Blog · Review Essay · Last reviewed June 16, 2026

The Mode of Information and the Database Subject

Mark Poster's The Mode of Information is a pre-web media-theory book that now reads like a map of the ground under AI systems. Its central claim is not that computers make communication faster. It is that electronic mediation changes how people, institutions, language, and reality are assembled. Databases do not merely store subjects. Electronic writing does not merely transmit meaning. Computer-mediated language helps create the persons and worlds it appears to describe.

The useful term is database subject: the person translated into fields, identifiers, scores, permissions, histories, and traces that institutions can search, rank, join, sell, correct, deny, or automate. AI does not invent that subject. It gives the database subject a conversational surface and, increasingly, an action layer.

The Book

The Mode of Information: Poststructuralism and Social Context was published by the University of Chicago Press in 1990. BiblioVault lists the Chicago edition with cloth ISBN 9780226675954 and paper ISBN 9780226675961; Google Books lists a 1990 Polity Press edition at 179 pages, with ISBNs 0745603262 and 9780745603261. WorldCat identifies the book as a 1990 English print book by Mark Poster, published in Chicago by University of Chicago Press. Internet Archive's library record gives the same author, year, publisher, subject areas, and Chicago ISBNs.

Poster was a historian and media theorist at UC Irvine. The University of California Academic Senate memorial identifies him as Professor of History and Film & Media Studies, a founding chair of UCI's Film & Media Studies department, and a major figure in the study of media culture, including television, databases, computing, and the Internet. That institutional location matters. Poster is not writing as a product analyst or computer scientist. He is writing as a critical theorist trying to understand why electronic media demand new accounts of language, power, and subject formation.

The book's structure is explicit. BiblioVault's table of contents lists chapters pairing TV advertising with Baudrillard, databases with Foucault, electronic writing with Derrida, and computer science with Lyotard, after an opening chapter on postindustrial society. That sounds abstract, and often is. But the abstraction is aimed at a concrete question that still matters: what happens when communication systems stop being channels and become environments that constitute social life?

The Mode

Poster's title plays against Marx's mode of production. Instead of treating industrial labor and ownership as the only way to understand late capitalism, he asks how social relations are reorganized when communication is mediated by electronic systems. The move is not a replacement of economics by language. It is an attempt to see that information systems become part of production, power, identity, and memory.

The useful point is that electronic mediation changes more than speed or reach. It changes address. A phone call, database record, television advertisement, bulletin board post, command line, or computer file does not simply carry a preexisting self from one place to another. It asks the self to appear in a format: caller, viewer, consumer, suspect, account holder, data subject, user, author, operator, record, profile.

That definition keeps the book from becoming only a period piece about early computing. A mode of information is the social arrangement in which people must become addressable by information systems before institutions can recognize them. In 1990, that meant files, databases, phones, television, terminals, and early networked writing. In 2026, it means identity graphs, customer-data platforms, vector stores, learning records, risk scores, model memories, audit logs, permissions, prompts, and agent traces.

This is why the book belongs beside Discipline and Punish, The Digital Person, My Mother Was a Computer, and Sorting Things Out. It gives media theory a way to see the person as something made through records, categories, interfaces, and the language games that institutions require.

Databases and Subjects

The strongest AI-era chapter is the one on Foucault and databases. Poster is interested in the database not only as storage, but as a system for making people knowable. A prison file, credit record, census entry, personnel system, customer profile, welfare case, medical record, school record, or police index does not merely describe a person after the fact. It selects attributes, stabilizes categories, enables comparisons, and lets institutions act at a distance.

The important twist is participation. Modern data systems often need the person to help produce the record. People fill forms, update profiles, choose categories, accept defaults, generate usage traces, correct errors, confirm identities, submit documents, click prompts, answer security questions, and adapt to the categories that will later govern them. The subject is not only watched. The subject is recruited into becoming machine-readable.

Poster gave this arrangement a name: the Superpanopticon. Where Foucault's panopticon disciplined bodies confined in space, under a tower whose gaze might always be watching, Poster argued that the database achieves the same effect with no walls, towers, or guards at all. It needs only traces: credit-card activity, phone bills, loan applications, welfare files, library records, traffic tickets. And unlike Foucault's prisoners, the data subject helps build the cell, because, as Poster put it, "surveillance by means of digitally encoded information constitutes new subjects by the language employed in databases." The line matters for the chapters that follow. If the database makes people through its language, then a system that now generates language at scale is not a neutral upgrade to the Superpanopticon. It is the same machine learning to write its own records.

