Abstract
Currently billions of dollars are being pumped into artificial intelligence (AI). 1 This matters for urban scholars not simply because of the money being spent, but also because making profits will require having cities serve as testbeds and a reworking of core urban systems. A lucky few will strike it rich. Many will watch their money vanish, casualties of a new, fast-moving creative destruction. The rush into AI is creating worries about another tech-fuelled market collapse. 2 After the dot-com boom burst, the NASDAQ was down 75% from previous highs and $5 trillion of wealth was destroyed. This time things are moving even faster. Corporate spending on AI is estimated to reach $500 billion in 2026. 3 Such is the hype surrounding AI, CEOs are investing huge sums to avoid being left behind. Nobody wants to be the next Kodak, Blockbuster, or Sears – corporate relics in a world that moved too fast for them.
Even a casual observer can see the herd stampeding. The divergent innovation pathways of a healthy market economy are presently being replaced with highly speculative AI-based alignments. Rather than having winners and losers, we could end up with everyone losing. Such worries are now regularly soothed by anxious market commentators. 4 Reassurance is provided by the central role of large profitable companies (e.g., Google, Microsoft) in backing the AI boom and the seemingly obvious profit-driving applications for AI. 5 Concerns about monopolies and power over the collective mind will seemingly have to wait.
Here I want to focus on the question of how AI will impact the city and what constitutes urban research. When we look at AI this way, major epistemological challenges come into focus. AI brings with it new technical capacities and has the potential to fundamentally change how cities are known, governed and contested. Emergent urban scholarship is convincingly showing that applying AI means transforming the urban experience (Caprotti et al. 2024; Maalsen, 2025). If workforces are rationalized by AI, the city changes. If AI shifts how urban services are provided, our experience of the city changes. We can already see this with AI-enabled apps, websites and user interfaces. 6 So, how do we understand this shift? The first step in answering this question should begin with us asking just what AI is, and what it might become.
At this point disagreement emerges (Colther and Doussoulin, 2024; Ratiomachina, 2025). Large language models (LLMs) generate text via predictive algorithms, thereby mimicking human understanding. Applications such as ChatGPT do not understand what they are telling you. Their output is a product of algorithmic modelling that has been trained on existing textual content. It is therefore not subject to the traditional forms of epistemic testing that we associate with knowledge in the social sciences. Nevertheless, LLMs are excellent for summarizing, drafting and synthesizing. 7 If you need to quickly understand a particular legal statute or want a general overview of a complex literature, something like ChatGPT is an extremely helpful tool. But it is doubtful whether this output itself constitutes knowledge. So, we cannot expect AI to understand the city for us.
AI is a particularly advanced epistemic instrument. It does not replace human knowing but rather re-shapes how knowledge is produced and disseminated. It is a technology that has expanded our cognitive capabilities, making it akin to the book or personal computer. The philosopher Andy Clark (2003, 2024) has been a notable proponent of this view. For Clark (2025) much of the recent talk about the dangers of AI misses a crucial point. In an article written with David Chambers (Chambers and Clark, 1998), Clark previously argued that human cognition has long been ‘extended’ by human action. Progressively humans have evolved how they engage with and produce knowledge: In the cases we describe, by contrast, the relevant external features are active, playing a crucial role in the here-and-now. Because they are coupled with the human organism, they have a direct impact on the organism and on its behaviour. In these cases, the relevant parts of the world are in the loop, not dangling at the other end of a long causal chain (9).
The point being made is that these external ‘things’ are integral to knowing, and thus knowledge production itself. Annie Murphy Paul (2022) talks about these things as ‘extra-neural resources’. Without these resources it is impossible to imagine living in an urbanized world like ours: The gap between what our biological brains are capable of, and what modern life demands, is large and getting larger each day. With every experimental discovery, the divide between the scientific account of the world and our intuitive ‘folk’ understanding grows more pronounced (3).
Murphy presents civilization as a process of epistemological evolution and adaptation where extra-neural resources have been incorporated into our acts of world-making. Andy Clark (2025) encourages us to see AI as just the latest step in this process: The brain's core skill set thus includes launching actions that recruit all manner of environmental opportunity and support – from scribbling on a sticky note to firing up an AI. If the best uncertainty-minimizing suite of actions involves a bit of internal brainwork and a bit of bodily work (such as tapping some keys on a laptop) that is the sequence that gets chosen. The brain itself is unconcerned about where and how things get done (np).
