Blog · Review Essay · Last reviewed June 25, 2026

The Smart Enough City and the City That Refuses to Become a Dashboard

Ben Green's The Smart Enough City is a compact argument against the fantasy that urban life becomes better when it becomes more computationally visible. It is not anti-technology. It is anti-substitution: against replacing democratic judgment, public services, local knowledge, and institutional repair with apps, sensors, algorithms, and optimization theater.

A smart enough city, in this review, is a city that lets technology earn its place. The test is not whether a system is advanced, automated, or data-rich. The test is whether it increases resident power, public capacity, repair, equity, and accountable judgment without turning the city into a vendor-controlled dashboard.

The practical question is therefore not "how smart can the city become?" It is "which public duty is this system supposed to strengthen, and what evidence would prove that residents can still contest, refuse, repair, or govern it?"

The Book

The Smart Enough City: Putting Technology in Its Place to Reclaim Our Urban Future was published by the MIT Press in 2019, with a paperback edition in 2020. MIT Press lists the book in its Ideas series and describes it as an argument for using technology to promote democracy and equity without treating technology as an end in itself. The open-access edition was made available through MIT Press and MIT Libraries support.

Green writes from the borderland between data science, public policy, urban governance, and civic technology. The University of Michigan School of Information identifies him as an assistant professor of information and public policy whose work focuses on the ethics of government algorithms, algorithmic fairness, human-algorithm interaction, and AI regulation. That institutional location matters because the book is not a detached complaint about gadgets; it is a practitioner-scholar's demand that cities ask what public institutions owe before asking what a system can optimize.

The book moves through smart-city promises about traffic, policing, public services, participation, and innovation, then asks what those promises hide when the city is treated as an optimization surface. The open-access edition makes the structure especially clear: chapters on the livable city, democratic city, just city, responsible city, innovative city, and the final framework of the "smart enough" city.

That structure matters. Green is not simply warning that sensors can surveil or algorithms can discriminate. He is arguing that the category "smart city" often starts in the wrong place. It asks what technology can do for the city before asking what the city owes its residents.

A useful working definition follows: a smart-city system is not just a sensor, model, app, camera, kiosk, portal, or digital twin. It is an institutional arrangement that decides which urban facts are collected, which problems are modeled, which residents become visible, which vendor controls the interface, which public office owns the outcome, and how people can refuse, correct, appeal, or reshape the system.

That definition keeps the review from treating "smart city" as a gadget category. A parking sensor, benefits chatbot, predictive maintenance model, surveillance camera, fare-gate analytics system, and digital twin are different tools, but they pose the same civic question: does the technology make a public duty more accountable, or does it move discretion into a layer residents cannot inspect?

Current Context

As of June 25, 2026, smart-city governance has moved beyond the old question of whether cities should use sensors and apps. The stronger question is whether city technology can be made reviewable as public power. New York City's 2023 AI Action Plan treats municipal AI as a city-government governance problem with risks including privacy, cybersecurity, bias, disparate impact, and opaque accountability. The Dutch government Algorithm Register publishes information about impactful algorithms used by public organizations, and Helsinki's AI Register describes itself as a window into city AI systems with feedback paths for residents. Those registers are imperfect, but they show the direction Green's book points toward: public systems should leave public evidence.

The broader policy stack now says the same thing in different vocabularies. OECD's smart-city data-governance work emphasizes data governance capacity, security, stakeholder complexity, and public value. OMB's 2025 federal AI-use memorandum requires risk-management practices for high-impact AI used by federal agencies, and its AI acquisition memorandum emphasizes portability, interoperability, performance monitoring, and vendor-lock-in controls. The EU AI Act makes defined uses in critical infrastructure and essential public services high-risk. ISO/IEC 42005 gives an impact-assessment frame for AI systems. None of these sources makes a city democratic by itself. They make one thing harder to excuse: deploying civic AI without a record of purpose, risk, responsibility, and recourse.

The current urban risk is not a single dystopian dashboard. It is a chain: cloud contract, data feed, camera or form, model, routing rule, case note, chatbot answer, retention setting, procurement renewal, and public-record exemption. A resident may never see the chain, but the chain can decide which complaint is urgent, which street is policed, which benefit case is flagged, which permit is delayed, which call is triaged, or which neighborhood becomes a target for optimization. A smart enough city keeps that chain short, inspectable, contestable, and accountable to people who live with its outputs.

