What does it mean to ask AI to see the city how Henri Lefebvre saw it?1 This question is either a productive experimental gesture or a category error of some consequence, and it is not obvious in advance which one it is. In attempting to answer it, the conditions under which such a simulation fails, and the nature of those failures, are analytically revealing. Such experiments with urban theory propose AI as a probe: a device whose misfires, distortions, and unexpected outputs can make legible the assumptions embedded both in the theoretical frameworks and in the models themselves.
Urban theory does not begin with description, but with a decision about what the city fundamentally is, which of its dimensions are analytically relevant, which phenomena constitute evidence, and which relations we should trace. Theoretical frameworks do not filter pre-existing data, as if the city were already there waiting to be sorted; it actively makes data legible by establishing what counts as data in the first place. Georg Simmel’s account of metropolitan life does not report a pre-existing fact called nervous stimulation.2 It selects the intensity of sensory experience as the relevant dimension of urban modernity and constructs the blasé attitude as its interpretive consequence. One cannot arrive at the blasé attitude by looking at the city without Simmel’s framework any more than one can see the production of space without Lefebvre’s distinction between conceived, perceived, and lived dimensions of spatial practice.
According to Karen Barad’s concept of intra-action, apparatuses of observation do not reveal pre-existing phenomena from the outside, but participate in their material production.3 A theoretical framework is therefore not a lens held in front of a pre-existing city, but an apparatus that participates in producing specific urban phenomena as knowable objects.
This situatedness is not an invention of late twentieth-century science studies. José Ortega y Gasset argued as early as 1914 that reality is not a spectacle to be contemplated by a detached subject but something enacted through the radical engagement of the self with its circumstance.4 The city, in this view, does not pre-exist the situated gaze that encounters it; it is co-constituted with the practices that bring it into focus.5 Donna Haraway made the political stakes of this explicit: the claim to a neutral, universal standpoint is a fantasy that conceals specific locations and interests. All knowledge is partial and situated, and acknowledging this partiality is not a retreat from rigor, but the condition for a more accountable kind of knowledge. Urban theory is distributed across human geography, sociology, urban planning, and cultural studies, with each tradition producing frameworks grounded in distinct ontological commitments. Its plurality is therefore not a sign of its immaturity, but a map of the political stakes encoded in analytical choice.
Within the current climate of increasing political and ideological polarization and extreme scientific specialization, social and urban plurality is often experienced as something disturbing rather than as a productive diversity. In this context, we need methods for considering multiple frameworks simultaneously, comparing what each makes visible and what each suppresses, without collapsing them into equivalence. The following experiments attempt to operationalize this. If theoretical frameworks enact rather than describe their objects, what does it mean to ask a computational system to perform this enactment?
Seeing the City like…
Simulations translate theoretical frameworks into forms that can be enacted. This translation is not self-evident, and the obstacles it encounters are instructive about what theoretical understanding actually consists of. Robin Evans argued that architectural drawings are never simply transparent windows onto an intended building, but active mediators that transform what passes through them.6 Every medium introduces its own logic alongside what it is asked to carry. Gilbert Simondon’s philosophy of technical objects extends this insight to the apparatus itself.7 A technical object does not merely transmit or represent a pre-existing reality; it possesses a mode d’existence enacted through a process of “concretization,” generating new realities through the operative coupling of human and material agencies. Vision-language models are therefore not a transparent medium through which urban theory is conveyed, but a technical being whose own internal logic participates in what it produces. The same argument applies to the translation of theoretical frameworks into computational models.
To operationalize Lefebvre’s concept of spatial production for a vision-language model is not to capture Lefebvre in caricaturesque form, but to produce a partial and necessarily reductive analogue through a set of detectable visual indicators that approximate the kinds of features his theoretical framework might focus on. While this loss must be acknowledged directly, it does not mean that the translation is worthless. Harun Farocki’s concept of “operative images” is useful here: images that are not produced for contemplation or representation, but to trigger operations of classification, measurement, or decision.8 The computational model therefore produces a set of analytical procedures that, when applied to urban imagery, generate outputs structured by the logic of that framework rather than by the default logic of the model’s training distribution.
Furthermore, when one translates a theoretical framework into explicit, computable operations, you are forced to articulate what the framework is actually doing: which visual features are measured, how competing indicators are weighed, what counts as a strong signal. This articulation is not available in the same way when the framework is developed and applied through the tacit, accumulated competence of scholarly reading. The loss of theoretical depth is real, but so is the potential gain in analytical accountability.
There is, however, a deeper risk. When a vision-language model is prompted to analyze a city through the lens of Lefebvre, there is no guarantee that it is performing Lefebvre’s analytical operations, and not simply matching patterns to the rhetorical surface of Lefebvre in its training data. The theoretical framework would then not redirect the model’s interpretive gaze, but merely dress it up, cover the model’s default cultural assumptions in the vocabulary of critical urban theory and lend them a misplaced authority. Furthermore, Lefebvre’s spatial triad has been substantially revised by Edward Soja’s thirdspace, challenged by feminist geographers, and received differently across linguistic spheres.9 For this reason, it is necessary to read any results forensically, especially moments of failure and implausible coherence, as sites where underlying assumptions become visible.
