Blog · Review Essay · Modified July 10, 2026 · Last reviewed July 10, 2026

Smart Mobs and the Crowd That Learned to Compute

Howard Rheingold's Smart Mobs is a pre-smartphone book about the social force of always-on coordination. Its enduring value is not that every forecast landed cleanly. It is that Rheingold saw mobile phones, wireless networks, reputation systems, sensors, peer production, and location-aware media converging into a new kind of collective actor: groups that can find, signal, trust, swarm, buy, protest, harass, flee, and govern through networked devices.

A smart mob, in this review, means a collective assembled through mobile signaling, reputation cues, location awareness, and shared computational infrastructure. It is not automatically wise, democratic, or emancipatory. It is a coordination form whose politics depends on who can signal, rank, verify, surveil, monetize, and intervene.

The sharper definition is a computational crowd: a public whose ability to notice, gather, trust, and act is partly produced by devices, platforms, rankings, identity systems, provenance layers, and now AI agents. The safety question is not whether crowds are good or bad. It is whether the machinery that assembles them is inspectable, contestable, and resistant to capture.

The governance object is the coordination record: initiating signals, amplification surfaces, identity cues, provenance claims, location or contact-graph use, agent actions, paid influence, moderation moves, and appeals. Without that record, a crowd can be mobilized, steered, or discredited before anyone can tell which public actually acted.

The Book

Smart Mobs: The Next Social Revolution was originally published by Perseus in 2002 and later appeared from Basic Books. Current publisher records list the Basic Books trade paperback, ISBN 9780738208619, as a 288-page edition on sale October 16, 2003. Library and archive records preserve the original publication history and the book's range of subjects: technology and civilization, communication and culture, internet social aspects, cellular telephones, wireless networks, reputation, collective action, and the final question of an "always-on panopticon" versus a cooperation amplifier.

Rheingold wrote before the iPhone, before mainstream social media had settled into platform form, before app stores, ride-hailing, livestreamed protest, QR-code check-ins, large-scale geolocation advertising, and generative AI agents. That makes the book useful as a fossil of perception. It catches the moment when mobile computing still looked like a set of loose pieces rather than a global behavioral layer.

The book's strongest claim is that social coordination changes when communication, computation, identity, location, and reputation travel with the person. A crowd with phones is not just a crowd plus gadgets. It is a crowd with a nervous system, and that nervous system can be civic, commercial, coercive, or all three at once.

Current Context

As of July 10, 2026, the loose mobile-media stack Rheingold described has become ordinary platform infrastructure. Phones, maps, cameras, payments, contact graphs, encrypted groups, recommendation systems, ad delivery, bot accounts, synthetic media tools, and delegated AI agents now sit inside the same practical coordination field. The crowd is no longer only connected. It is ranked, measured, monetized, simulated, and partially automated.

The regulatory vocabulary has caught up only partly. The European Commission describes very large online platforms and search engines under the Digital Services Act as services with more than 45 million monthly users in the EU and says they face the DSA's most stringent rules, including transparency around advertising, recommender systems, content moderation, systemic-risk assessment, risk mitigation, and data access. The Commission's supervised list of designated very large platforms and search engines was updated on July 10, 2026, which makes the platform layer a live governance object rather than a settled archive. The Commission's AI Act guidance says transparency rules for AI interaction and generated content come into effect on August 2, 2026, after this review date. Those rules are relevant to smart mobs because public scale, urgency, and authenticity can now be manufactured or amplified through recommendation, generation, and targeting.

NIST's AI Agent Standards Initiative, announced in February 2026, moves the issue from social-media coordination into delegated software action. NIST frames agents as systems capable of autonomous actions, secure action on behalf of users, and interoperability across the digital ecosystem; its project page emphasizes open protocols, authentication, authorization, identity infrastructure, and security evaluations for human-agent and multi-agent interaction. A smart mob with agents is not simply a faster crowd; it is a crowd in which some participants may be non-human principals acting through credentials, tools, and logs.

The U.S. Federal Trade Commission's September 2024 staff report supplies the data-governance backdrop. It found that major social media and video streaming services engaged in broad surveillance to monetize personal information, had weak minimization and retention practices, and gave users little or no ability to opt out of how their data fed automated systems. That matters because collective action built on location, contact, reputation, and attention data can also become collective vulnerability.

