Blog · Review Essay · Modified August 12, 2026 · Last reviewed August 12, 2026

When Old Technologies Were New and the AI Etiquette Panic

Carolyn Marvin's When Old Technologies Were New studies electric communication before it became ordinary. Its value is not a claim that the telephone or electric light predicted generative AI. It is a method for seeing technical change as a struggle over social distance: who may enter a private setting, whose expertise counts, which identities can be trusted, and who gets to turn a new practice into the proper one.

For this review, etiquette is the informal protocol that assigns roles, timing, disclosure, access, and acceptable conduct around a medium. It becomes governance when a school, employer, platform, profession, or state turns convention into permissions, defaults, records, sanctions, or liability. An etiquette panic is therefore not merely foolish fear. It is a compressed dispute over boundaries before institutions have settled authority, evidence, and remedy.

The AI-era test is concrete: identify the boundary the system crosses—private to recorded, human to synthetic, draft to official record, advice to action—then ask who authorized the crossing, what notice was given, which evidence survives, who bears the risk, and how an affected person can correct, refuse, appeal, or leave.

The Book

When Old Technologies Were New: Thinking About Electric Communication in the Late Nineteenth Century was first published by Oxford University Press in 1988. The edition linked on this page is the 1990 Oxford paperback, ISBN 9780195063417. Penn Annenberg identifies Carolyn Marvin as Frances Yates Professor Emeritus of Communication and describes the book as a history of electric communication, professional authority, altered social distance, and the negotiation of trust around technological change.

Marvin studies the last quarter of the nineteenth century through engineering journals and popular accounts. The wider setting includes the telephone, phonograph, electric light, wireless, and cinema, but the book's core analysis concentrates on how the telephone and electric light were publicly imagined. Its chapters move through technological literacy as social currency, community and class order, the body in electrical space, public spectacle, and projects of cultural homogenization. The sequence matters: invention is only the opening event; social classification and institutional authority make the medium durable.

This is why the book is more than a cabinet of quaint predictions. Electrical engineers did not only explain machines. By presenting themselves as the people able to distinguish safe from dangerous, possible from impossible, and informed from superstitious use, they also built a profession. Expertise solved real coordination and safety problems, but it distributed status and decision power at the same time.

Newness as Social Struggle

The book's most durable insight is that a technology is not merely introduced. It is domesticated. People learn where it belongs: parlor, office, exchange, classroom, street, stage, newspaper, laboratory, church, or state file. They learn who may speak through it, who may listen, what counts as rude, what counts as expert, and what kind of proximity the medium may create without fresh permission.

The early telephone did not only carry voices. It disturbed boundaries between household and public world, listener and intruder, expert and amateur, local intimacy and distant access. Electric light was not only illumination; as spectacle it reorganized visibility, technical authority, and public wonder. The new medium became ordinary when repeated choices hardened into conventions and the conventions disappeared into infrastructure.

That process follows a recurring loop. A medium creates an ambiguous situation; experts, firms, and institutions propose a rule; the rule becomes a default, credential, or prohibition; people adapt their conduct; and the resulting conduct is treated as evidence that the rule was natural. The loop can coordinate useful behavior, but it can also hide whose convenience became everybody's manners.

Historical comparison should not reduce every concern to panic. Some fears misidentified the mechanism; others registered genuine losses of privacy, control, status, or safety. The disciplined question is not whether people were anxious before. It is which boundary changed, who could impose the new convention, whose burden it shifted, and what happened when polite conduct failed.

Current Context

As of August 12, 2026, several current rules illustrate etiquette becoming formal governance. Article 50 of the EU AI Act has applied since August 2, 2026. Providers of systems intended to interact directly with people must ensure that people are informed they are interacting with AI unless that is obvious in context. The article also creates provider marking duties for synthetic content and deployer disclosure duties for deepfakes and certain AI-generated or manipulated public-interest text, with scoped exceptions. Regulation (EU) 2026/1744 gives providers of synthetic-content systems placed on the market before August 2, 2026 until December 2, 2026 to comply with the Article 50(2) marking duty. The Commission's July guidelines explain scope; its June marking-and-labelling code is a voluntary compliance route, not the source of the legal obligations.