Clark's epistemological position here is one that Karl Popper would have likely endorsed. Popper created ontological distinctions that Clark's theory of extended cognitions relies upon. For Popper, understanding how we produce knowledge necessitated making ontological distinctions. As a problem-solving species, Popper argued humans have created a world of objective knowledge that exists independently of their consciousness (1979; also see Bird, 1985).
Popper (1979) identified three distinct dimensions of reality. World 1 is the material world, the world of physical substances and processes. World 2 is the subjective dimension, that of our consciousness and experiences. Then there is World 3, the objective world of human knowledge. Here is one of Popper's most productive philosophical insights. World 3 can be viewed as a depository of human action, all their theories, languages, institutions and so on. Popper thought these products were ‘objective’ because they were artefacts that could be identified, examined and critiqued using scientific methods. In other words, they have an objective existence independent of human consciousness (World 2).
When Popper talked about World 3, he would often reference books. Their existence is objective and independent. For example, someone can write a book that sits on a shelf unread for decades. It exists as an object, even if no one ever picks it up. Should someone eventually do so, this World 3 object can be engaged by World 2 and, if necessary, subject to scrutiny and used to propel the further creation of World 3 objects.
Although cities are material spaces, and thus exist as World 1 entities, they are also part of Popper's World 3. Cities have definitive materiality, but this is inseparable from the accumulated bodies of knowledge – plans, rules, classifications – that organize how urban space is produced and used. Cities are World 3 objects, imagined in World 2 and manifest via a transformation of World 1. But cities are not singular objects. Rather they are aggregations of countless World 3 objects: laws, architectures, plans, institutions, cultures and so on. Many of these urban products are the consequences of World 2 action, where certain ideas (e.g., ‘smart cities’ or ‘automobility’) are crafted from the material world. The city therefore exemplifies the interrelated nature of Popper's three worlds. All intertwine to create our lived reality. This has complex historical dimensions with us all living within the physical manifestations of preceding collective intelligence. The knowledge infrastructure of our forebears is in our roads, sewers, architectures and schools.
The rapid deployment of AI creates new problems for Popper's ontology. When Popper uses books as exemplars of World 3, we are presented with a somewhat static object. Sat on the bookshelf, the order of words cannot magically transform. Put differently, any changes in the meaning of the book will come from how it is read, not the grammatical composition of the World 3 object. The urban environment can provide analogous examples. Across the USA, cities are scarred with the mid-century preference for automobility (see Caro, 1975; Lewis, 2013). In my home city of Worcester, Massachusetts, I-290 sliced the city in half. This World 3 object, a product of immense engineering skills, has institutionalized a whole set of human behaviours and perceptions. Now, in less autophile times, the road's static presence creates new urban problems, prompting the need for new World 3 outputs.
But AI is not like a book or a massive interstate. AI brings unprecedented dynamism to Popper's World 3 by continuously updating itself in response to new data, training and feedback loops. This compresses the distance between, and reorganizes the role of human intervention, between knowledge production, application and revision. The great promise (and threat) of AI is that it can engage in knowledge production in ways distinct from humans (see Maalsen, 2025). AI is becoming an active partner in the creation of World 3 itself. Some folks are already imagining a world where AI is used to autonomously develop planning code, zoning guidelines and traffic systems 8 (Salazar-Miranda and Talen, 2025). If machine learning becomes a larger part of the urban process, the resulting synthetic reasoning will require us to change how we think about agency and change.
Andy Clark (2025) argues this AI boom is best viewed as an historical progression, not some revolution in our ontological relations. We have, Clark (2025) argues, continually evolved our cognitive geographies, extending this activity into the world around us. In this view, historical forms of urbanization have deployed all sorts of cognitive devices to drive development and growth: manuals, tools, notebooks, timers, sensors. One role for future urban research will therefore be to decipher how the extended cognition enabled by AI will shift urbanization processes that already operate at a distance from World 2.
Current investments in AI suggest this is a quite uneven process. Massive sums are being shepherded into ventures that can make enterprise more efficient and profitable. The impacts of this investment will vary across sectors. In the digital world, a business model based on dopamine-driven attention capture (Auchincloss, 2025) will likely see AI used to fine-tune what we see and experience online (De et al. 2025). In the business of urban management, AI is being applied to complex, data-rich problems. For example, now one can imagine a real-time sequencing of traffic lights where AI responds to data inputs on traffic flows to optimize that evening's commute home. AI is doing a type of ‘thinking’ that can only be handled by machines, all with the aim of improving our urban experience.