Smart Is Not a Politics

The word "smart" is doing too much work. It can mean instrumented, efficient, responsive, automated, optimized, data-driven, or merely attractive to vendors and grant writers. Green's central move is to separate intelligence from wisdom. A city can gather more data while becoming less democratic. It can respond faster while hearing fewer people. It can make services more convenient while shifting power toward the firms that own the interface.

This is why the book pairs well with recurring concern about legibility and institutional substitution. Smart-city rhetoric often turns politics into a service-design problem: congestion becomes routing, civic engagement becomes an app, public safety becomes predictive policing, and urban inequality becomes a dataset awaiting better features. Each reframing can produce useful tools. Each can also narrow the moral field until the only visible problems are the ones a system can measure.

The sharper distinction is between a public problem and a product-shaped problem. A public problem may require housing, shade, transit frequency, language access, repair crews, budget authority, union staffing, public meetings, or legal change. A product-shaped problem is the portion that can be captured in a procurement line: a platform, dashboard, API, camera feed, prediction model, or analytics contract. Green's warning is that smart-city programs often let the second definition swallow the first.

Green's alternative is deliberately modest. The city should be smart enough: capable of using technical systems where they serve a public purpose, but not so enchanted by computation that it forgets housing, sidewalks, schools, labor, privacy, democratic control, and the unequal history already embedded in urban space.

The practical rule is public purpose before data capture. A city should be able to state the nontechnical reform the tool supports, the public office accountable for the outcome, the residents likely to bear error, the data it will not collect, and the reason a less invasive repair would not be enough. Without that record, "smart" becomes a license for institutional drift.

That makes "smart enough" an adequacy standard. The city should ask whether the technology is necessary, proportionate, accessible, reversible, and governable for the public duty at hand. If a sidewalk can be repaired with staffing, budget, and transparent scheduling, the city does not need a predictive platform to discover that broken pavement exists. If a dispatch model is used because the call center is underfunded, the city should say so rather than calling austerity innovation.

Urban Legibility

The strongest parts of the book show how smart-city systems make some urban realities visible while pushing others out of frame. A 311 app can make reported potholes and complaints easier to track, but the map of complaints may also reflect who has time, language access, trust, digital access, and confidence that the city will listen. The data is not simply the city speaking. It is the city filtered through participation, power, and interface design.

That distinction is essential for any AI-era public system. A dashboard can make city government feel empirically grounded. It can also create a second city: a model city composed of reports, categories, coordinates, risk scores, service tickets, camera events, inferred patterns, and predicted needs. Administrators then face the temptation to govern the model because the model is clearer than the place.

Green's value is that he does not romanticize ignorance. Cities do need records, measurements, maps, budgets, and data systems. But measurements have to remain accountable to the lived city. A public interface should help residents contest and improve the system, not merely convert them into signals that administrators and contractors can process.

The safety problem is not only false data. It is missing people. A complaint system can undercount people without broadband, stable housing, free time, English fluency, disability access, or trust in city hall. A camera network can overcount visible behavior while missing indoor harm, landlord neglect, heat stress, or informal care. A digital twin can simulate traffic while ignoring fear, displacement, mourning, memory, and the political meaning of a place. The absence of a signal is not evidence that nothing happened.

That is why resident testimony is not a soft supplement to urban data. It is a counter-instrument. A city that cannot reconcile sensor evidence with lived evidence will gradually learn to privilege the record it can query over the person it is supposed to serve.

The city also needs a rule for nulls. No report is not no harm. No camera event is not no conflict. No app use is not no need. No complaint in English is not no complaint. Every urban dataset should carry an account of who could not or would not enter the record, and every dashboard should identify the decision it is not qualified to make.

The Machine-Learning City

The chapter on machine learning is the book's most direct bridge to contemporary AI governance. Green explains predictive systems through concrete municipal examples, including crime analysis and predictive policing. The useful case is not magic automation. It is a bounded collaboration in which a model supports expert inquiry, inherits assumptions from human practice, and remains tied to the question being asked.

The dangerous case appears when vendors and agencies treat prediction as neutrality. Historical data is not raw truth. It records enforcement patterns, reporting patterns, institutional discretion, social inequality, and the categories that agencies already use. When machine-learning systems learn from those records, they can convert old priorities into new objectivity.

