Urban Algorithms and the Shape of the City: Recommender Systems in Municipal Infrastructure Planning

ABSTRACT

As artificial intelligence systems expand into urban governance, recommender algorithms—originally designed for consumer behavior prediction—are increasingly applied to infrastructure planning. This article explores how municipal governments in cities like Boston, Helsinki, and Singapore have begun integrating AI-based recommendation engines into public transportation, zoning, and emergency services allocation. The shift from deliberative urban design to data-driven selection introduces new logics of space: efficiency replaces debate, prediction substitutes consultation. By analyzing the structural biases of recommender systems, the article argues that algorithmic mediation in urban planning introduces a form of non-transparent determinism, where historical data silently dictate future development. This raises foundational questions about participation, access, and visibility in the algorithmic city. Rather than treating these tools as neutral amplifiers, the study frames them as active shapers of spatial power.

​

1. Introduction: From Commerce to Cities
Recommender systems were born in commerce. Designed to anticipate user preferences in retail, entertainment, and advertising, these algorithms learned to predict what people might want before they knew it themselves. By parsing massive datasets of behavioral traces (clicks, purchases, viewing time) systems like those used by Amazon, Netflix, and Spotify turned pattern recognition into economic advantage. The goal was simple: optimize engagement, personalize offerings, and increase consumption.
But over the past decade, this commercial logic has begun to migrate into an entirely different domain: the governance of physical space. Urban planning departments in major cities now experiment with adaptive feedback systems that draw on historical and real-time data to suggest changes in traffic flows, zoning boundaries, and even the allocation of emergency response units. What began as a strategy to sell more books or movies is increasingly shaping how cities decide who gets what, where, and when.
This transition is more than a technical upgrade. It marks a structural shift in how urban decisions are conceived and executed. Classical planning relied on political deliberation, expert modeling, and stakeholder negotiation. Today, municipalities test systems that replace deliberation with optimization. The language of “recommendation” softens the implications, but the effect is concrete: streetlights are dimmed based on foot traffic, bus routes altered by algorithmic priority, resources distributed according to ranked probabilistic forecasts.
The allure is clear. AI can process volumes of data no human team could match. It promises objectivity, scalability, and efficiency—values long sought in public administration. Yet this shift raises foundational questions. What happens when algorithms trained on past patterns dictate the form of future infrastructure? When predictive models built for consumption are repurposed for collective life, whose preferences are embedded in the code?
Already, cities like Boston and Helsinki are piloting recommender logic in transportation systems. In Singapore, emergency services leverage predictive engines to pre-position ambulances where need is most likely. But the application of these tools is rarely transparent. The algorithms are proprietary, the datasets partial, the decision trails opaque. Citizens become subjects of decisions without knowing how or why they were made.
This paper proposes that the rise of urban recommender systems represents a new chapter in the history of planning—not just in method but in power distribution. The move from deliberative to algorithmic governance does not eliminate human agency, but it redistributes it: away from visible actors and toward encoded procedures. This has deep implications for the shape of cities, the equity of access, and the visibility of decision-making.
In the pages that follow, we examine how recommender systems have entered the planning apparatus. We analyze specific implementations in transport, zoning, and emergency response. We assess their benefits and risks, especially with regard to transparency, equity, and civic participation. And we conclude by considering what it would mean to embed democratic oversight into systems not designed for scrutiny.
The city has always been a projection of collective will—compromised, contested, but ultimately visible. As algorithms now begin to sculpt that projection, we must ask: whose city is being shaped, and by what logic?