A pressing problem with all these applications is that AI has a different relationship with criticism. While AI systems are designed to optimize themselves according to predefined objectives, they do not ‘think’ about the normative assumptions embedded in these objectives. Nor are they capable of weighing the moral concerns about these ends. For Popper, knowledge and politics both rely on a continual process of conjecture and refutation (2002): we propose understandings and solutions then subject these to criticism. If the criticism sticks, we come up with new solutions. As a devolved and/or semi-autonomous form of decision-making, AI potentially eludes this process. AI is a closed system of algorithmic calculation that can evolve faster than human thought. But it cannot rationally scrutinize itself. For example, it cannot deliberate whether making automobility more attractive is an ethically good thing. Nor can it assess whether the enforced alienation of car transport is preferable to more socialized means.
A significant threat accompanying AI deployment is therefore that it increasingly replaces a deliberative process of social engineering with something akin to algorithmic optimization. Here World 3, as a product of human action, appears as a hybridized epistemic object that is co-produced between humans and machines. These are entities with different political and moral status. How much this hybridized World 3 can remain transparent and open to criticism is debatable. This is particularly true in a world where AI is produced by profit-seeking big-tech companies that have little interest in the iterative and democratic modes of social change Popper endorsed.
As AI is applied across numerous parts of the urban process (see Maalsen, 2025), we are therefore shifting the qualities of what Clark and Chambers (1998) call the ‘extended mind’. If books were emblematic of this extension for Popper, the static nature of World 3 they embodied is being made significantly more dynamic – and potentially semi-autonomous – by AI. For World 3 to be transformed, we no longer need an engagement with World 2. AI can transform the contents of World 3 without human intervention. In Andy Clark's (2025) terms, the act of cognition now significantly extends beyond the human mind and body. Popper's World 3 becomes less a depository and more an active partner in knowledge production.
Understanding where AI intervenes in cities and where our critical attention needs to concentrate should therefore become research priorities. As AI distributes cognition in new and more dynamic ways, it does so via multiple channels. We are already seeing this in cities today. Sensors are feeding algorithms that are digitally tailoring parts of today's urban experience (Caprotti et al. 2024). What marks these out compared to historical forms of cognitive extension is their active participation in the urban process. Apps and platforms are actively modifying services, digital networks providing real-time feedback and problem solving and urban governance is becoming regulated and evolved in real time. The city is therefore increasingly integrated into a hybridized cognitive ecosystem.
Such a vision of AI-driven urbanization relies on concepts like prediction, optimization and augmentation (see Liu et al., 2025; Maalsen, 2025; Raymond et al. 2025). These are not neutral terms. Often, they sit upon value judgements that themselves require ongoing scrutiny. At this point it is unclear what the intervention points are for ensuring that democratic deliberations remain a part of this emergent cognitive ecosystem. As we offload the cognitive work of understanding and coordinating complex urban systems, how do we maintain the ability to ensure embedded value judgements are transparent and debatable?
In his work on digital ethics, Luciano Floridi (2014, 2023) has developed a vocabulary to help tackle some of these problems. Floridi coined the term ‘onlife’ to describe the merging of digital and physical life. We must, he argues, be attentive to the ways in which ethical and epistemic questions are embedded within our digital lives. Democratic practice can no longer be purely associated with institutions and deliberations. The extended-cognition city is an example of Floridi's onlife, a place where humans and machines interact and remake the world. Citizens must therefore navigate the city's infosphere in ways not commonly associated with democracy. This will include developing understandings of how AI systems and algorithmic reasoning regulate urban social relations in ways that make them relevant to the democratic process.
Again, this might not be totally novel. Rather it can be viewed as analogous to knowing the rules of public hearings or parliamentary protocols. Our political systems have evolved to regulate democratic processes. Indeed, we have debates about whether these systems are adequate: Do speaking limits in council sessions stymie the democratic process? Do house speakers effectively moderate cross-party deliberations? When we propose changes to these rules to pursue political ideals, we can use experience to test whether they work or not. This is Popper's process of conjecture and refutation. The challenge with AI and ‘onlife’ is not that they mediate political processes. Rather, it is that how they intervene in processes of social regulation and mediation is likely less transparent and, in some cases, locked behind corporate firewalls making them immune to deliberation.