That pattern is everywhere in contemporary AI. A hiring model inherits resumes and performance records. A school model inherits assignments, attendance logs, proctoring signals, and learning-platform behavior. A welfare system inherits case files and verification documents. A companion chatbot inherits disclosure. An enterprise assistant inherits chat histories, tickets, documents, permissions, and corrections. The model may be new, but the database subject was already there.

The sharper governance point is that a database subject is not identical to the person. It is an institutional proxy with a maintenance problem. It can be stale, partial, adversarially generated, wrongly joined, over-retained, over-shared, or made from categories the person cannot refuse. When an AI system uses that proxy to summarize, rank, infer, or trigger an action, it inherits the proxy's violence and its convenience at the same time.

This also explains why privacy language can feel too narrow. The problem is not only whether information is hidden or exposed. The problem is whether a record becomes a working version of the person. Once a profile routes service, risk, price, suspicion, opportunity, or attention, the database is part of the social body. It is not just about you. It becomes one of the places where you are made actionable.

Electronic Writing

Poster also sees electronic writing as a change in the status of language. Text on a screen is editable, networked, copied, transmitted, stored, searched, recombined, and detached from a stable page. In 1990, that meant word processors, conferencing systems, databases, electronic files, and early networked communication. Today it means prompts, generated answers, collaborative documents, chat transcripts, embeddings, model memory, autocomplete, agent logs, and text that triggers tool calls.

The AI-era lesson is that language is no longer safely downstream of action. A phrase can retrieve records. A prompt can call a tool. A summary can update institutional memory. A model-generated note can enter a medical chart, customer file, ticket, contract draft, risk register, or court record. An instruction can move money, send a message, change a database field, or open a door to another system.

That makes The Mode of Information useful for reading AI interfaces. The danger is not only that generated language may be false. It is that generated language may be operational. A model's sentence can become the next record, and the next record can become the source for the next model. Once language is stored and reused by machines, words become infrastructure.

The AI Reading

Poster was not writing about large language models. He was writing before the consumer web, before search became the main public interface to knowledge, before social media trained billions of people to live inside feeds, and before models could generate fluent prose on demand. The book's value is precisely that it explains the substrate those systems inherit.

AI systems need information subjects. They need users, records, labels, histories, embeddings, accounts, roles, permissions, examples, preferences, corrections, evaluations, and feedback. They also need language to be treated as computable material. A prompt is not just expression. It is input. A document is not just reading matter. It is context. A label is not just description. It is training signal. A generated answer is not just speech. It is often a proposed institutional action.

This changes how "human in the loop" should be read. The human is not an outside sovereign who occasionally approves machine output. The human is often already inside the information system as user, subject, annotator, exception handler, source, customer, worker, patient, student, claimant, or risk object. Governance cannot begin at the final approval click. It has to begin where the person is first translated into data and addressable by the system.

Recursive Reality

The book also clarifies recursive reality: the process by which descriptions produced by information systems reshape the world that those systems later describe.

A database category changes how a person fills a form. A recommendation changes what a user watches. A risk score changes where inspectors look. A model summary changes what a manager remembers. A generated FAQ changes what customers ask next. A school dashboard changes what teachers document. A legal search result changes which precedent feels available. A companion chatbot changes what a user discloses, and the disclosure changes the model's next response.

The old distinction between representation and reality starts to leak. The system records behavior, then routes behavior, then uses routed behavior as evidence. It stores a description, then people adapt to the description, then the adapted world confirms the system's categories. This is why the database chapter still matters. The record is not passive memory. It is a lever.

Institutions and Labor

Poster's strength is language, subjectivity, and mediation. That is also his limit. The book is less interested in procurement, organizational incentives, labor conditions, infrastructure, environmental cost, or the political economy of who owns the systems. Those questions now have to be brought back in.

A database does not appear by theory alone. Someone designs the schema, buys the vendor product, cleans the data, resolves duplicate identities, writes the policy, pressures the worker, answers the complaint, audits the system, or refuses to fund an appeal path. A model-mediated institution is built through budgets, contracts, workarounds, legal constraints, staff shortages, managerial desires, and maintenance labor.