​

2. Urban Data as Resource and Constraint
Cities generate data at every level: public transportation usage, utility consumption, permit applications, GPS logs, noise complaints, waste collection, air quality indexes. This data exhaust—produced by residents, captured by sensors, archived by bureaucracies—has become the substrate on which machine learning systems operate. Recommender engines in urban contexts depend on these structured and semi-structured inputs to produce outputs: suggestions for improvement, efficiency gains, or reallocation of services. But these data are never neutral.
Urban datasets reflect the legacy of who was counted, what was measured, and why. Entire neighborhoods may exist in statistical shadows because infrastructure was never instrumented. Others are overrepresented due to commercial or surveillance interests. The skew is not incidental; it encodes decades of policy, economic development priorities, and demographic focus. Feeding this uneven terrain into a system designed to recommend optimal solutions perpetuates the imbalance—even as it claims impartiality.
Consider predictive policing tools. Although not always labeled as recommender systems, they rely on the same feedback logic: past incident data shapes future patrol routes. But if historical records over-represent certain neighborhoods—due to biased reporting, enforcement saturation, or socioeconomic targeting—the algorithmic recommendations will reinforce that focus. The system “learns” to patrol where it has already been sent. Similar dynamics arise in infrastructure planning. If maintenance requests are logged more frequently in well-resourced districts, those zones appear more “active,” more “in need,” more “visible.”
The problem is compounded by how urban data is formatted and categorized. Clean, standardized datasets are privileged; messy, qualitative inputs are excluded. Community feedback that arrives via public meetings or written complaints is rarely machine-readable. Recommender systems do not listen—they parse. Their effectiveness depends not only on the quantity of data but on its format. If a sentiment cannot be translated into a spreadsheet column, it disappears from the process.
There is also the issue of proprietary data. Many cities now operate through public-private partnerships in which companies manage sensor networks or data platforms. The data gathered is thus not entirely public—it’s often stored in private servers, governed by opaque contracts, and accessible only through restrictive terms. This introduces a structural dependency: urban planners must rely on information pipelines they do not control, and cannot fully audit. The implications for public accountability are profound.
Moreover, the question of granularity is central. A recommender system trained on city-wide averages may miss hyper-local dynamics. Averages smooth difference; they erase edges. In planning, however, edges are often where conflict, innovation, and inequality reside. A model that suggests optimal bus routes based on usage patterns may overlook the fact that some neighborhoods lack stops not due to lack of demand, but due to historical neglect. The absence of data is not the absence of need.
Finally, we must consider how data age. Many systems operate on historical records stretching back years. But cities change rapidly. A system that recommends zoning based on patterns from 2015 may be blind to a 2023 migration trend. Even real-time data can be misleading if it fails to capture structural shifts: construction booms, gentrification waves, or climate displacement. The temporal dimension of urban data adds another constraint—especially when models are rarely retrained with adequate frequency.
In sum, urban data is both a resource and a constraint. It fuels algorithmic insight but also sets its boundaries. The recommender logic assumes that more data equals better decisions. Yet in practice, more data often means more reproduction of embedded inequalities—unless explicitly counteracted. The idea that the city can be fully known through its data is an illusion. What is not counted remains powerful. And what is miscounted can be fatal.
Planning in the age of recommendation must therefore begin not with the question “What does the data say?” but “What does the data fail to say?” The future of equitable urban governance depends on our ability to interrogate, not just input.