So, it is worth stressing that similar challenges have been faced before. When we view the city in history through Andy Clark's lens of extended cognition, we see it as an output and vehicle for the cognitive development of the human species. In Ancient Athens, the Agora served to externalize civic knowledge, providing a location whereby the democratic process could be enacted without relying on certain individuals. Athens’ citizens used laws, monuments and protocols to (re)create their political life. Knowledge was literally being embedded in the urban fabric. Centuries later, our cities are saturated with equivalent cognitive reach. Our technological and infrastructure augmentation is immense: railways, telecommunications, simulations of all kinds. In many cases, we have found ways to ensure that this World 3 making has remained actively political, despite the growing complexity of its extended cognition.
The great novelty we face is that this utilization of ‘cognitive scaffolding’ (Clark, 2025) is now less depository and increasingly co-produced. AI systems operate at speeds and levels of opacity that exceed the capacity of most citizens to observe, understand and/or contest. This shift will have countless implications, both small and large. Given the prospective advances, these implications will inevitably be faced. Some are now dreaming of being able to capture and control the vast complexities of the urban process (Xu et al., 2026). Others (see Noveck and Mujica, 2023) are suggesting that AI can bring whole new capacities to urban governance, thus enabling city governments to become more responsive and effective. We will not be able to critically examine all these changes using existing approaches. Our object of analysis is becoming even more of a moving target.
The extent of this shift will likely depend on the capabilities and development of AI itself. If AI remains a largely pattern seeking technology based on LLMs, its applications will be confined to efficiency gains and automation. If the more optomistic futurist view is realized (see Tulka, 2024) and AI technologies become capable of creative thinking, we will not be dealing with the cognitive expansion of humans (see Clark, 2025), but rather the creation of new forms of cognition.
For now, we are confined to the applications of LLMs within the urban process. So, we will have ameliorative AI agendas (i.e., applications for government efficiency and democratic transparency) working alongside commercially orientated efforts. It is anyone's guess which of these applications has the most significant impact on urban life. What is clear is that the investment in AI is already shaping cities. 9 More junior and lower-salaried white-collar jobs are particularly vulnerable to AI replacement (Leopold, 2025). The current drop of graduate recruitment in white-collar industries will eventually show up in gentrified neighbourhoods across large cities as the flow of new customers and renters slows. We must also be attentive to the unintended and unanticipated consequences of the current boom in AI. For example, as big-tech companies burn through their cash reserves to fuel AI investment (Karma, 2025), how this impacts future urban development in cities such as San Francisco and Boston will likely be significant.
Despite the inevitable hype surrounding AI, we can be in little doubt that it is bringing with it significant change to the urban process. As we identify and critically examine these shifts, we must also recognize that the process of cognition is itself being altered. How we produce and store knowledge (i.e., (re)make Popper's World 3) is being transformed. If the city is a living repository of World 3 objects, AI is making it into a more dynamic and self-regulating system. Understanding all this will necessitate us to revisit epistemological questions that have been long neglected in urban scholarship. It is my great hope the Dialogues in Urban Research can provide a space for this conversation.
On the issue
In this issue of Dialogues in Urban Research, we have two forums that take up different dimensions of the field's past and present. In Justus Uitermark's (2026) forum, we go back in time to the Chicago School of Sociology. The Chicago School, Uitermark argues, has become something of a hidden presence with urban scholarship. Rather than assign the work of people like Park and Burgess to history, Uitermark argues that much is to be gained from a re-engagement with the work. In particular, he argues that the careful consideration of context and methodological pluralism practiced by the Chicago School merits a more visible place in today's urban scholarship.
Within the context of AI, Uitermark's thoughtful evaluation of the Chicago School is immensely productive. While the techno-optimists might believe in radical reinvention, Uitermark's paper reminds us that the roots of urban scholarship run very deep, and that we would be foolish to neglect this inheritance. Our commentators largely agree, although their differences signal to the difficulty involved in deciphering what parts of historical scholarship to actively maintain, and which parts can safely be assigned to historical accounts of the field.
Our second paper by Ana Santamarina and Anthony Ince (2026) asks us to reconsider the politics of one of the Chicago School's core concepts: the neighbourhood. Santamarina and Ince show how the neighbourhood is a constituent part of today's far-right politics and thus ask us to disassociate localism with progressivism. Again, some fundamental and often unspoken assumptions are being revisited. Although much is now made of the role of social media in stoking tribalistic political antagonisms, Santamarina and Ince help us see how the urban and digital intertwine, and how the local becomes pivotal to stoking certain types of political positions. Of course, how this relationship evolves in the context of AI, and the AI-enabled local urban environment, will become an increasingly relevant question for us to answer.