That is why Poster should be read with books such as Automating Inequality, Atlas of AI, The Costs of Connection, and Ghost Work. He helps name the subject produced by information systems. Those other books help locate the extraction, bureaucracy, labor, and material systems that make the subject useful to power.

Governance and Safety

As of June 16, 2026, the database subject is no longer only a media-theory diagnosis. It is also a legal and operational governance problem. The European Data Protection Board's automated decision-making and profiling page records that the EDPB endorsed the GDPR-related WP29 guidelines on May 25, 2018. Those guidelines sit behind the practical demand that people affected by profiling and automated decisions need information, safeguards, and contestability, not merely a privacy notice no one can act on.

The Court of Justice of the European Union sharpened the point in its February 27, 2025 press release for Dun & Bradstreet Austria, Case C-203/22. The Court said an automated credit-assessment subject is entitled to an explanation that lets them understand and challenge the decision, and that the controller must describe the procedure and principles actually applied so the person can understand which personal data were used and how. In plain governance terms: a database subject needs enough explanation to contest the record that is being used as them.

The EU AI Act adds system duties around the same problem. Article 10 requires high-risk systems using training, validation, or testing datasets to address data origin, preparation, assumptions about what the data measure, bias examination, mitigation, and data gaps. Article 12 requires high-risk systems to allow automatic recording of events across their lifecycle so operation can be traced, monitored, and reviewed. NIST's AI Risk Management Framework similarly makes context, impacts, knowledge limits, third-party data, and monitoring part of the risk map.

The safety implication is concrete. If a deployed AI system acts on database subjects, it needs records that answer at least these questions: who is represented, which fields and identifiers define the representation, where the data came from, what purpose justified collection, which joins and inferences were made, what model or workflow used them, what action followed, what logs preserve the event, and how the person can correct, challenge, or escape the frame. A system that cannot answer those questions may still be efficient, but it is not answerable.

Where the Book Strains

The Mode of Information is not an easy book. Its theoretical machinery is dense, and its examples sometimes feel like they are being pulled toward a master concept. Charles J. Stivale's 1991 review in Criticism is useful because it treats the book as important while also questioning whether the mode of information becomes another totalizing framework. That criticism holds up. The concept is powerful when used diagnostically; it becomes weaker when asked to explain everything.

The book is also dated in ways that matter. The Internet arrives mostly as horizon rather than settled environment. Smartphones, social media, cloud platforms, search engines, recommender systems, biometric identity, vector databases, generative AI, and agentic tool use are not the objects of analysis. Readers have to translate from television ads, databases, electronic writing, and computer science into contemporary stacks.

The translation is worth doing, but it has to be concrete. The point is not to say that every AI interface is poststructuralism with a chat box. The point is to inspect how electronic systems create categories of personhood, move language across contexts, and turn records into institutional action.

What This Changes

The practical lesson is to audit subject formation, not only outputs.

For any AI-mediated system, ask what kind of person the system requires in order to work. Does it need a claimant, prospect, patient, student, debtor, risk, user, worker, creator, suspect, subscriber, or dependent? Which fields define that person? Who chose them? Which categories can be refused? Which records can be corrected, deleted, or contextualized? Which inferences become action? Which summaries become memory? Which actions feed the next model?

For institutions, the book sharpens a harder question: when does a database stop being a record system and become a social reality engine? That line is crossed when the stored representation changes access, obligation, suspicion, price, care, speech, or trust. AI does not create that problem from nothing. It accelerates it, makes it conversational, and hides it behind helpful language.

Poster's book remains valuable because it catches the moment when electronic mediation stops looking like a neutral channel. The database, the screen, the file, the prompt, and the model are not merely between people. They help decide what a person is allowed to be inside an institution. That is where governance has to begin.

Source Discipline

This review separates three layers of evidence. BiblioVault, library records, and UC sources support book metadata, table of contents, and Poster's institutional context. Contemporary reviews support reception and limits. Current governance claims come from EDPB, CJEU, the AI Act Service Desk, and NIST; those sources are not treated as proof that Poster anticipated present law, but as evidence that the database-subject problem now has regulatory vocabulary.

The AI-era application is an interpretation. This page does not claim that present AI systems are conscious, divine, or AGI. It treats AI systems as institutional media: systems that translate people into records, use records to generate language or action, and then write the resulting action back into the record.

Sources

Book links are paid affiliate links. As an Amazon Associate I earn from qualifying purchases.

Buy on Amazon Browse Books

Return to Blog · Return to Books