Check out the episode:
You can find the shownotes through this link.
Are you interested in the evolution of urban digital twins?
Our debate today works with the article titled What have urban digital twins contributed to urban planning and decision-making? From a systematic literature review toward a socio-technical research and development agenda from 2025, by Shervin Azadi, Dena Kasraian, Pirouz Nourian, and Pieter van Wesemael, published in the MDPI Smart Cities journal.
This is a great preparation to our next interview with Ruthbea Yesner in episode 458 talking about the opportunities with new technologies, such as digital twins for better urban futures.
Since we are investigating the future of cities, I thought it would be interesting to see how digital twins have been helping urban decision-making. This article advocates for urban digital twins that are interdisciplinarily integrated, consensually contextualised and procedurally operationalised.
[intro music]
Welcome to today’s What is The Future For Cities podcast and its Research episode; my name is Fanni, and today we will introduce a research by summarising it. The episode really is just a short summary of the original investigation, and, in case it is interesting enough, I would encourage everyone to check out the whole documentation. This conversation was produced and generated with Notebook LM as two hosts dissecting the whole research.
[music]
Speaker 1: When you look at a traditional architectural blueprint, there is a certain comfort in its precision. It is clean, it’s static, and it behaves exactly the way the math says it should.
Speaker 2: It represents a perfectly closed, predictable system. The variables are just entirely within the engineer’s control.
Speaker 1: Exactly. But then you take that blueprint and lay it over an actual living city, and immediately that closed system breaks wide open. You have competing political interests, historical context, and just messy human agency. A blueprint doesn’t really know what to do with a community that fundamentally disagrees with how a physical space should be used. And today, we are looking at the modern digital equivalent of that blueprint: urban digital twins, or UDTs. A massive new systematic literature review out of the journal Smart Cities by Azadi and colleagues just looked at 88 different peer-reviewed projects from the last seven years, and they found a glaring problem.
Speaker 2: It’s a huge gap.
Speaker 1: Seven years into the hype of these digital replicas, there is just a massive gap between the grand ambitions of UDTs and their actual implementation in real world urban planning.
Speaker 2: And it is a fascinating gap to explore, really because it forces us to ask what these models are actually for.
Speaker 1: Yes, exactly. So the central question we are tackling today is why this gap exists. Are these digital twins failing because we are treating cities as purely technical problems rather than social ones? Or, and this is the flip side, are the current technical achievements exactly the foundational progress we should expect while the proposed social integrations are just frankly completely unrealistic? I argue that UDTs are failing to reach their potential precisely because of this purely technical focus. If we want them to be useful, we have to urgently adopt what the review calls the augmented urban planning framework, which forces the integration of social and procedural complexities.
Speaker 2: And I come at it from a slightly different way. I argue that the current technical achievements in urban digital twins are highly valuable on their own. Burdening an emerging technology with the impossible task of solving deep-rooted human and bureaucratic conflicts, it risks paralysing the actual measurable technological progress we’re making right now.
Speaker 1: Let me start by looking at the hard data from the review because I think it paints a very stark picture of where we are. The authors looked closely at 40 articles that detailed actual physical implementations of UDTs. Out of those 40, exactly zero demonstrated integration into official city planning and decision-making processes. Zero.
Speaker 2: It’s a tough number.
Speaker 1: It is. We are spending massive amounts of computing power and capital building these incredibly sophisticated digital replicas of our cities, but they aren’t making it to the mayor’s desk. The review argues that a purely technical approach inevitably leads to digital universalism.
Speaker 2: Which is a heavy term.
Speaker 1: It is, yeah. But conceptually, it just means oversimplifying the city. It means forcing the vibrant, chaotic reality of a city into a single sterile digital frame, and that inevitably marginalizes underrepresented groups because they just don’t fit neatly into the algorithm I believe UDTs have to pivot to this augmented urban planning agenda. Specifically, they need what the authors call consensual contextualization.
Speaker 2: Which means bringing citizens in.
Speaker 1: Actually bringing citizens into the modelling process and procedural operationalization, which means legally and functionally integrating them into real government workflows. Otherwise, these are just incredibly expensive, utterly useless lab experiments.
Speaker 2: Okay, I see the gap you are pointing to with the zero out of forty statistic, but I have to push back hard on characterizing these as useless lab experiments. Let’s look at what is actually happening in those exact same forty implementation articles. Eighty percent of them successfully focused on interdisciplinary integration. We are talking about major tangible breakthroughs here. They are successfully modelling the intersection of urban water systems with district decarbonization efforts or layering dynamic urban mobility data over energy consumption grids.
