Long before academics and pundits began mourning the loss of the Habermasian public sphere, Walter Lippmann, America’s dean of journalism, questioned whether such a sphere had ever existed. Writing in the 1910s and 1920s, Lippmann argued that the very notion of a civic society capable of understanding and advocating for the common good was a fantasy. The public was a “phantom” and society a “bewildered herd,” with individuals too trapped in their subjective experience, too reliant on the distorted pictures of the world furnished by mass media and politicians, to govern themselves in any meaningful sense. Democracy couldn’t mean collective self-rule, according to Lippmann; it had to mean periodic ratification of decisions by experts who knew what was good for everyone else. Lippmann’s contempt for the masses would go on to underpin a certain way of organizing our society’s information environment: someone had to enact the “manufacture of consent.” The mass media did so, as Edward Herman and Noam Chomsky later outlined, through structural filters — concentrated ownership, reliance on official sources, the systematic privileging of elite perspectives — that ensured Lippmann’s favored sort of democracy flourished. 

John Dewey, the pragmatist philosopher and public intellectual who critiqued Lippmann in the pages of The New Republic in the 1920s, before publishing his full reply in The Public and Its Problems (1927), took issue with this pessimistic architecture. Properly functioning democracy, Dewey thought, wasn’t prevented by incompetent publics, but by broken infrastructure for determining and articulating their interests. Publics, he argued, formed around shared consequences. When the actions of some people produced effects that spilled over onto others, those affected had the basis for a common interest and, potentially, collective action. But this process required institutions that helped people identify those shared consequences, trace their causes, and recognize themselves as a public with reason to act. Dewey thought the press could be transformed into such an instrument — that it could evolve beyond an organ for elite opinion-formation to instead practice genuine social inquiry and report on the actual consequences of industrial and political decisions in ways that allowed communities to find each other. Where Lippmann saw knowledge as a resource for experts to wield on behalf of populations, Dewey saw it as a tool for communities to organize around. The formation of the herd, in Dewey’s telling, wasn’t natural — it was the result of arrangements that kept people cut off from the information that might reveal their commonalities.

It is no secret whose ideas won out. The press that Dewey hoped could enable public formation was instead consolidated to prevent it. The newspaper did not become an instrument of inquiry; it manufactured consent for a status quo that prioritized profit at almost any cost and condoned war for almost any reason. The fourth estate never belonged to the public, and it had always been in the business of managing populations, not enabling deliberation or political participation. The spaces that typified the Habermasian public sphere — town halls, newspapers, coffeehouses — excluded most of the population while congratulating themselves on their openness.

Today, what people confuse for a digital public sphere is neither open nor clearly delineated — it’s largely private and entangled. What we see is less a sphere than a pyramid. At the bottom, group chats and direct messages across various communication services create digital enclaves where people actually form views. Above that, the platform layer: where slivers of private talk get surfaced, largely by algorithmic mediation. This layer establishes via volume what feels urgent and generates memetic shorthand that institutions scramble to address. Next is the institutional layer — newsrooms, courts, legislatures — where arguments get established as serious, then become decisions and budgets and contracts. What matters is which information moves through the layers, controlled by whom, toward what ends.

Facebook’s promise to “bring the world closer together” and Twitter’s self-mythology as a “global town square” are marketing exercises that borrow democratic prestige to dignify websites concerned with generating revenue and directing the herd. When Facebook built a tool that it suggested employers could use to suppress labor organizing content, or deployed algorithms that systematically steered vulnerable users toward extreme content, or promoted content that helped to spur the Rohingya genocide in Myanmar, the platform was performing as expected. When Elon Musk started visibly boosting ideological allies (read: amateur phrenologists and spurned lovers of apartheid), apparently suppressing enemies, and seemingly adjusting algorithms by whim, he was not corrupting some neutral edifice. He made the governance question legible to people who’d preferred not to see it. The landlord had always been there.

It would be too easy — not to mention false — to conclude simply that platforms aren’t a public sphere. Institutions are still constrained by what circulates online, and opinions still make it from the group chat to the letter of the law. But this process is structured by advertising and data firms that surveil populations and segment them for engagement. According to the Herman-Chomsky propaganda model, advertising functioned as a disciplining filter that played a key role in the public sphere’s constraints on democratic deliberation. But in the transition to digital, the old advertising filter has metastasized into something Chomsky and Herman did not predict. Where concentrated ownership once limited the range of acceptable opinion across a few dozen major outlets, algorithmic ranking and engagement optimization now perform the same function across billions of individual feeds, calibrating what each person sees to maximize time-on-platform rather than informed citizenship. Social media platforms don’t produce a common public; they produce competing sub-publics, each with its own sense of what everyone knows. To answer the question Habermas never adequately posed: the public sphere, such as it is, functions for those who own it. 

Now, A.I. threatens to enclose what’s left of our threadbare epistemic commons — the bottom-level online communication where people still figure out what they think, together. A.I. agents require near-total access to users’ digital lives. The pyramid itself is being restructured: data from the digital enclaves where people form views feeds directly to the institutional layer, bypassing the intermediate spaces where those views might have been contested or refined. A.I. agents become new valve operators, stimulating consumption, monitoring behavior, determining what surfaces and what gets summarized. Dewey’s condition for public formation — communicative infrastructure that helps affected people find each other — grows more distant by the year.

It’s tempting to follow Dewey’s lead and search for fixes: federated protocols, decentralized alternatives that might allow a public to form. But technical configurations can’t substitute for politics. You cannot build your way to a public if the political economy is designed to prevent one. It’s equally tempting to catalog the failures of online communication while pretending we’re not ensnared in it. But there is no way to have a relationship to the online world in modern life without being implicated; you can delete an app, but your labor market, your housing market, your information environment, the institutions that govern your life are all shaped by these systems, whether you use them directly or not. The honest position is more uncomfortable: these layers are still where discourse happens. People stay because alternatives don’t exist at scale — and scale, as much as we may loathe that word and the way it’s deployed by our tech overlords, is what makes a public public.

The oligarchs who own our digital apparatus suffer no confusion about what it’s for. The rest of us labor under a delusion that undermines our ability to understand the present moment and chart a course forward. Dewey’s faith in the potential of institutions that treat democratic knowledge not as a problem to be managed but as a capacity to be cultivated is harder to carry right now. But the alternative is accepting Lippmann’s premises wholesale, conceding that the only question worth asking is who manages the herd and how.