Blog · Review Essay · Last reviewed June 25, 2026

Updating to Remain the Same and the Habit Loop of New Media

Wendy Hui Kyong Chun's Updating to Remain the Same: Habitual New Media is a theory of digital life after novelty has become routine. Its AI-era value is that it explains why the most powerful interfaces are often the ones that feel ordinary: the update prompt, the personalized feed, the notification, the profile, the phone check, the public self tuned for network circulation.

For this review, habitual media means media systems that govern by repetition rather than command: they train users to return, refresh, disclose, respond, personalize, update, and accept the platform's changing surface as ordinary life. The operational unit is the habit loop: cue, action, reward or relief, retained trace, interface adjustment, and easier repetition. The AI-era question is whether those loops expand agency or make dependence feel like convenience.

The Book

Updating to Remain the Same: Habitual New Media was published by the MIT Press in hardcover and ebook on May 27, 2016, and in paperback on August 11, 2017. MIT Press lists the hardcover ISBN as 9780262034494, the ebook ISBN as 9780262333788, and the paperback ISBN as 9780262534727. Open Library records the work under subjects including the social aspects of the internet, information society, digital media, and mass media technology.

Chun is a media theorist whose work connects software, memory, race, surveillance, networks, and political recognition. MIT Press's author note identifies her as the author of Control and Freedom, Programmed Visions, and Updating to Remain the Same; the SFU Digital Democracies Institute identifies her as Simon Fraser University's Canada 150 Research Chair in New Media and director of the institute.

The book belongs beside Programmed Visions, The Culture of Connectivity, The Twittering Machine, and Filterworld. It shifts attention from spectacular technological change to the repetitive routines through which technology becomes the background grammar of everyday life.

Current Context

As of June 25, 2026, Chun's habit argument applies directly to AI interfaces that do not arrive as a single dramatic replacement. They arrive as recurring prompts, summaries, memories, autocomplete suggestions, recommendation slots, notification digests, agent permissions, search answers, and profile settings. The user does not have to believe a grand story about artificial intelligence. They only have to repeat the small act until the interface becomes the normal route to thought, sociality, work, and self-description.

Current governance has started to regulate parts of that loop. The EU Digital Services Act's official text prohibits certain deceptive interface designs for online platforms under Article 25, requires recommender-system transparency under Article 27, and requires very large online platforms and search engines to offer at least one recommender option not based on profiling under Article 38. The FTC's dark-pattern report treats manipulative design as a consumer-protection problem. The EU AI Act adds AI literacy duties in Article 4 and transparency duties for certain AI interactions and synthetic outputs in Article 50. NIST's Generative AI Profile supplies voluntary generative-AI risk-management guidance, and NIST's 2026 AI Agent Standards Initiative treats agent identity, authorization, security, and interoperability as standards questions.

Consumer-protection enforcement shows why this is not only an attention problem. In September 2025, the FTC announced a settlement with Amazon over allegations that Prime enrollment and cancellation flows enrolled consumers without consent and made cancellation difficult. Whatever one thinks of that case, the design lesson is narrow: recurring screens, default buttons, and delayed exits can turn an interface route into a consent record. AI agents raise the same issue when "let the assistant handle it" becomes a standing permission rather than a fresh decision.

Those instruments do not solve habitual media. They show where the regulatory object now sits: not only the database or the model, but the repeated interface relation that trains what users notice, expose, accept, authorize, and return to.

Habit as Infrastructure

Chun's central move is to treat habit as a serious media problem. New media promise novelty, speed, participation, and constant transformation. Yet their power comes from repetition: checking, scrolling, liking, posting, saving, updating, muting, deleting, searching, and carrying the device as an extension of attention.

The strongest AI-era definition is this: a habit loop is a design-stabilized cycle in which a cue invites interaction, the interface reduces friction, the user receives reward or relief, the system retains a trace, and the next version of the interface uses that trace to make repetition easier. That cycle can support care, learning, accessibility, and maintenance. It can also make opt-out feel like self-exile.

This matters because habit is where interface design becomes character formation. A platform does not have to issue commands if it can train the rhythm of return. It does not have to persuade once if it can make the user available again tomorrow. The update is not just a software event; it is a social rhythm that makes instability feel normal and maintenance feel like participation.