Speaker 1: But if city planners aren’t actually using them to make decisions, what is the value?
Speaker 2: Value is that building the technical foundation for interdisciplinary integration is phenomenally difficult. Just getting a water management software to talk to a traffic simulation requires overcoming massive API hurdles, data standardization issues, and computational bottlenecks. Those are massive successes. The paper’s proposed sociotechnical agenda requires an unrealistic level of stakeholder consensus. UDTs are fundamentally meant to be highly sophisticated assessment tools.
Speaker 1: Interesting point, though I guess I would frame it differently because we have to ask what exactly we are assessing. You mentioned the technical achievements and the interdisciplinary integration. Let’s dig into what these models are actually doing right now on the ground.
Speaker 2: The objective data crunching, because the review notes that seventy percent of the implementations successfully included simulation and predictive models. For example, researchers are using UDTs for the dynamic control of district heating networks. They are running virtual pilots of logistic vehicle fleets to assess total cost of ownership, traffic impact, and carbon emissions before a single real truck even hits the road. That objective, incredibly complex data analysis is the true foundational purpose of a digital twin. It processes variables at a scale the human mind simply cannot handle, providing a baseline of objective truth.
Speaker 1: I see why you think that, but let me give you a different perspective on this idea of objective truth. Building a flawless simulation of traffic flow or heat dispersion that doesn’t account for the human beings actually living in that environment, it’s like using a thermometer to measure a city’s culture.
Speaker 2: A thermometer?
Speaker 1: Yes. Think about it. A thermometer measures one objective metric perfectly, temperature. But if you rely solely on that thermometer to understand a neighbourhood, you are going to look at a bustling, vibrant night market and a barren concrete parking lot, and if they are both seventy-five degrees, your model will tell you they are exactly the same. That is the danger of digital universalism. If a digital twin only measures heat exposure or vehicle throughput but completely ignores the citizens’ perception of that space, their preferences, or their agency, it isn’t actually objective. It is just reinforcing the blind spots of the modeler. We embed specific values into the math. If we optimize a model for, say, traffic flow, we might inherently be destroying a neighbourhood that relies on slow streets and pedestrian access, all while the model congratulates us for being efficient.
Speaker 2: That is a compelling argument regarding the limitations of a purely quantitative model. But have you considered the alternative the authors are proposing to fix this? They champion this idea of consensual contextualization. They want citizens deeply involved in the parameter setting of the twins. But the text itself exposes the fatal flaw in that approach. Out of those forty implementation studies we’ve been talking about, do you know how many involved actual real-world citizens?
Speaker 1: The review says three.
Speaker 2: Only three out of forty.
Speaker 1: Exactly.
Speaker 2: Only three, and the review explains exactly why. Real-world collaboration is plagued by deeply unbalanced power dynamics and a total misalignment of terminology. If you put a municipal water engineer and a local neighbourhood activist in the same room to build a model, they don’t even use a word like efficiency to mean the same thing.
Speaker 1: Which is exactly why the review calls for a plurality of UDTs. The solution isn’t to build one single monolithic master twin that sits on a server in city hall and dictates reality. The authors are arguing for multiple stakeholder-specific models. Different communities can curate their own data and co-create a shared understanding of their specific needs.
Speaker 2: But let’s look at the practical reality of trying to program that. The review points out that twenty-nine of those studies relied on hypothetical stakeholders. The researchers didn’t do that because they were lazy. They did it because building what you are describing is a computational nightmare. Creating custom reconfigurable UDTs for every single competing stakeholder group requires dynamic hyperparameters
Speaker 1: For those of us who aren’t machine learning engineers, break down what you mean by dynamic hyperparameters in this context. Why does that cause a nightmare?
Speaker 2: So in modelling, hyperparameters are the fundamental overarching rules that govern how the model learns and processes data. You set them before the simulation runs. If you have infinite competing stakeholder views, you are essentially asking the model to constantly change its own fundamental laws of physics depending on who is looking at the screen. So if community A wants the model to prioritize pedestrian safety above all else, and community B wants it to prioritize commercial delivery speed, the core parameters clash. You can’t just slide a scale back and forth. It fundamentally breaks the mathematical architecture of the simulation. Attempting to build dynamic hyperparameters for every conflicting community view is a recipe for complete paralysis. You would have endless town hall debates over model boundaries and data curation rather than actionable urban planning. If the goal is to reduce carbon emissions in a district heating network before winter, arguing over a bespoke digital representation for every single resident will ensure the network never gets built.
Speaker 1: I’m not convinced by that line of reasoning because ignoring the stakeholders in the name of speed doesn’t make the conflict go away. It just pushes it downstream. If you build a highly efficient district heating network that requires tearing up a neighbourhood, and you never modelled the community’s response to that disruption, the community’s just gonna protest and vote the project down anyway.