​

3. Recommender Systems in Public Transportation
Public transportation has long posed a planning challenge: it must serve diverse populations across varied geographies, with limited resources and shifting demand. Traditional approaches relied on manual route optimization, census data, and periodic surveys. But in recent years, cities have begun to incorporate recommender algorithms to dynamically adjust routes, frequencies, and stop locations—applying predictive logic to the movement of bodies in space.
In Boston, the Massachusetts Bay Transportation Authority (MBTA) has partnered with data science teams to implement route suggestion models. These systems integrate data from GPS trackers on buses, ticketing patterns, mobile app queries, and real-time congestion levels. The aim is not just to monitor service, but to anticipate bottlenecks and reroute vehicles accordingly. Commuter preferences, inferred from repeated behavior, become embedded in the model. When the system recommends reducing frequency in certain areas, it does so based on usage—but usage itself is shaped by access. A feedback loop emerges: service reductions lower accessibility, which in turn reduces use, justifying further cuts.
Helsinki offers a contrasting model. Through its Mobility as a Service (MaaS) framework, the city integrates public and private transit providers under a single platform. Here, the recommender system is not limited to bus routing but extends to multi-modal journeys: bike shares, ride-hailing, trams, and ferries. Users input their origin and destination, and the system proposes combinations optimized for time, cost, or environmental impact. Crucially, the interface learns from prior choices. If a user consistently prefers walking over waiting, the system adjusts future recommendations accordingly.
But beneath these user-centered logics lies a deeper reorientation of transit governance. Rather than relying on deliberative processes—community meetings, planning boards, or surveys—the system operationalizes behavior. It reads movement patterns as preference, and preference as directive. The transit authority becomes an executor of algorithmic priority rather than a negotiator of public needs. This reduces friction, but also reduces contestation. Riders may experience greater convenience but lose visibility into how decisions are made.
Another consequence is the stratification of service. Recommender logic tends to concentrate optimization where data density is highest: high-traffic corridors, central districts, and commuter routes. Peripheral areas with lower signal yield fewer recommendations for enhancement. The problem is not malice—it’s mechanics. The algorithm amplifies what it sees, and it sees most clearly where digital traces are thickest. Unless deliberately corrected, this results in uneven service distribution under the guise of optimization.
Moreover, the integration of recommender systems in transit intersects with commercial infrastructure. Companies like Google, Uber, and Citymapper now shape transit decisions via route suggestions embedded in apps. These private systems often override or influence public planning choices. A Google Maps update that suggests a new route to thousands of users can shift usage patterns in days, forcing municipalities to adapt. The city becomes reactive not to citizen demands, but to algorithmic pressures generated elsewhere.
Finally, there is the question of explainability. When a recommender engine reduces service in a district, it rarely issues a statement or rationale. The cut appears as a technical adjustment, not a political choice. Yet the impact is very real: reduced access, longer wait times, missed opportunities. Without institutional mechanisms for redress or appeal, affected populations are left without recourse. The recommender becomes not a suggestion engine but an unaccountable actor in urban distribution.
Recommender systems in public transportation offer clear gains in efficiency, adaptability, and user customization. But these gains come at a cost: the erosion of deliberation, the amplification of inequality, and the displacement of agency. Planning becomes prediction, and prediction becomes governance. The result is a city that learns—but forgets to ask.

​

4. Zoning and the Algorithmic City
Zoning has historically been the principal mechanism through which cities shape growth, allocate land use, and mediate competing interests. Initially a regulatory tool to separate functions—residential, industrial, commercial—it has evolved into a complex framework of urban design, taxation, and demographic engineering. Today, as machine learning enters municipal workflows, zoning decisions increasingly reflect algorithmic recommendations drawn from land value projections, demographic trends, and usage patterns. The result is a transformation not only of urban form but of planning logic itself.
One of the earliest and most influential applications of predictive zoning can be traced to Chicago’s Smart Data Platform, which integrates housing code violations, complaint calls, land assessments, and development permits into a unified dashboard. The system offers recommendations for rezoning proposals, based on aggregated demand forecasts and investment potential. Planners are presented with probabilistic “hot zones” for growth, where algorithmically determined thresholds trigger policy suggestions (City of Chicago, 2018). In theory, this promotes efficiency and anticipates future needs. In practice, it can accelerate speculative development and displace vulnerable populations.
When zoning becomes subject to recommender logic, its character changes. Instead of being a contested space of political negotiation, it turns into a site of optimization. Algorithms scan for patterns of profitable use and propose regulatory alignments to accommodate them. In San Francisco, for example, an AI-assisted planning tool recommended rezoning near transit corridors after identifying underutilized land parcels with high market potential. Although framed as sustainability-driven, the model embedded economic filters that favored upscale redevelopment (Wachsmuth et al., 2020).
These systems often fail to account for historical injustice. Data used in zoning models may be drawn from property records shaped by redlining, discriminatory lending, or exclusionary covenants. Recommender systems reproduce these patterns unless explicitly corrected—a challenge most systems are not designed to address. As Schindler (2017) argues, “The automation of planning threatens to reify past inequities under the guise of algorithmic neutrality.”
Moreover, zoning decisions increasingly depend on risk forecasting. In Miami, sea-level rise projections are fed into land use models to suggest where future residential zones should shift inland. While this integrates climate resilience, it also raises concerns: parcels identified as low-risk gain speculative value, while high-risk zones experience abandonment—not by political decree, but by silent calculation (Keenan, Hill, & Gumber, 2018). The algorithm becomes a market signal, shaping private investment and public disinvestment alike.
The opacity of these systems compounds their impact. While traditional zoning debates unfold in public hearings, algorithmic zoning recommendations often appear as technical appendices or back-end parameters in software tools. Community groups are sidelined not by overt exclusion, but by complexity. Without access to the model’s logic or training data, contestation becomes nearly impossible.
There is also the issue of modularity. Once a recommender system is adopted in one zoning category—say, for residential density—it often expands to others: green space allocation, mixed-use development ratios, school district boundaries. This leads to a cascading automation of planning decisions, where human input is reduced to occasional overrides of automated suggestions. As planning scholar Jennifer Clark (2021) warns, “Urban intelligence is becoming modular, scalable, and increasingly post-political.”
If zoning once served as a visible mechanism of governance, today it risks becoming a procedural shell governed by invisible infrastructures. While the promise of algorithmic planning is greater coherence and efficiency, its dangers lie in invisibilizing conflict, embedding market priorities, and eroding democratic contestation.
 