This is the recurring trap of model-mediated governance: the system appears to remove human bias by replacing discretion with computation, but it may actually preserve discretion in the training data, objective function, feature design, deployment context, and interpretation of outputs. The decision becomes harder to contest precisely because it now looks technical.

Green's later work on government algorithms sharpens the point: "human oversight" is weak if the human cannot see enough evidence, lacks time, is punished for override, or is asked to legitimate a system the institution already decided to use. In city government, oversight has to be institutional: the agency must justify the system before procurement, test whether people can meaningfully use or challenge it, and preserve the authority to narrow, pause, or retire it.

A city should therefore review the loop, not only the model. A policing, inspection, benefits, or traffic system changes future data by changing where staff go, which cases receive attention, which people learn to avoid contact, and which records become official memory. The evaluation question is not only "was the prediction accurate?" It is also "what did the prediction cause the city to see next?"

The AI-Age Reading

Since the book appeared, the smart-city stack has expanded into a broader AI governance stack. The same pattern now appears in municipal chatbots, automated benefits screening, camera analytics, traffic optimization, school risk systems, procurement tools, public-comment processing, emergency dispatch, and digital twins. The city is no longer only sensed. It is summarized, predicted, simulated, and answered back through interfaces that can sound authoritative.

Generative AI makes Green's warning sharper because it adds a conversational layer to the instrumented city. Residents may not see the database, model, vendor contract, retention policy, or uncertainty threshold. They may encounter a confident answer from a public chatbot, a risk score embedded in a workflow, or a service portal that silently routes their case. The city becomes legible to the system while the system becomes less legible to the resident.

The official sources named above have caught up with much of Green's concern, but their value depends on implementation. A city can publish an AI strategy and still procure opaque systems. It can maintain an algorithm register and still omit the tools residents most need to understand. It can run an impact assessment and still ignore testimony from people who will bear false positives. The smart-enough test is whether these artifacts change deployment decisions, not whether they decorate them.

The post-2019 smart-city record also gives Green's argument a cautionary case. Sidewalk Labs said in May 2020 that it would no longer pursue Toronto's Quayside project, and Waterfront Toronto confirmed the withdrawal. That episode should not be flattened into a slogan that every urban technology project fails. Its more useful lesson is institutional: data governance, public trust, financial viability, public authority, and local legitimacy are not side issues that can be solved after a vendor's urban operating system is already imagined.

A smart enough AI city would start from different questions. Is the public problem actually technical? Who benefits if it is defined that way? What nontechnical reform is being postponed? What data should not be collected? Who can appeal? Who maintains the system? Which vendor incentives shape the interface? What public capacity is lost when the city rents intelligence rather than building accountable institutions?

Governance and Safety

The governance implication is that smart-city review has to begin before the procurement pitch. A city should publish a problem statement that names the public duty, the affected residents, the nontechnical alternatives considered, the data proposed, the legal authority, the vendor dependency, and the office with power to stop the project. If that statement cannot be written plainly, the system is not ready for purchase.

For AI and algorithmic systems, the minimum control set is concrete: a public inventory entry, an algorithmic impact assessment, privacy and civil-rights review, accessibility review, security review, data-retention limits, model and data documentation, pilot boundaries, independent evaluation where stakes justify it, public notice, appeal or correction routes, incident reporting, and sunset criteria. The city should keep a non-digital or human-access channel for essential services, because civic rights should not depend on passing through a vendor interface.

A useful minimum artifact is a civic technology file: problem statement, legal authority, affected residents, alternatives considered, data categories rejected, vendor and subcontractor list, model or rule version, public owner, retention schedule, accessibility test, security review, public notice, appeal path, incident contact, evaluation results, renewal date, and exit plan. The point is not paperwork for its own sake. It is to keep the city from forgetting where public judgment moved.

Procurement is part of safety. OMB M-25-22 is federal guidance, not a city law, but its logic travels well: buyers need data portability, interoperability, vendor-lock-in protections, documentation, performance monitoring, and disclosure when contractors use AI. A city that cannot export its data, inspect system changes, investigate incidents, or switch vendors has not bought intelligence. It has leased dependency.

