Structured Absence

Abstract

This essay argues that while generative artificial intelligence (AI) can be described as a platform, that description is analytically insufficient for understanding where power operates and how it displaces democratic accountability. The platform concept directs attention to the interface—application programming interfaces (APIs), app stores, and developer ecosystems—while the decisive conditions of generative AI lie upstream in compute supply chains, mineral extraction, semiconductor chokepoints, energy and water infrastructures, immigration flows, and corporate-state arrangements that concentrate control. To name this formation, I introduce the concept of structured absence: the organized diminution of democratic institutional capacity in which governance persists while public authorization disappears. Generative AI is not ungoverned; it is extensively governed through export controls, industrial policy, labor regimes, and private infrastructures that remain largely inaccessible to democratic publics. I argue that “predatory inclusion” and authorization are central analytic terms for understanding this shift: communities are incorporated into AI systems on unequal and opaque terms, while the standing to deliberate over those systems is withheld. The essay concludes by calling for democratic infrastructures capable of authorizing technological change.

ChatGPT is a platform. That may be the least important thing about it.

OpenAI built an application programming interface (API), launched a GPT store, cultivated a developer ecosystem—the recognizable architecture of a two-sided market. Platform studies know how to study all of this. The analytical vocabulary developed over the past fifteen years—from Gillespie’s (2010) insight that “platform” was always a strategic corporate claim, to Srnicek’s (2016) account of platforms as business models built on data extraction, network effects, and cross-subsidization, to Nieborg and Poell’s (2025) analysis of how platforms generate institutional relations of dependency—can be extended, with modification, to describe the interface layer of generative artificial intelligence (AI). The platform concept applies to the generative AI turn. Herein lies the problem. The easy applicability of “platform” to generative AI risks directing attention to the surface of a power structure whose decisive features lie elsewhere.

What matters most about generative AI are not questions about its digital distribution network, such as the GPT store, but everything beneath it: supply networks dependent on exploitative mineral extraction, energy and water resources under extraordinary strain, talent pipelines shaped by revanchist immigration policy, geopolitical contests over semiconductor exports, and entwined corporate-state structures designed to concentrate control. Consider the semiconductor chokepoint itself: the extraordinary concentration of advanced chip fabrication in firms like the Taiwan Semiconductor Manufacturing Company (Miller, 2022) alongside U.S. export controls over high-end semiconductors and NVIDIA’s dominance in compute, means that generative AI is governed long before a model, tool, orsystem reaches the public. Its conditions of possibility are shaped upstream through value chains, industrial policy, and geoeconomic strategy rather than through app stores or content moderation rules. ChatGPT is legible as a platform. The network of power that makes ChatGPT possible is often indistinct.

Suchman (2007) argued that technical systems always configure their users—scripting assumptions about who they are and what they need. Expanding this insight, our studies of platforms have been similarly configured by their object. Because social media presented itself as regulable—U.S.-based companies, identifiable executives who could be put under oath, declared yet dynamic terms of service, a statutory framework to contest, and congressional hearings—the field was primed to ask regulatory questions. These were often the questions the platform firms wanted us to ask (Klonick, 2019). This did not ultimately make social media transparent, only more publicly contestable: there were visible sites of conflict, identifiable actors, and institutional venues in which claims could be made. Generative AI does not present itself as governable in this same way. Its governance is declaratively eschewed yet substantively underway—in arenas that platform studies and social media scholars have barely examined and democratic publics can barely access. As we argued in “Five Theses on the Gravity of Platforms,” AI is a “platform whose weight sucks so many things into its orbit, scales so tremendously it exceeds our capacity to address it using older methods and metaphors” (Aidinoff et al., 2024; Nelson, 2026a). The response is not retreat but ambition: “Platform production,” we contend, “has become much more sequestered and private, therefore, we need to compensate for this by being theoretically more adventuresome, pushy, and big.” What was social media? The field needs new concepts (Nelson, 2026a).

One of the most important forms of power in the generative turn is exercised neither through formal regulation nor through direct control over regulatory institutions. It operates through the absence of institutional accountability— an absence that is announced even as control is exercised through other levers of the AI value chain: compute, export controls, standard-setting, security review. This is structured absence. This is the organized diminution of democratic institutional capacity, in some areas, alongside its structural reconstitution in others. ¹ The erosion of the Federal Trade Commission’s technical capacity, the defunding of the Office of Technology Assessment, the decline of civic associational life, and the hollowing out of public science infrastructure—these absences preceded the platform era. ² The institutional void was already there. Platforms expanded into it, generative AI is deepening it. In this void, advantage and disadvantage are distributed by those who occupy it, interests are served and foreclosed, futures are enabled and extinguished. Structured absence heightens inequality.