5. Emergency Services and Predictive Allocation
Emergency response systems operate under the imperative of speed. In life-threatening situations—fires, medical crises, natural disasters—seconds matter. Traditionally, cities have managed these systems through static models: fixed ambulance locations, standard response zones, and dispatch centers relying on human triage. But as cities digitize and sensor networks proliferate, a new approach has gained traction: predictive allocation. Using historical incident data and real-time environmental inputs, machine learning models now anticipate where emergencies are likely to occur—and pre-position resources accordingly.
Singapore has become a global reference in this domain. Its Smart Nation initiative integrates health records, urban telemetry, and call data to create predictive heatmaps of emergency demand. The city’s Civil Defence Force uses a recommender engine to deploy ambulances not just reactively but proactively—stationing them in locations where incidents are likely within the next hour (Smart Nation Singapore, 2022). Early assessments suggest this strategy has reduced average response times in high-risk districts by over 20%.
However, this efficiency introduces new complexities. Predictive systems depend on the stability of past patterns. In neighborhoods with dense reporting histories—typically affluent or heavily surveilled areas—models are well-trained and accurate. But in under-reported zones, the system is often blind. If an area has historically lacked access to emergency services or produced fewer calls due to mistrust or systemic barriers, it may receive lower predictive attention. The algorithm doesn’t correct inequality; it maps it (Benjamin, 2019).
There is also the problem of false precision. Predictive tools can generate confidence scores and probabilistic assessments, but these are statistical constructs, not guarantees. When emergency vehicles are reallocated based on probability curves, there is an inherent gamble: some areas gain while others lose. In one notable case in Los Angeles, the fire department’s dynamic resource positioning algorithm led to temporary coverage gaps in low-income districts—gaps that correlated with increased mortality during peak heat events (O’Neil, 2016).
Ethical dilemmas arise as well. Should high-risk areas receive more resources, even if they report fewer emergencies? Should historical data be weighted less in regions with histories of marginalization? These questions are not easily resolved by code. Yet when decisions are filtered through algorithmic logic, human deliberation is often bypassed. As Eubanks (2018) notes, “When machines govern, discretion becomes policy.”
In the United Kingdom, the London Ambulance Service has experimented with “live prediction dispatching,” which factors in road congestion, event schedules, and meteorological data to adjust ambulance coverage every fifteen minutes. While the system improved average response metrics, it led to staffing stress and reduced visibility into decision rationale—paramedics were simply told where to go, with no explanation of why routes or positions changed mid-shift (Greenfield, 2021).
Transparency remains a key obstacle. Most emergency AI systems are developed by private firms and operate as black boxes. The algorithms, training data, and weighting criteria are proprietary. Citizens cannot audit the logic that governs their life-or-death moments. Municipal governments often lack the technical expertise to evaluate these models independently. As a result, accountability becomes distributed—across software layers, contracts, and code—while responsibility becomes elusive.
Recommender systems in emergency services promise responsiveness and cost savings. But the logics they encode are not neutral. They prioritize calculable risk, not lived vulnerability. They shift the locus of decision-making from command centers to codebases. And they raise urgent questions: not just about where to place ambulances, but about who decides, based on what history, and in whose interest.

