Smart cities, urban big data, digital twins and other buzz terms
Outline
Big data vs. just data
Smart cities
Urban dashboards
Digital twins
Artificial intelligence in cities
Big data vs. just data
Digital traces human activities leave behind
Accidental (Arribas-Bel 2014)
Why?
Computing power
Digital turn
Geospatial technologies [important for urban researchers]
Big data vs. just data
Volume [how big is big?]
Velocity, being created in or near real time
Variety, being structured and unstructured in nature
Exhaustive in scope, capture entire populations or systems
Fine-grained in resolution
Relational in nature, common fields that enable the merging of different data sets
Flexible, extensionable (can add new fields easily) and scalable (can expand in size rapidly) (Kitchin 2013).
Or, whatever doesn’t fit in an excel spreadsheet (Batty, anecdotal)
Big data vs. just data
Biases
Need for new methods
Accidental
Big data can speak for themselves. Not really (Kitchin 2013)
Smart cities, different approaches
The digital technologies approach
Urban infrastructure and services are managed computationally
Networked digital instrumentation embedded into the urban fabric
Continuous streams of data that dynamically feed into management systems and control rooms
New forms of governmentality
Based on Kitchin (2019)
Smart cities, different approaches
The outputs approach
Strategic use of information and communications technology (ICT)
➔ smarter citizens, workers, policy, and programs
➔ innovation, economic development, and entrepreneurship
➔ urban resilience and sustainability
Based on Kitchin (2019)
Smart cities, different approaches
The just approach
ICT-led, citizen-centric model of development
➔ social innovation and social justice, civic engagement and activism
➔ transparent and accountable governance
Based on Kitchin (2019)
Smart cities, different approaches
In reality…
… a blend of all these approached
Rio de Janeiro, Brazil: The first smart city
Smart cities, different approaches
Rio, ten years on: a critical postscript
Built by IBM/Cisco in 2010 and marketed globally as the smart-city control room
Since criticised as “surveillance theatre” and a flawed model of smart-city emergence (Gaffney and Robertson 2018)
Centralised, real-time monitoring did not translate into better everyday urban outcomes for residents; political priority and investment have since receded
Take-away: a control room tells us about power and visibility, not automatically about better governance – keep this in mind for AI-powered versions of the same idea, later in this lecture
Smart cities and data
(Smart) cities generate much more data nowadays
- e-government systems, city operating systems, centralized control rooms, digital surveillance, predictive policing, intelligent transport systems, smart grids, sensor networks, building management systems, civic apps … (Kitchin and Moore-Cherry 2021)
Smart cities and data
Local authorities are under pressure to release open data for:
public scrutiny
civic apps
Data reuse:
internally facing
externally facing
Dashboards
A visual display of the most important information needed to achieve one or more objectives; a consolidated and arranged on a single screen so the information can be monitored at a glance (Few and Edge 2007)
Dashboards
Transparency, open government philosophy, accountability (Young, Kitchin, and Naji 2022)
Improve user’s ’span of control of large and complex data
Share outputs
Dashboards
Not just neutral, technical tools
Translators and engines rather than mirrors
Reductive in nature (vs. the complex nature of cities)
Decontextualize cities (Kitchin, Maalsen, and McArdle 2016)
Bristol dashboard
Source: https://experience.arcgis.com/experience/bcf5a6312bc04ffeb43db67cd57f5439/
NYC Dashboard
Digital twins
A mirror image of a physical process that is articulated alongside the process in question, usually matching exactly the operation of the physical process which takes place in real time
A variety of digital simulation models that run alongside real-time processes that pertain to social and economic systems as well as physical systems
Models are simplifications not replications of the real thing
Based on Batty (2018)
Digital twins
Source: (Bolton et al. 2018)
Digital twins: a concrete (and partial) example
Bristol Harbour digital twin (Mott MacDonald, for Bristol City Council)
UAV, uncrewed surface vessel and LiDAR surveys of 300+ assets across 21km of waterway
Over 1 terabyte of point-cloud, mesh and photographic data, accessed through a web platform for remote condition assessment and deterioration tracking
Genuinely useful: replaces some in-person inspections; supports year-on-year change detection
Artificial intelligence in cities
So far: data as the raw material of smart cities
Now: AI, and especially generative AI, as the tool increasingly used to make sense of, and act on, that data
Narrow AI / machine learning is not new
What is new since c. 2022 is generative AI: Large Language Models (LLMs) such as ChatGPT, Claude, Gemini and Copilot, which can read, write, summarise, code and reason over text, spatial and tabular data too
Agentic AI
These LLMs are one instance of a broader shift towards foundation models: models pretrained on very large, broad data and then adapted to many downstream tasks (Bommasani et al. 2021)
See also Batty (2026), reflecting on AI from within this unit’s own home journal
AI applications for cities: some examples
Planning practice: drafting and summarising planning applications, synthesising public consultation responses, generating scenario narratives for local plans (Zheng et al. 2025; Liu et al. 2025)
Citizen-facing services: e-government chatbots, 311/multilingual service requests, accessibility of council information (Cortés-Cediel et al. 2023)
Urban monitoring: computer vision on street-level and satellite imagery (e.g. detecting pavement damage, green space, informal settlements) – see Biljecki and Ito (2021) for a street-level imagery review
Mobility and infrastructure: demand forecasting, predictive maintenance, incident detection layered on top of the dashboards
Digital twins: generative models increasingly used to simulate “what if” urban scenarios inside twins, rather than just visualising current conditions, like the harbour twin above (Weil and Bibri 2023)
AI’s own urban footprint
AI is not just applied to cities – it also physically lives in them
Every LLM query runs on a data centre somewhere: real buildings, with real electricity, water and land demands
Recent evidence: almost all US data centres sit within or close to existing urban areas, drawn there by pre-existing power grids and fossil-fuel industrial infrastructure, not built on empty land (Ancona et al. 2026)
This is a location-economics question, not just a technology one – think back to agglomeration economies and von Thünen from Lecture 1: AI infrastructure clusters where power, fibre and cooling are already cheap and available
More broadly, economic geographers are now asking how AI itself predicts, prescribes and programs economic futures – reshaping labour, geopolitics and regional development unevenly across places (Zook, Loomis, and Lim 2026)
So: the “cloud” has a location, a landlord, and a local planning application
A critical perspective: whose city does the algorithm see?
Recall Kitchin’s three approaches to smart cities from earlier in this lecture: digital technologies, outputs, justice (Kitchin 2019)
AI intensifies all three: more automated infrastructure, more claims of efficiency/innovation gains, but also new governance questions – and, as with Rio, new versions of the control-room logic, now AI-powered
Tedeschi (2024): datafication (now increasingly AI-mediated) can produce new forms of urban (in)justice – whose data trains the model, whose behaviour it optimises for, who is legible to it and who is not
Practical risks: bias and unequal error rates across neighbourhoods/groups; opacity and unaccountability of automated decisions; labour and environmental costs of large models (see the data centre point above); over-trust in fluent-sounding but wrong outputs (Kerche, Zook, and Graham 2026)
Big data (and now AI) cannot speak for themselves
Recall from earlier: “Big data can speak for themselves. Not really” (Kitchin 2013)
The same applies, arguably even more so, to AI: a fluent answer is not the same as a correct or contextually appropriate one
Your task, in this unit and beyond, is the same whether the tool is a GLM or an LLM: use it critically, understand what it is actually doing, and be able to justify your methodological