Contracts should also separate municipal data uses. Data needed to operate a service is not automatically available for vendor product improvement, model training, benchmarking, resale, or cross-client analytics. Public records, resident complaints, mobility traces, emergency calls, permit applications, and service histories should not become a training commons because the procurement team accepted a default cloud term.

Public-space systems deserve extra caution because refusal is hard. A person can avoid an app more easily than a camera-lined street, transit gate, license-plate reader, curb-management sensor, or automated dispatch workflow. The higher the unavoidable exposure, the stronger the burden should be for data minimization, public notice, local democratic authorization, and deletion or retention limits.

Impact assessment should also have consequences. If the evidence shows unreliable performance, inaccessible design, disparate burden, weak appeal, excessive retention, vendor lock-in, or loss of public capacity, the remedy should be delay, redesign, narrowing, human staffing, or cancellation. An assessment that cannot stop a deployment is a ritual, not governance.

Where the Book Needs Care

The book's restraint is also its limitation. Green gives a pragmatic civic framework rather than a full political economy of urban technology markets. Readers looking for a deeper account of real estate, austerity, procurement capture, platform monopoly, or metropolitan inequality will need companion texts. The smart-city problem is not only bad design ideology; it is also a question of who funds infrastructure, who owns data, who can litigate, and who gets to leave when experiments fail.

Even so, that practical register is part of the book's usefulness. It is written for people close enough to municipal work to make decisions: city staff, civic technologists, planners, vendors, researchers, and public-interest advocates. It does not ask them to reject technical systems wholesale. It asks them to stop letting technical possibility set the agenda.

The book also predates the current generative AI boom, the data-center backlash around AI infrastructure, and the spread of AI procurement into ordinary public administration. Its method survives that change, but the object has grown. The smart city is no longer only a plan for a sensor-rich district. It can now be a cloud contract, a chatbot front desk, a camera model, a benefits triage workflow, an automated records search, a digital twin, or a vendor suite quietly embedded inside city operations.

What This Changes

The most important lesson is that a public institution can be captured by its own interface. Once a city learns to see through tickets, sensors, apps, rankings, and predictions, the administrative image can begin to outrank the resident's account of reality. The map starts asking to be obeyed.

The Smart Enough City gives a useful discipline for resisting that drift. Put technology in its place. Treat models as instruments, not authorities. Keep politics visible. Preserve friction where friction protects people. Build channels for refusal, appeal, repair, deletion, and public memory. Above all, do not confuse a city that is easier to compute with a city that is easier to live in.

That discipline turns recurring site themes into municipal controls. Public memory becomes a register of deployed systems, not a vague promise of transparency. Human oversight becomes a workflow with time, evidence, authority, and override power. Vendor governance becomes exit rights, audit access, model-change notice, and data portability. Source discipline becomes a separation between vendor claims, official records, evaluation evidence, resident testimony, and legal duties.

That makes the book a necessary review for the AI era. The future city will be full of models. The question is whether those models remain servants of public judgment or become the quiet grammar through which public judgment is allowed to speak.

Source Discipline

This review separates four kinds of sources. Book claims come from MIT Press, the open-access edition, and author or university pages. Current governance claims come from official policy, standards, and regulator sources such as OECD, OMB, EUR-Lex, NIST, and ISO. Interpretive claims are this site's analysis of what Green's framework means for AI-era city systems. Vendor decks, smart-city marketing pages, and press releases can show how a system is sold, but they cannot prove democratic legitimacy, safety, fairness, or public benefit.

For any city technology claim, the evidence record should name the system, vendor, contract, data source, affected population, evaluation method, authority to act, appeal path, retention rule, and update history. A pilot announcement is not an audit. A dashboard is not a public record by itself. A model card is not an impact assessment. A procurement award is not democratic consent.

Municipal AI strategies and algorithm registers should be read with the same discipline. They are evidence of a governance practice, not proof that every listed system is fair, effective, or complete. They are strongest when they connect a named system to an accountable office, an affected population, data flows, human oversight, appeal, feedback, and update history.

Legal and policy sources should also be kept in their lane. OMB memoranda bind covered federal agencies, not every municipality; the EU AI Act applies by role, market, use case, and timeline; NIST and ISO provide risk-management and impact-assessment frameworks rather than proof that a particular system is safe. This article 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