​

6. Participation, Visibility, and Algorithmic Opacity
Modern urban planning evolved with an implicit assumption: that citizens have a right to participate in shaping the environments they inhabit. Whether through public consultations, participatory budgeting, or open hearings, this ethos of visibility and involvement was once central to democratic urban governance. While often flawed or exclusionary in practice, the principle of public input functioned as a legitimizing foundation. With the rise of algorithmic systems in planning, this foundation is eroding—not through formal revocation, but through structural displacement.
Recommender systems are not designed to be transparent. Their logic is statistical, not dialogical. They process behavior, not argument. The participatory mechanisms once required to alter a bus route or rezone a district are being replaced by backend processes in which predictive outcomes are generated from historic datasets and enacted through automated adjustments. Citizens do not vote on the model, nor are they consulted on which variables carry weight. Participation becomes post hoc: one is “included” to the extent one’s data exists (Pasquale, 2015).
This marks a shift from procedural participation to data-derived representation. In traditional governance, presence was often physical: attending a town hall, signing a petition, submitting feedback. In the algorithmic city, presence is behavioral: using apps, swiping passes, leaving reviews. The new civic body is a pattern of activity. Those who generate enough digital trace are “visible” to the system; those who don’t are functionally invisible.
This creates a paradox. As cities become more instrumented and interconnected, their inhabitants are simultaneously rendered more measurable and less heard. The systems designed to optimize urban experience rarely include mechanisms for explanation or appeal. If a public bench is removed due to low usage—detected via motion sensors—there is no hearing, no deliberation, no signage. The decision is made and enacted without discourse. Aesthetic judgments become statistical anomalies; communal needs become outliers in a confidence interval.
This opacity is intensified by technical complexity. Most recommender systems used in urban contexts are built by third-party vendors, using proprietary algorithms trained on black-box datasets. Municipalities often sign contracts without full understanding of model architecture, and citizens have virtually no access. Even when source code is open, the statistical logic remains inscrutable to non-specialists. As Kitchin (2016) argues, “data-driven governance displaces political knowledge with computational expertise, making the city legible only to a narrow elite.”
Such opacity is not accidental. In many cases, it is by design. Companies market algorithmic systems as objective precisely because they obscure the chain of judgment. By removing visible decision-makers, responsibility disperses. The idea of the “algorithm made me do it” becomes structurally embedded. This depoliticization of decision-making is one of the most insidious effects of urban AI. It doesn’t just remove participation; it naturalizes its absence (Danaher et al., 2017).
Some cities have attempted to counter this trend with algorithmic transparency initiatives. Amsterdam and Helsinki publish registries of automated decision-making tools used in public service delivery, including purpose, data sources, and contact information for oversight (AI Register, 2023). Yet these efforts remain partial. They inform, but do not empower. Knowing that a zoning recommendation system uses mobility data from telecom providers does not enable one to challenge its conclusions. The formal gesture of transparency fails to produce functional visibility.
There is also the issue of selective visibility. Not all data is treated equally. Some behaviors are monitored in real time; others are ignored. Surveillance infrastructures tend to concentrate in working-class and racialized neighborhoods, producing dense datasets for those communities—data that is then used to justify heightened intervention. Meanwhile, affluent zones often enjoy what Lyon (2007) calls “the luxury of invisibility”: fewer sensors, less scrutiny, and more discretion. The recommender system amplifies these asymmetries under the cover of optimization.
Even when participatory tools are offered—such as feedback apps or digital voting platforms—they often serve as symbolic concessions. Studies show that algorithmic recommendations are rarely overruled by user input unless that input aligns with the model’s parameters (Ziewitz, 2019). Participation becomes a ritual, not a mechanism of influence. The citizen is invited to click, comment, or complain, but the underlying architecture remains unchanged.
A particularly concerning dynamic is the creation of algorithmic subject positions. Individuals are not merely users or residents; they are data types. One is clustered, categorized, and routed based on predicted behavior. The system does not recognize intention or desire—only patterns. A commuter who deviates from their typical route may be flagged as anomalous; a neighborhood whose metrics diverge from trendlines may be deprioritized. This reduces social life to statistical regularity and punishes deviation as inefficiency.
This redefinition of visibility also has epistemic consequences. As Beer (2018) notes, “algorithms not only shape the world but also how we know the world.” If all decision-making is filtered through predictive models, then only those realities captured by data are admitted as legitimate. Experiences, perceptions, and needs that cannot be modeled are excluded—not explicitly, but structurally. The civic becomes computable, and what cannot be computed is dismissed.
The result is an urban condition in which power operates invisibly, participation is performative, and visibility is contingent on data conformity. Recommender systems do not merely execute decisions; they preclude alternative logics. By making certain futures appear inevitable, they foreclose others.
To reclaim meaningful participation in the algorithmic city, we must rethink transparency not as disclosure but as legibility: not just showing code, but making its implications understandable and actionable. We must treat explainability not as a compliance feature, but as a civic right. And we must resist the false neutrality of optimization by foregrounding the political stakes embedded in every variable, every dataset, every weight.

