On AI, Power, and the Question of Public Agency

AI ethics debates focus on constraining automated decisions. They rarely ask who gets to wield AI as a political force. What is needed is not better governance of AI's outputs, but public ownership of the infrastructure that produces them.

Old Regime: Capital as Gate to Action

In industrial and early mass-media capitalism, owning capital meant owning the means to act in the world: factories, newspapers, broadcast channels, and the organizational capacity to convert money into durable power. Capital was a precondition for shaping the public sphere.

But no single actor could simultaneously control all the channels through which political communication moved. That structural dispersion left at least some cracks where counter-publics, unions, parties, and movements could wedge themselves in. Reform remained possible.

New Regime: Capital + AI + Media

The new regime closes those cracks. Not by suppressing counter-publics directly, but by controlling the infrastructure through which attention itself is organized.

The means of action directed at the public are no longer simply capital. They are capital fused with data, algorithmic infrastructure, and platform control. A small number of firms and political actors can now simulate, predict, and steer public attention at scale, minute by minute.

They do so through micro-targeted content and what might be called synthetic publics. The concept names a specific collapse: when polling, focus groups, and A/B tests are applied recursively at scale, the distinction between discovering public opinion and manufacturing it becomes operationally meaningless.

Measurement feeds directly back into targeting; targeting shapes what is measured. That fusion collapses the distance between owning assets and directly shaping the mental frames through which people make political choices. Political power and private capital have merged.

The public, as a collective actor, is no more.

What the Current Debate Misses

Much writing on AI-enabled harms focuses on how destructive specific applications are. It rarely identifies the political subject acting through these tools. That omission leads, almost inevitably, to normatively thin prescriptions: better regulation, ethics boards, expert oversight.

Each addresses symptoms rather than root causes. Each also presumes the relevant agent is either the state or the firm. None addresses the broader public as a collective actor with a legitimate claim to direct how these technologies are built and deployed.

Re-centering the Public as Actor, Not Audience

The real lost alternative is not a better-regulated AI. It is the public's capacity to act on and through the techno-economic infrastructure, rather than merely being shielded from its effects.

The public need not be morally good in any liberal-virtue sense. Whatever it actually is, we have to live with it. No external authority can legitimately set a better public against the existing one and govern in its name.

That move — invoking a higher conception of the good to override actual public preferences — is available to any power that wishes to rule without consent. It is not a safeguard. It is the oldest justification for elite governance.

The question, then, is not whether the public will make good decisions once it regains agency over AI. It is whether decisions made without that agency can be legitimate at all.

The public, as it is — fragmented, contradictory, sometimes wrong — remains the only legitimate source of political direction available. The task is to restore its structural capacity to act, not to guide it toward correct conclusions.

Toward Real Public Agency

The direction is clear even if the path is not. Parts of the AI stack — foundational models, training data, critical compute — need to be held in common rather than privately owned, changing who can act rather than merely who can complain.

Publics need collective rights over how their data is used for political targeting: rights attached to communities, not individuals, with power to block particular uses, impose conditions, and demand representation.

They also need dedicated institutions through which to contest AI deployments on an ongoing basis, rather than being periodically surveyed by those who run them. These are not technically novel proposals. They are the application of familiar democratic instruments — public ownership, collective rights, institutional accountability — to a domain that has so far escaped them.

Such directions are justified not because they guarantee good outcomes, but because they restore the conditions under which governance can be legitimate at all. The public may hold these powers and still make decisions that prove wrong. That is the nature of democratic politics.

The alternative — decisions made by a small number of firms and individuals, without public mandate or accountability — is not safer. It is simply unaccountable.

Conclusion

The standard for AI governance is not moral perfection. It is the same standard we apply to every other domain of collective life: that those who are governed have a meaningful say in how they are governed.

We have not met that standard yet.

Sahand Hagi (Political Scientist, MBA and Startup Founder)