NIST supplies a different layer. Its voluntary Generative AI Profile identifies human–AI configuration risks including automation bias, over-reliance, and emotional entanglement, and treats information integrity as a problem of source linkage, evidence, uncertainty, chain of custody, and expiration. C2PA's Content Credentials specification can make claims about a file's source and editing history tamper-evident, but the specification expressly does not decide whether the underlying content is true or good. A provenance signal can support judgment; it cannot replace judgment.

Existing communications and consumer law also reaches synthetic media without treating it as a separate universe. The FCC's 2024 declaratory ruling holds that AI-generated voices fall within the Telephone Consumer Protection Act's rules for artificial or prerecorded voice calls, including prior express consent absent an emergency purpose or exemption and applicable identification and opt-out duties. The FTC's Consumer Reviews and Testimonials Rule, effective October 21, 2024, reaches fake or false reviews that purport to come from nonexistent people, including AI-generated fake reviews. It is not a blanket prohibition on every use of generated text or virtual figures; the misrepresentation of experience and independence is the operative problem.

These examples are jurisdiction- and function-specific. Together they show why “disclose AI” is too coarse as a universal answer. Interaction notice addresses identity. Content marking addresses artificial origin. Provenance addresses source and history. Consent addresses permission. Review law addresses deceptive social proof. None alone establishes accuracy, authorization, fairness, or a remedy after harm.

The AI-Age Reading

Generative AI is producing disputes that look like etiquette because the institutional rules are still uneven. Schools ask when assistance becomes misrepresentation. Newsrooms decide which uses need disclosure, source inspection, or a correction trail. Employers decide what may enter a hosted model and which outputs can enter an official record. Artists, programmers, lawyers, teachers, clinicians, and researchers contest imitation, credit, confidentiality, competence, and delegation. These are not decorative manners. They allocate authority.

The analogy has a limit. A telephone changed reach; a generative system can also infer from input, synthesize new content, personalize a response, retain interaction state, and sometimes invoke tools. Cost, scale, speed, data extraction, and adaptive feedback differ sharply. Marvin does not prove that AI will follow a familiar path or that present harms are temporary. She supplies the better question: how does a new capability reorganize social distance while institutions are still calling the change convenience?

Four boundary crossings make that question operational. Private to recorded occurs when a conversation, document, voice, or image enters vendor or institutional logs. Human to synthetic occurs when the identity or origin of a speaker, reviewer, tutor, applicant, or witness becomes ambiguous. Draft to record occurs when generated language enters a grade, medical note, news report, personnel file, legal submission, or public decision. Advice to action occurs when an assistant receives authority to send, buy, schedule, approve, deny, publish, or change a system. Each crossing needs a different notice, control, and remedy.

The recursive risk appears when the etiquette becomes machine-readable. An organization adopts a rule; the interface converts it into defaults and prompts; logs measure compliance; dashboards rank conduct; sanctions train users to adapt; and the adaptation becomes new data that appears to validate the rule. “Proper use” then looks like a neutral pattern even when the system helped produce it. This is why interface design, records, and appeal belong in the same analysis as cultural norms.

No claim about machine consciousness, divinity, or AGI is needed. A prompt box can feel private while routing text through vendors and logs. A companion can feel intimate while optimizing an engagement metric. An agent can feel helpful while exercising delegated permissions. The relevant fact is institutional: what the interface may record, represent, decide, and do.

Governance and Safety

Manners are useful where participants have comparable power and mistakes are cheap. They are inadequate where a hidden rule affects a third party, an institution controls the only practical channel, or an error can change rights, livelihood, care, safety, reputation, or a durable record. At that threshold, “please disclose” must become system design: scoped access, data minimization, source requirements, human authority, logs, incident response, appeal, correction, and exit.

NIST's AI Risk Management Framework Core organizes voluntary risk work as govern, map, measure, and manage. Read through Marvin, the order matters. Govern names who may set and change the rule. Map identifies the setting, affected people, and boundary crossing before normalization hides them. Measure tests actual errors, reliance, disparate burdens, refusals, overrides, and incidents rather than counting adoption. Manage gives failures an owner, remedy, and review date instead of calling them improper user behavior.