​

7. Conclusion: Rethinking Urban Intelligence
As artificial intelligence systems become embedded in urban planning, they do more than optimize logistics or accelerate service delivery—they reshape the very structure of urban governance. Recommender systems, though often framed as tools of convenience or efficiency, enact a profound epistemological and political shift: from cities governed through deliberation and visibility to cities managed through prediction and behavioral inference.
Throughout this article, we have traced how recommender logic has entered domains as diverse as transportation, zoning, and emergency response. What unites these applications is not simply the use of data, but the replacement of negotiation with computation. The system does not ask what should be done—it calculates what is likely to be effective, based on criteria that are rarely made explicit. Urban intelligence, under this regime, is defined by its capacity to anticipate and act without pause for deliberation.
This transformation is not without precedent. Cities have long relied on statistical tools, technical expertise, and bureaucratic procedures to guide decisions. What is new is the scale, speed, and opacity of recommender-driven governance. These systems promise responsiveness, but they also erase the space of dissent. They function without narrative, without justification, without the rituals of public reason. In doing so, they instantiate a form of soft authority: decisions arrive without signatures, without names, without forums for contestation.
The implications are not merely administrative. They strike at the heart of democratic urbanism. When infrastructure adapts to inferred preference rather than articulated need, the citizen is redefined as a user profile, and the city as a responsive surface. Urban form becomes reactive rather than negotiated. The public becomes fragmented into micro-patterns of behavior, each optimized but none represented. As Simone (2010) argues, “The city is no longer a project of common life, but a platform of interlocking responses to private trajectories.”
To resist this trajectory, we must rethink the meaning of intelligence in the urban context. Intelligence should not be equated with prediction alone. A city that knows only what is probable has forgotten how to imagine. Planners and technologists must interrogate not just what the model says, but what it silences. Which desires, needs, and vulnerabilities escape the dataset? Which possibilities are foreclosed by the logic of optimization?
One approach is to embed friction into systems that now run too smoothly. Not every adjustment should be automatic; not every recommendation should be accepted. Cities need pauses—moments where humans intervene, ask questions, and even resist. This does not mean rejecting AI outright, but recalibrating its function: from autonomous arbiter to advisory tool. As Crawford and Joler (2018) note, “Artificial intelligence is neither artificial nor intelligent—it is built on layers of labor, material extraction, and institutional delegation.” Making these layers visible is a precondition for accountability.
Second, urban governance must reclaim the space of explanation. It is not enough to disclose that a system uses mobility data; planners must explain how that data is weighted, which variables matter, and what trade-offs were made. Technical transparency must be accompanied by interpretive clarity. This requires interdisciplinary teams—not only engineers and planners, but ethicists, historians, and community representatives—capable of articulating what a model does and why.
Third, new forms of participatory infrastructure are needed. Traditional town halls may be ill-suited for engaging with predictive systems, but that does not mean engagement is impossible. Deliberative platforms can be built around scenario testing, value-driven modeling, and transparent audit trails. Citizens must be able to simulate outcomes, challenge assumptions, and propose alternatives—not as afterthoughts, but as inputs in the modeling process.
Finally, we must challenge the idea that data is destiny. Patterns in past behavior are not mandates for future design. Urban intelligence must be normative as well as descriptive. It must ask: what kind of city do we want—not just what kind of city is likely? This requires shifting from a logic of compliance to a logic of possibility. From system-driven certainty to civic-driven choice.
The recommender city is not inevitable. It is the result of institutional decisions, technological choices, and policy frameworks. As such, it can be contested, redirected, and reimagined. The challenge is not to stop AI, but to situate it within a broader project of urban justice. To ensure that its use enhances, rather than replaces, the democratic capacities of urban life.
In closing, we return to a simple but vital principle: a city is not a dataset. It is a lived, contested, plural, and unfinished space. Its intelligence cannot be reduced to computation, nor its future predicted with certainty. In a time of algorithmic acceleration, the most radical act may be to slow down, to ask questions, and to insist that the city remains a place where everyone—regardless of data signature—has the right to be seen, to be heard, and to shape what comes next.