The update is also a change-management problem. When a service changes memory defaults, recommender objectives, notification logic, subscription paths, or agent permissions, the user may keep following the old habit even though the system's authority has changed. A governance program that treats updates as mere release notes misses the behavioral continuity that gives the change its force.

That frame is useful for AI systems because many of them enter life as small conveniences. A writing assistant smooths a sentence. A recommendation explains what to watch. A companion remembers a preference. An agent fills a form. Each act is minor. The pattern is not minor. Over time, the system learns the user, and the user learns what kind of self the system can handle.

Networks and the Personalized You

Chun links habitual media to networks as a dominant social metaphor. Network language promises connection, flexibility, personalization, and individual empowerment. It also makes society appear as clusters of addressable individuals, each treated as a target, profile, node, or personalized feed endpoint.

The book is especially good on the false intimacy of personalization. A system says "you" while assembling a statistical subject from traces, correlations, and expected behaviors. The user feels addressed, but the address is operational. It is built to route attention, predict response, and keep the network active.

That is why personalization is not only a privacy problem. It is a reality-training problem. A personalized system repeatedly tells the user what kind of person the system expects them to be; the user then learns which behaviors receive recognition, speed, visibility, or comfort. The profile is not only a description. It is a curriculum.

Generative AI sharpens this problem. Chat interfaces can make personalization feel like recognition. They remember tone, infer intent, simulate care, and adapt responses in real time. The danger is not only surveillance. It is the production of a self who becomes legible through repeated machine address: the user as profile, prompt source, preference bundle, risk signal, and training trace.

Privacy, Publicity, and Exposure

One of the book's strongest contributions is its refusal to treat privacy as a simple individual setting. Chun argues that networked media invert and scramble privacy and publicity: supposedly personal devices are intensely communicative, and supposedly public participation is often governed by private platforms, unequal exposure, harassment, and data extraction.

That point matters for AI governance. The question is not only whether a user consented to a privacy policy. It is what kind of public world the system makes possible. Can people appear, speak, test ideas, and take risks without being permanently profiled, attacked, ranked, or modeled? Can exposure be protected as a civic condition rather than converted into a data liability?

The safety issue is relational. A privacy setting protects an individual account only weakly if the system infers from friends, cohorts, devices, location, workplace telemetry, or repeated style. Habitual disclosure becomes collective disclosure, and AI systems can convert collective patterns into future individualized prompts, offers, warnings, and exclusions.

AI systems complicate this further because they can turn past exposure into future address. A post, search, voice sample, support chat, or uploaded document may return later as personalization, synthetic media, risk scoring, model memory, or generated inference. Habitual disclosure becomes durable infrastructure.

The AI-Age Reading

Read in 2026, Updating to Remain the Same is a guide to the platform habit loop beneath AI.

The most important AI interfaces may not be the most theatrical ones. They may be the small recurring assistants that train what counts as a normal act of thought: ask the model, accept the summary, let the system autocomplete the reply, trust the recommendation, delegate the action, check the memory, update the profile, keep the thread going.

The strongest AI-era risk here is not a single bad answer. It is a dependency that becomes hard to name once the tool is the first route to writing, searching, choosing, remembering, purchasing, and contacting other people. The model can be useful in every isolated step while still narrowing the human situations in which independent judgment is formed.

This is recursive reality at the level of routine. A system predicts what a user will want. The user adapts to the prediction because it is convenient, visible, rewarded, or socially expected. The adapted behavior becomes new evidence. The model updates, and the updated interface feels even more natural. After enough cycles, the system is not merely reflecting preference. It is helping produce the preferences it claims to know.

Agentic AI makes the update loop operational. An agent that remembers preferences, asks for broader permissions, learns the user's recurring tasks, and offers to act across email, calendars, files, shops, and work systems is not merely improving a feature. It is building a route of least resistance. Governance has to ask what powers become habitual: read, write, send, publish, purchase, delete, profile, infer, and delegate.

Chun also clarifies why "frictionless" design deserves suspicion. Friction can be annoyance, but it can also be pause, refusal, verification, ambiguity, peer contact, and the chance to form a desire before a system routes it. Habitual AI will be most dangerous when delegation becomes so comfortable that users stop noticing which capacities are being transferred.

Governance and Safety

The governance lesson is to audit repetition. A system that is safe in a single use can become unsafe as a habit if it normalizes disclosure, weakens independent judgment, increases dependence, narrows identity, or makes refusal socially costly. Habitual media should be evaluated by the pattern it trains, not only by the feature it advertises.