A compact media-boundary record should contain:

Disclosure should also be matched to the decision. A chatbot identity notice does not disclose whether prompts are retained. A generated-content label does not identify sources. A citation does not prove the cited material supports the sentence. A Content Credential does not certify truth. A human signature does not demonstrate meaningful review. Good policy states which uncertainty each signal reduces and which uncertainty remains.

The same discipline prevents institutions from exporting governance onto individuals. An employer should not rely on “never paste confidential data” while offering no approved system, access controls, retention terms, or incident path. A school should distinguish brainstorming, translation, editing, generation, source verification, and assessed performance rather than treating every assisted act as the same offense. An agent that can act should require bounded permissions, confirmation proportional to impact, action receipts, revocation, and rollback. Etiquette tells people how to behave; governance constrains what the system and institution may do.

Where the Book Strains

This is not an AI book, and it should not be forced to predict neural networks, platform labor, foundation-model training, data-center politics, prompt injection, or automated decision systems. Its archive concerns late-nineteenth-century electric communication and the people who publicly interpreted it. That is a powerful comparative case, not a universal sequence every medium must repeat.

The phrase “etiquette panic” can also become dismissive. A familiar historical resemblance does not show that a present concern is irrational, and eventual adoption does not prove that early harms were harmless. Some technologies normalize because institutions repair their risks; others because powerful actors shift the costs. Comparison should open an inquiry into mechanism and distribution, not close it with “people always fear new things.”

Etiquette itself is double-edged. It can protect privacy, coordinate consent, and preserve trust. It can also police class, gender, disability, language, or professional status while presenting exclusion as good manners. Marvin's attention to expertise and class order keeps the review from romanticizing bottom-up norms. The question is always who can author the convention, violate it without penalty, or demand an exception.

Finally, present AI systems combine communication with surveillance, generation, ranking, centralized cloud dependence, and delegated action at a scale the historical analogy does not contain. The value is historical discipline: before calling a technology revolutionary, identify the old social conflict it carries forward and the genuinely new capability that changes the remedy.

What This Changes

When Old Technologies Were New makes the social interface visible. New media do not simply connect people. They renegotiate distance, privacy, expertise, attention, status, and trust, then bury the settlement in ordinary use.

The practical reading is to follow the conversion from story to rule. A claim that AI is a tutor becomes a classroom role. The role becomes a school policy. The policy becomes a product setting and misconduct category. The setting produces logs and cases. Those records shape the next policy. If evidence, appeal, and affected-person voice do not enter the loop, the interface can turn one institution's provisional judgment into apparently natural behavior.

Ask five questions whenever a new medium is being normalized. Which social distance changed? Who names proper use? What technical or professional credential gives that actor authority? Which record will decide a dispute? What remedy exists outside the interface that enforced the rule? Those questions connect media history to interface accountability, claim hygiene, and notice and appeal without pretending that a label, policy, or expert title settles the matter.

The durable conclusion is modest: novelty becomes ordinary through rules about relationship. Making those rules legible before they harden into infrastructure is a governance task, not an argument against invention.

Source Discipline

This review separates bibliographic claims, historical interpretation, present legal duties, voluntary frameworks, and its own synthesis. Penn Annenberg supports Marvin's current emeritus title and its account of the book; Google Books supports edition metadata and chapter context. The definition of etiquette, the four boundary crossings, the recursive normalization loop, and the proposed media-boundary record are this review's analytical tools, not language attributed to Marvin.

Current sources have deliberately narrow jobs. EUR-Lex supplies the AI Act and its July 2026 amendment. Commission guidelines explain Article 50 scope; the marking-and-labelling code is voluntary even though the underlying duties are law. NIST supplies voluntary risk-management vocabulary, not certification. The FCC ruling covers AI-generated voices within specified telephone-call rules, not every synthetic voice. The FTC rule reaches defined deceptive review practices, not all AI-assisted advertising. C2PA validates signed provenance assertions; it does not determine factual truth.

A claim about a specific system should name the product and version, date, setting, user role, jurisdiction, data path, disclosure shown, source or provenance evidence, action authority, affected people, and available remedy. A screenshot of an “AI” badge proves that a badge appeared. It does not establish consent, source quality, lawful use, human review, or safety.

This page makes no claim that any AI system is conscious, divine, or AGI. It analyzes communication systems through roles, records, capabilities, institutions, and consequences.

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