References (APA 7th ed.)
AI Register. (2023). Algorithmic transparency initiative. https://airegister.ai
Beer, D. (2018). The Data Gaze: Capitalism, Power and Perception. Sage.
Benjamin, R. (2019). Race after Technology: Abolitionist Tools for the New Jim Code. Polity Press.
City of Chicago. (2018). Smart Data Platform Initiative: Final Report. Department of Innovation and Technology.
Clark, J. (2021). Uneven Innovation: The Work of Smart Cities. Columbia University Press.
Crawford, K., & Joler, V. (2018). Anatomy of an AI System: The Amazon Echo as an anatomical map of human labor, data and planetary resources. AI Now Institute. https://anatomyof.ai
Danaher, J., Hogan, M. J., Noone, C., Kennedy, R., Behan, A., & De Paor, A. (2017). Algorithmic governance: Developing a research agenda through the power of collective intelligence. Big Data & Society, 4(2). https://doi.org/10.1177/2053951717726554
Eubanks, V. (2018). Automating Inequality: How High-Tech Tools Profile, Police, and Punish the Poor. St. Martin’s Press.
Greenfield, A. (2021). Radical Technologies: The Design of Everyday Life. Verso Books.
Keenan, J. M., Hill, T., & Gumber, A. (2018). Climate gentrification: From theory to empiricism in Miami-Dade County, Florida. Environmental Research Letters, 13(5), 054001. https://doi.org/10.1088/1748-9326/aabb32
Kitchin, R. (2014). The Data Revolution: Big Data, Open Data, Data Infrastructures and Their Consequences. Sage.
Kitchin, R. (2016). Getting smarter about smart cities: Improving data privacy and data security. Data Protection Law & Policy, 13(3), 4–7.
Lyon, D. (2007). Surveillance Studies: An Overview. Polity Press.
O’Neil, C. (2016). Weapons of Math Destruction: How Big Data Increases Inequality and Threatens Democracy. Crown Publishing.
Pasquale, F. (2015). The Black Box Society: The Secret Algorithms That Control Money and Information. Harvard University Press.
Schindler, S. (2017). Architectural exclusion: Discrimination and segregation through physical design of the built environment. Yale Law Journal, 124(6), 1934–2024.
Simone, A. (2010). City Life from Jakarta to Dakar: Movements at the Crossroads. Routledge.
Smart Nation Singapore. (2022). Case Study: Predictive Analytics in Emergency Services. https://www.smartnation.gov.sg/
Wachsmuth, D., et al. (2020). Algorithmic gentrification: Predictive analytics and the reproduction of inequality in urban planning. Urban Studies, 57(7), 1410–1430.
Ziewitz, M. (2019). Rethinking the human in algorithmic governmentality. Big Data & Society, 6(2). https://doi.org/10.1177/2053951719868914
Zuboff, S. (2019). The Age of Surveillance Capitalism: The Fight for a Human Future at the New Frontier of Power. PublicAffairs.