Coordination Becomes Infrastructure

Rheingold's recurring examples include Japanese texting culture, Finnish mobile communities, wireless commons, eBay reputation, peer-to-peer networks, distributed computing, protest coordination, and the use of text messaging in the Philippines during the demonstrations against Joseph Estrada. The connecting thread is not mobile novelty. It is the lowering of coordination costs.

When people can signal quickly, discover one another, update plans, and read shared cues in near real time, collective action changes shape. It can become more flexible, more improvisational, and harder for institutions to anticipate. A market can form around trust ratings. A protest can reroute. A flash crowd can appear. A rumor can move faster than verification. A state can watch the same signals and learn to intervene.

That is the article's first useful definition: coordination cost is the friction between wanting to act together and actually being able to do so. Mobile networks cut that friction by shortening the path from signal to gathering, from event to witness, from rumor to cascade, and from isolated complaint to visible public pressure. They also shorten the path from observation to intervention.

The modern version is a coordination stack: device identity, contact graph, maps, payments, cameras, message routing, recommendation, cloud storage, and trust signals all working together. Each layer reduces one friction while creating a new dependency. The smart mob can gather because the stack remembers names, locations, links, and reputations; the same stack lets platforms, employers, police, advertisers, and adversaries reconstruct how the gathering happened.

A usable coordination record should answer ordinary operational questions: who initiated the signal, which surface amplified it, what identity or provenance cues were shown, what paid or automated assistance was used, what location or contact data was consumed, what moderation or ranking intervention happened, and what remedy exists when the crowd signal is false, harmful, or misattributed. That record is not a demand for total surveillance. It is the minimum evidence needed to govern systems that already shape crowds at scale.

The stack also decides which crowd becomes visible. A trending module, map pin, push notification, recommendation queue, group invite, creator payout, verified badge, or agent-suggested next step can turn diffuse interest into coordinated motion. That means coordination infrastructure is also an allocation system: it allocates attention, credibility, proximity, urgency, and sometimes physical risk.

This is why the book belongs beside work on networked publics and media theory rather than only mobile technology. Smart Mobs is about the interface between social psychology and technical architecture. The practical question is what people become able to do together once the network is carried in the pocket, attached to place, and mediated by rankings that decide which signals count.

Reputation as a Machine

One of the book's most important threads is reputation. Rheingold saw that cooperation at scale depends on signals of trust: seller ratings, moderation histories, social graphs, public contributions, persistent names, and other cues that help strangers decide how to act around one another.

That insight aged well, but not innocently. Reputation systems can support cooperation, but they can also become status machines, exclusion machines, disciplinary machines, and markets in artificial credibility. They make trust visible by compressing it. The compression is useful, and the compression is dangerous.

The AI-era version is sharper. Ranking, scoring, verification badges, follower graphs, review systems, recommender signals, contributor histories, and model-generated summaries now feed into automated decisions. A reputation cue can determine visibility, access, credit, employment opportunity, moderation priority, or whether an AI agent treats a request as trustworthy. Once reputation becomes machine-readable, it becomes governable by systems most users cannot inspect.

That puts Smart Mobs in direct conversation with recommender systems and platform governance. The problem is not only that bad actors can fake reviews or inflate followers. It is that technical systems turn reputation into an input for distribution, risk scoring, content moderation, fraud detection, search rank, and agent delegation. Trust becomes a measurable asset, and measurable assets invite capture.

That is why reputation governance must include contestability. A rating, badge, fraud score, moderation history, or agent trust score should have a record of source, purpose, expiry, appeal path, and downstream use. It should also be scoped: a marketplace delivery dispute should not silently become a general trust mark for civic participation, financial access, or agent delegation. Otherwise the cue that helps strangers cooperate becomes a portable stigma or a synthetic credential that follows people across systems without context.

For agentic systems, the reputation cue also needs an identity boundary. If an assistant is allowed to join a group, spend money, rate a seller, book travel, post a reply, or call another agent, the system should preserve whether the action came from a human, a delegated agent, a service account, a paid campaign, or a synthetic participant. Without that record, a smart mob can become a smart-looking blur of borrowed authority.