Speaker 2: That’s fair.
Speaker 1: The model didn’t save you time. It just gave you a false sense of security. But let’s say for the sake of argument that you and I could actually agree on a model. Let’s say we navigate the hyperparameters and build a twin that a community accepts. We still have a massive problem, which brings us to the third pillar of the review’s agenda, procedural operationalization.
Speaker 2: The integration into government.
Speaker 1: For a UDT to be a planning tool, it has to be embedded directly into existing municipal governance frameworks. Right now, the review shows UDTs are only contributing to highly isolated practices. They do a bit of environmental sensing, a bit of virtual experimenting for specific engineering firms, but they are completely failing to impact long-term visioning, zoning policy formulation, or actual government workflows. The review argues that if they aren’t procedurally operationalized, if they aren’t a legal part of the planning procedure, then they aren’t planning tools at all. They are just video games for urban engineers.
Speaker 2: I understand the frustration there, but the text details severe, highly tangible roadblocks to exactly that kind of deep procedural integration. It isn’t just bureaucratic stubbornness keeping these models out of city hall. We are talking about immense financial cost to maintain the servers, the massive environmental footprint of the heavy computation required to run real-time city data, and perhaps most critically, severe cybersecurity risks.
Speaker 1: Cybersecurity is definitely a barrier. But surely if banks can secure their data, city planning departments can secure a digital twin.
Speaker 2: It is fundamentally different. A digital twin, by its very nature, requires pulling in massive amounts of interconnected infrastructure data: power grids, water supply, traffic control systems. If you deeply integrate that into a city’s operational workflow, it becomes a central point of failure. The review specifically cites the DUET Framework’s T-cell model as a rare example of actually trying to address the security architecture needed for UDTs.
Speaker 1: I am familiar with the DUET project, the Digital Urban European Twins, but what exactly is a T-cell model in this context? That sounds like biology, not data architecture.
Speaker 2: It is a biological analogy, yeah, and a very accurate one. In the human immune system, a T-cell can identify and isolate foreign threats. In the DUET framework, a T-cell data architecture is used to encapsulate and protect highly sensitive government data from the broader open environment of the digital twin. So it allows the model to query the data and get an answer without ever actually copying, moving, or exposing the raw underlying data to the wider network. It prevents systemic infection or hacking. The fact that the review highlights this as a rare exception proves my point. The vast majority of UDTs do not have this kind of sophisticated T-cell architecture. Therefore, deeply integrating highly experimental, broadly connected digital twins into opaque, complex government processes right now is technically dangerous. The technology readiness just isn’t there yet for full procedural operationalization. We have to secure the foundation before we build the house.
Speaker 1: I agree that technology readiness is a serious challenge, and the security concerns you raise with the T-cell example are entirely valid. But we cannot let the difficulty of the task serve as an excuse to abandon the goal. Without addressing the complexities of human context and government’s procedures, without fully embracing augmented urban planning, these digital replicas are going to fail to deliver any real societal value. They will remain incredibly expensive dashboards that exist only in academic papers and corporate pitch decks. UDTs are at a crossroads right now. They have to evolve from technical toys into socially integrated tools, even if that integration is messy and computationally frustrating.
Speaker 2: And I maintain that we should celebrate and systematically refine the concrete technical achievements of UDTs first. Interdisciplinary integration, just getting the disparate physical systems of a city to mathematically align in a single simulation is incredibly difficult, and we are making real, vital progress there. Burdening an emerging fragile technology with the impossible task of fixing human nature, resolving historical political conflicts, and curing bureaucratic dysfunction is a guaranteed recipe for failure. We need to let the tools mature before we demand they solve society’s deepest problems.
Speaker 1: Despite our different lenses on how to move forward, we both acknowledge the stark reality exposed by this eighty-eight article review. There is a profound, undeniable gap between the utopian hype of smart cities and their current realized implementation on the ground. It forces us to seriously interrogate our technological ambitions and ask what it is we are really trying to build.
Speaker 2: Absolutely.
Speaker 1: And the source material offers even more layers for us to explore in the future, particularly regarding the need for open data standards across municipalities and the very real boundaries we need to set regarding data collection. But for now, we leave it to you, the listener, to consider. When we map our digital cities, are we just drawing cleaner, more complex blueprints, or are we finally ready to model the messy living reality of the people inside them?
[music]
What is the future for cities podcast?
Episode and transcript generated with Descript assistance (affiliate link).
What is The Future for Cities? Podcast 
Leave a comment