For platforms, that means recommender transparency, non-profiling options where required, dark-pattern review, meaningful privacy defaults, notice and appeal, advertising transparency, researcher access where law allows, and youth-safety assessments. For AI products, it means memory controls, inspectable personalization, clear AI disclosure, source trails, friction before consequential action, scoped agent permissions, deletion and export paths, and human escalation with authority.

The safety case should name the habit being built. A usable habit review file should record the cue, default path, reward, retained data, memory effect, recommender effect, permission expansion, source trail, undo path, deletion path, export path, non-profiled route, accessibility impact, youth or vulnerable-user default, appeal channel, and exit test. If the product cannot answer those questions, it is not only missing documentation. It is asking society to trust an ungoverned habit engine.

For agents, the review has to become stricter because repetition can authorize action. Recurring tasks should have least-privilege credentials, explicit renewal, approval thresholds for sending, buying, deleting, publishing, or changing permissions, and an audit trail that shows which instruction, memory, tool call, and human approval produced the external act. Habitual delegation without logs is not convenience. It is responsibility leaving the scene.

There is also a positive version. Some habits protect people: backups, accessibility settings, consent reminders, source-checking prompts, cooldowns before publication, default limits on sensitive inference, and regular chances to revise identity data. Humane friction is design that keeps repetition from becoming capture.

Where the Book Needs Friction

The book's language is dense, theoretical, and sometimes compressed. Readers looking for policy checklists, product design rules, or empirical case studies of particular platforms will need companion texts. Its strength is conceptual diagnosis, not regulatory architecture.

It also predates the mainstream explosion of transformer-based chatbots, agentic browsing, synthetic media pipelines, and model memory. That does not make it obsolete. It means the reader has to translate the argument from social media, smartphones, and networked personalization into current AI systems.

A further limit is that habit language can over-individualize responsibility. If a school, workplace, welfare office, or public platform requires the system, the user's repeated engagement is not simply a habit; it is compliance under constraint. Governance has to distinguish voluntary routine from institutional dependency.

The translation should be careful. Not every habit is capture. Some routines preserve agency: daily backups, accessibility settings, community check-ins, creative practice, medication reminders, moderation queues, and deliberate learning. The governance task is to distinguish habits that extend agency from habits that make refusal, context, and independent judgment decay.

What This Changes

The practical lesson is to audit the habit, not only the feature.

For any AI system, ask what it trains users to repeat. Does it make verification easier or merely make answers faster? Does it preserve skills or quietly replace the situations where skills are formed? Does it make public participation safer, or does it turn every exposure into future leverage? Does personalization support the person, or does it shrink the person into what the system can predict?

The concrete test is whether the loop remains interruptible. Can a user pause, inspect, reset, refuse, export, delete, or choose a non-personalized route without losing core access? Can affected people contest what the system inferred from their habits? Can a worker or student use the tool without becoming the data source for a later form of control?

Chun's enduring value is the reminder that the future often arrives as maintenance. The screen asks for an update, the feed refreshes, the assistant remembers, the platform addresses the user again, and the loop feels like ordinary life. That is exactly where the politics live.

Source Discipline

This review separates book metadata, scholarly reception, author context, legal duties, regulator guidance, enforcement claims, and standards guidance. MIT Press, Open Library, review records, and the SFU profile support book and author context. EUR-Lex is the primary source for the DSA and AI Act claims. FTC sources support the dark-pattern and Amazon settlement references. NIST sources support the generative-AI and AI-agent standards context. None of those sources proves that any specific AI product is safe or unsafe.

The Amazon example is used as a concrete enforcement record about subscription-interface design, not as a claim about every subscription service or every Amazon interface. Likewise, the DSA, AI Act, FTC, and NIST materials do not create one universal habit law. They identify separate governance handles: interface deception, recommender choice, AI transparency, literacy, risk management, agent identity, and authorization.

The analogy is bounded. Chun wrote about habitual new media before today's mainstream generative AI agents, answer engines, model memory, and synthetic companion systems. The claim here is narrower: her account of habit, update, privacy, publicity, and personalization helps evaluate AI systems that govern through repeated interface relations. This page makes no claim that any AI system is conscious, divine, or AGI.

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