporting - but millions of creators and readers.

The lesson echoes one we've learned again and again throughout history: enduring systems balance order with freedom, structure with autonomy. They are rigorous enough to be safe, yet open enough to be alive.

Humanity is on the cusp of this next revolution with Artificial Intelligence poised to alter, shift, redirect our work, education, play and creativity. Some for the good like the internet itself.

Good: Think access to knowledge from medical to traffic patterns to brainstorming a better life.

Bad:Think arguments with strangers.Information aligned by ideological capture.And the time suck of too many cat pictures.

Thinking about this structure now is key to our future and how we choose to hold the reins of artificial intelligence may well define the kind of thinkers, and the kind of society, we become. The choice is worth our deepest attention.

And the future is now.