Structured absence is not accidental. It is maintained—not by the absence of governance but by its proliferation in forms that are unaccountable and insulated from democratic contestation. Social media platforms, for all their failures, were at least legible as sites of governance. Publics could identify the company, name the executive, contest the policy, and invoke a statute. The generative turn has reorganized this entirely. Its infrastructure is organized around chokepoints—in processing, fabrication, and computing—where power concentrates and selective governance already operates. But it operates through mechanisms available to geopolitical elites and largely inaccessible to democratic publics (Nelson, 2026b). As with the self-regulating market of the nineteenth century—laissez-faire was planned—the mirage of deregulation is itself the political achievement: the active construction of a domain that presents itself as beyond governance while being intensively governed by other means (Polanyi, 1944). Regulatory arbitrage and industry ethics boards, transparency reports, and voluntary commitments fill the perceptual space where democratic regulatory institutions might otherwise be demanded (Nelson, 2025b). The institutional void persists not because generative AI is ungoverned but because the governance that exists endeavors to make democratic institutional capacity appear unnecessary—or unimaginable. This is what distinguishes the generative turn from the platform era: not the scale of the technology but the disappearance of arenas, even imperfect, in which its organization can be democratically contested.

What does social theory owe to this moment? Platform studies narrated disruption: platforms disrupted media, democracy, and labor markets. This narrative must itself be disrupted (Nelson, 2025b). Taylor’s (2019; also Cottom, 2020) theory of “predatory inclusion” helps clarify why. Communities are included in generative AI—as training data, as subjects of automated decision, and as the low-wage labor that makes the systems function—on terms that are deliberately opaque, shielded by trade secrecy and technical complexity. In this sense, predatory inclusion is one mechanism through which structured absence operates: people are incorporated into systems they cannot meaningfully contest because the institutional capacity to demand accountability has already been diminished. Disruption asks, what did the technology break? Predatory inclusion asks, who was incorporated, on what terms, and to whose benefit?

One important strain of the study of platforms has been organized around governance: rules, oversight, regulation. By that accont, generative AI is actively governed as a platform. But it is almost entirely unauthorized as a political formation. If governance asks how technology should be regulated, authorization asks who has the standing to decide and whether a decision carries democratic legitimacy or merely administrative force. Authorization is the political consequence of structured absence: when democratic institutions are diminished, decisions persist, but public standing disappears.

The debate over technology governance—from platform regulation to generative AI policy—is trapped in a false binary: regulate or defer. Both sides assume the existing repertoire of institutional forms is the only one available. Every proposal reproduces a familiar template—the regulatory agency, the advisory board, the transparency report—because these are the forms the field knows how to think. What is needed is institutional imagination: the capacity to envision forms of shared authority that do not yet exist. Not advisory boards but civic epistemic infrastructure with genuine investigative and legal authority. Not consultation but shared governance that distributes decision-making power. But institutional imagination must reckon with its own conditions. The populations most subject to algorithmic governance—in housing, hiring, policing, and labor—are also those with the fewest resources to contest it. What structured absence degrades is not only democratic standing but also the capacity to reason about and shape the conditions of one’s own life. Institutional forms that do not account for this are democratic in principle and inaccessible in practice. The task is not to imagine institutions that ought to exist but to build ones that can function in the world structured absence has already made.

The generative turn is not the next chapter of the platform story. It is a rupture that demands new terms of analysis. The platform model applies to generative AI—and in applying, it directs attention to the interface while power operates beneath it. The task is not to regulate what exists but to build what does not, in a void and under conditions that make building the most difficult for those who need it most.

For a decade, Social Media + Society has described platform power with considerable sophistication. That work is not over. Platforms remain consequential, and the social inequalities they produce remain urgent. But the field must now also learn to study absence: the structured absence of democratic institutional capacity, the missing authorization, and the institutional forms not yet built. This means theorizing not only what AI platforms do but also what is not there to contest them—and who benefits from its absence. It means specifying what does not exist but must be built, and insisting, against the prevailing narratives of inevitability, that democratic publics retain the authority to decide what kind of technological society they inhabit. That is what social theory owes to this moment.