The Cooperation Amplifier and the Panopticon

The book is often remembered for its enthusiasm, but its final tension is darker: the same always-on media that enable cooperation also enable surveillance. Rheingold's Guardian interview in 2004 made that ambivalence explicit. He described mobile and internet coordination as empowering, but also warned that it could amplify the capacities of individuals, organizations, and states for harmful action.

This double edge is the book's most durable discipline. The mobile network does not decide whether it is civic infrastructure or control infrastructure. Design, law, business models, police practice, labor relations, and social norms decide. The same location trail that helps friends gather can help employers monitor workers, police map crowds, marketers infer vulnerability, or abusive people track targets.

That is the bridge to contemporary surveillance and AI governance. A phone is no longer only a communication device. It is a sensor package, identity token, payment interface, camera, workplace terminal, authentication device, social graph, and behavioral record. The Federal Trade Commission's 2024 staff report on major social media and video streaming services is useful here because it treats data collection, targeted advertising, automated decision-making, and weak privacy controls as one institutional system rather than isolated features. Add machine learning and the crowd's nervous system becomes readable from above.

The AI-Age Reading

Read in 2026, Smart Mobs looks like a prehistory of synthetic coordination.

The original smart mob was made of humans using devices. The AI-era smart mob includes bots, recommendation systems, generated media, automated accounts, agentic browsers, payment agents, moderation models, synthetic respondents, and tool-using assistants that can coordinate tasks across platforms. Collective action is no longer only people finding people. It is people, models, and institutions acting through shared computational surfaces.

NIST's 2026 AI Agent Standards Initiative shows why this is no longer only a social-media question. Its framing treats agents as systems capable of autonomous actions and emphasizes standards, open protocols, secure action on behalf of users, interoperability, authentication, and identity infrastructure. For smart mobs, that moves the coordination problem from "who is in the crowd?" to "which humans, bots, assistants, service accounts, and delegated permissions are acting inside the same apparent crowd?"

This changes the old problem of the crowd. The question is not simply whether a public is wise or irrational. It is whether the public is being assembled by ranking systems, memory systems, synthetic participants, invisible experiments, and automated persuasion. A crowd can be real and still be partially generated. A consensus can be socially consequential while being seeded, amplified, summarized, or simulated by nonhuman systems.

The danger is not that a model is a member of the public. The danger is that institutions may treat machine-amplified social proof as public consent, public outrage, public panic, or public expertise. The governance problem is therefore evidentiary before it is metaphysical: what counted as participation, who could inspect the count, and how quickly could a mistaken signal be corrected?

Rheingold's phrase still helps because it keeps attention on coordination rather than content alone. The decisive AI harms will not always look like false statements. They may look like faster mobilization, more believable social proof, automated harassment, synthetic legitimacy, distorted reputation, and institutional panic in response to signals whose human basis is unclear. That is the same problem raised by synthetic respondents and synthetic publics: generated social evidence can steer real institutions before anyone has established who, or what, it represents.

Governance and Safety

The governance implication is concrete: systems that assemble publics need controls for identity, provenance, ranking, data access, and redress. The European Union's Digital Services Act gives one current vocabulary for large platforms: systemic risk assessment and mitigation for very large platforms and search engines, transparency around recommender systems, advertising repositories, and data access for regulators and vetted researchers. Those duties do not solve the smart-mob problem, but they name the platform layer where coordination is shaped.

The EU AI Act adds a second vocabulary for generated interaction and generated media. Article 50 transparency obligations, scheduled to apply from August 2, 2026, address direct interaction with AI systems and marking or labeling certain AI-generated and manipulated content. In smart-mob terms, that matters because a crowd's apparent scale, mood, urgency, and authenticity can now be manufactured or assisted by systems that speak, summarize, imitate, and remix at volume.

Standards and risk frameworks supply narrower tools. C2PA provenance can attach signed claims about the source and history of media, while the NIST AI Risk Management Framework organizes risk work around govern, map, measure, and manage. Neither is a truth machine. Provenance can be missing, stripped, forged outside the trust boundary, or ignored by platforms; risk frameworks can become paperwork. Their value is operational: they turn "was this real?" into inspectable questions about source, transformation, model use, custody, distribution, and accountability.

