Lippmann coined 'manufacture of consent' in Public Opinion (1922) with a meaning distinct from Chomsky-Herman's later conspiratorial inflection. Lippmann's argument: consent is manufactured through the architecture of information distribution, not through coordinated propaganda. Editors select stories (space is finite), wire services compress events (telegraph charges by word), politicians simplify positions (attention is scarce). Each structural constraint introduces bias—not partisan bias necessarily but systematic deviation from completeness. Consent emerging from this process is not propaganda's product but architecture's product. The information environment has a shape; the shape shapes opinions forming within it, the way a riverbed shapes water flow—not by intention but by constraint. The AI discourse was manufactured in precisely this structural sense by manufacturers Lippmann could not have imagined: the AI industry (emphasizing empowerment narratives), the media (emphasizing drama), and the algorithmic feed itself (optimizing for engagement). No conspiracy coordinated these forces—manufacture was emergent, produced by independently operating structural incentives converging on vivid, coherent, systematically misleading pictures.
The industry's manufacture operated through emphasis rather than fabrication. Narratives of AI democratization, empowerment, augmentation were constructed from genuine evidence: the Lagos developer, Trivandrum engineers, solo builders shipping in weekends. But selection of that evidence, framing of that evidence, relentless repetition in keynotes, blog posts, orchestrated launches constituted consent manufacture operating not through lies but through emphasis. What the industry narrative structurally could not emphasize: costs. The Berkeley burnout documentation, Han's philosophical diagnosis, skill erosion engineers experienced, work intensification data showed. These facts existed in the same information environment as empowerment narratives. Industry's structural incentives ensured empowerment received keynote-level amplification while cost received footnote-level acknowledgment.
The media's manufacture operated through engagement incentives. The AI story generating most engagement was not the most accurate but the most dramatic: trillion-dollar wipeouts, engineers who could not stop, philosophers refusing smartphones, twelve-year-olds asking existential questions. Each story was real. Selection of these from the vastly larger set of possible stories—the quiet adaptation, gradual recalibration, uneventful Tuesday using tools competently—constituted consent manufacture privileging the dramatic over the representative. The public's picture was constructed not from representative samples but from curated collections of extremes.
The algorithmic feed's manufacture was the dimension Lippmann could not anticipate. The feed manufactures consent not by choosing what to say but by choosing what to show—and the choice is made not by a questionable human but by an algorithmic process whose criteria are proprietary and emergent. Even engineers who designed the system cannot fully predict the selection pattern it produces. This is new consent manufacture: the manufacturers are not identifiable (no editor, no wire service, no political leader), biases are systemic rather than personal, emergent rather than chosen, distributed across millions of micro-decisions no single human made or can reconstruct. The consent it manufactures is more difficult to recognize as manufactured because there is no manufacturer to point to, no editorial meeting where selection was decided, no political interest the selection serves in simple ways. And yet the manufacture is real.
Lippmann developed the concept from WWI experience. Governments did not need to lie to the press—they needed only to provide steady information supporting their preferred narrative while restricting access to information complicating it. The resulting coverage was not fabricated but manufactured: assembled from genuine materials, organized by structural incentives, presented to the public as complete pictures of selectively represented reality. The phrase 'manufacture of consent' appeared in Public Opinion Chapter 15 ('Leaders and the Rank and File'), describing how public opinion is organized rather than how it spontaneously forms.
The concept's appropriation by Edward Herman and Noam Chomsky in Manufacturing Consent (1988) gave it more conspiratorial inflection—a model of deliberate propaganda. Lippmann's original meaning was structural, not intentional: the manufacturing happens through the built environment of information distribution, through incentives and constraints operating independently of any coordinating intelligence. The distinction matters for the AI moment: the discourse is being manufactured, but not by a cabal—it is being manufactured by structural forces (industry incentives, media incentives, algorithmic incentives, cognitive stereotypes) operating simultaneously, producing confident polarized pictures of an ambiguous evolving reality.
Structure, not conspiracy. Consent is manufactured through the architecture of information environments—editorial selection, compression requirements, speed incentives—independently of deliberate coordinated propaganda.
Three manufacturers. The AI industry emphasizes empowerment, the media emphasizes drama, the algorithmic feed optimizes engagement—each selecting from the same pool of genuine facts, each producing different pictures through different structural biases.
Emphasis over fabrication. The most effective consent manufacture uses real evidence, selecting and framing it according to institutional incentives that diverge from accuracy—producing pictures that are not false but are radically incomplete.
Emergent, not coordinated. No one decides which demonstrations, horror stories, statistics, testimonials to combine into which pseudo-environments. The assembly is produced by convergence of independently operating structural forces.
Algorithmic manufacture is opaque. When the selector is an algorithm whose criteria are proprietary and emergent, the manufactured quality of the resulting picture becomes nearly impossible for users to recognize—no editor to question, no bias to map.