The minimum governance record for smart-mob infrastructure should preserve the coordination path: initiating signal, sponsor or operator, paid or unpaid status, bot or agent identity, recommender surface, ranking objective, provenance signal, location and contact-data use, moderation or amplification actions, crisis escalation, and appeal or correction route. The record should be privacy-minimized, but absence of a record should not become a shield for platforms or institutions that shape public action.

For public institutions, procurement and deployment of tools that monitor public sentiment, route emergency response, detect events, summarize social reaction, or flag coordinated behavior should require provenance support, bot and agent separation, source sampling, uncertainty reporting, human review for high-impact interventions, and a non-automation fallback. A government should not convert a dashboard spike into policing, benefit administration, disaster response, or public communication without knowing whether it is seeing people, automation, paid influence, or platform ranking.

A useful safety checklist for smart-mob infrastructure now includes bot and agent disclosure, agent identity and authorization records, rate limits on coordinated inauthentic behavior, archives for paid public-influence campaigns, crisis escalation paths, notice and appeal for moderation, provenance display for contested media, data minimization for location and contact graphs, researcher access where public risk is high, and audit logs when agents act on behalf of people or organizations. The goal is not to stop collective action. It is to make collective action less dependent on invisible ranking, invisible data capture, and unverifiable social proof.

Where the Book Needs Friction

The book sometimes inherits the early internet habit of treating more participation as a default good. Later platform history makes that harder to sustain. Participation can produce mutual aid, open knowledge, and democratic pressure. It can also produce pile-ons, rumor cascades, gamed metrics, leaderless fragility, and publics that can mobilize before they can deliberate.

Rheingold was not naive about danger, but the political economy of platform capture was still underdeveloped in 2002. The major question now is not only whether people will be users or consumers. It is whether the infrastructure of collective action will be owned by firms that sell attention, inference, compute, identity, payments, cloud access, app distribution, and model-mediated trust.

The book also needs to be read alongside critics of surveillance, labor, racialized classification, and institutional power. Coordination tools do not land on a flat society. The people most exposed to monitoring, policing, workplace discipline, immigration enforcement, and automated scoring experience the smart mob's infrastructure differently from the people who experience it as convenience or expressive freedom. The same live map can be a mutual-aid tool, a protest resource, a police target list, an employer dashboard, or an advertising segment.

What This Changes

The core lesson is that coordination is a form of reality construction.

When an interface shows who is nearby, what is trending, who is trusted, where to go, what others believe, which route is safe, which post is rising, which merchant is reputable, or which source an agent recommends, it is not merely reporting the social world. It is helping assemble the social world it reports.

Smart Mobs remains worth reading because it saw the crowd become computational before the computational layer became ordinary. The next version is not just mobile. It is generative, reputational, surveillant, and agentic. That makes the old question more urgent: will networked coordination widen human agency, or will it make publics easier to summon, score, steer, and sell?

The answer cannot be found in enthusiasm for connection or panic about mobs. It has to be built into the coordination layer: source trails for media, accountable identities for delegated action, limits on location and contact-graph data, public-interest access for risk researchers, and remedies for people harmed by ranking, impersonation, harassment, or mistaken reputation.

Source Discipline

This review separates three claims. The book claim is bibliographic: publisher, author, library, interview, and scholarly records establish what Rheingold wrote and how the book was received. The interpretation claim is analytical: smart mobs are treated here as coordination infrastructure, not as a guaranteed democratic force. The current-governance claim is sourced to regulators and standards bodies, including platform law, AI transparency rules, provenance specifications, AI risk-management guidance, and agent standards. Those sources do not endorse Rheingold's argument or this site's interpretation.

Current claims should keep the layer visible. The DSA is a platform-governance law, not a general theory of crowds. The EU AI Act transparency provisions concern AI interaction and certain generated or manipulated content, not every form of social coordination. C2PA supports provenance claims, not truth claims. NIST's agent work is standards and research infrastructure, not a finding that agents are independent persons. The FTC report is evidence about covered companies' data practices, not proof that every coordinated crowd is surveilled in the same way.

No claim here requires treating AI systems as conscious, divine, or AGI. The relevant issue is institutional: generated media, recommender systems, agents, reputation systems, and mobile sensors can assemble signals that people and organizations treat as public reality. The safety problem is therefore about evidence, accountability, consent, and control.

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