Over the past three years, I have undertaken a series of experiments that have tried to understand what AI systems actually know about cities. CLIP and the City (2023) fed 360-degree urban panoramas of Rome into large vision-language models, extracting embeddings to map clusters and discontinuities in how the model encoded the city’s visual culture.10 WorldGist (2023–24) generated synthetic imagery for every world capital from an identical noise seed—using the same initial random input (like an image of Gaussian noise) so that differences in output reflect the knowledge encoded by the model for each city, rather than variation in the prompt—which revealed that these systems encode dramatically unequal knowledge across geographies.11 In City in a Bottle (2024–25), we operationalized urban theory directly, extracting key concepts from Henri Lefebvre, Jane Jacobs, Walter Benjamin, Georg Simmel, and more into perceivable visual features, translating, for instance, Lefebvre’s triad of conceived space (planned, regularized, abstract), perceived space (paths, traces of everyday practice), and lived space (symbolic and informal appropriation) into visual indicators, which then could function as measurable proxies for testing theoretical frameworks against urban imagery.12
The experimental pipeline for City in a Bottle combined three vision-language model architectures. OpenCLIP was used to extract latent semantic alignments between urban imagery and operationalized theoretical concepts, providing embedding-based measurements of how strongly each image corresponded to each framework’s core vocabulary.13 GPT-4o performed multi-step interpretative reasoning over the same image corpus through structured theoretical prompts, generating detailed readings organized by framework-specific indicators.14 LLaVA served as an open-source baseline to assess the degree to which interpretative variation depended on model architecture rather than the structure of the prompting protocol.15 For each theoretical framework, key concepts were translated into measurable visual indicators and organized into concept bottlenecks composed of explicit reasoning steps, visual cues, and evaluative questions.
What all vision-language models encode is far from neutral. It is a specific cultural view of cities, shaped by the distribution of data that built them, which is predominantly from and of the Global North, and weighted toward cities already legible within the aesthetic conventions of platform capitalism. They see the city, but in a particular way, and the assumptions embedded in that way of seeing are rarely declared. These experiments take this diagnosis as their starting point and ask what might be possible to do about it, not by seeking to contain the biases of AI systems, but by deliberately multiplying the interpretive frameworks operating within them.
What Is Seen
Taking the experiment of City in a Bottle one step further, an image database was constructed through a stratified comparative sampling strategy across three cities: Rome, Madrid, and Montréal. Rather than selecting sites for visual diversity, the database was organized around five functionally equivalent urban tissue types present in each city: the monumental civic node, the dense mixed-use street, the modernist or planned residential fabric, the infrastructural hinge zone, and the urban edge condition. For each location, three types of images were collected: a ground-level view, a standard satellite image, and an atmospheric variant.16 This produced a tight collection of forty-five images, allowing interpretation to be compared across scale, modality, and urban condition. In Rome, images were collected of Piazza del Popolo, Via dei Giubbonari, the Laurentino 38 estate, the Ostiense rail hinge, and the Tiber embankment. In Madrid: Sol, Fuencarral, Moratalaz, Chamartín, and the Madrid Río waterfront. In Montréal: Place d’Armes, the Plateau, Jeanne-Mance, the Turcot interchange, and the Lachine Canal.
Each image was then processed through a multi-layer pipeline composed of eight theory-specific modules condensing operationalized versions of canonical urban theoretical frameworks: spatial production (Lefebvre), historical layering (Benjamin), perception analysis (Simmel), political economy (Harvey), assemblage logic (DeLanda), Latourian actor-network reading (Latour), genericity (Koolhaas), and atmosphere (Sloterdijk). Each module was built around four to eight measurable visual indicators that the theory would treat as analytically significant. The Simmelian module, for instance, scored each image for sensory density, the legibility of individual faces in crowd situations, evidence of anonymous or purposeful movement, and markers of overstimulation such as advertising saturation or acoustic clutter. The Lefebvrian module, conversely, scored the presence of markers for “conceived space” (such as signage, planned infrastructure, and signs of abstract geometric patterns), “perceived space” (traces of everyday spatial practice like paths worn by pedestrians and habitual routes), and “lived space” (such as signs of symbolic appropriation and of imaginative or affective use). Outputs were scored against three metrics: presence (whether a theoretical indicator was visible or not), evidence (the density and specificity of supporting visual features), and confidence (the model’s own stated certainty about its reading). This allowed comparison across frameworks, regardless of how many and diffuse or few and concentrated their outputs are. Contradictions between them were treated not as failure but as relevant logical outcomes, and cases where frameworks produced weak or inconclusive results were retained as part of the analytical record.
To map the strength of each theoretical framework for each city, we positioned the eight frameworks in a two-dimensional parameter space. The horizontal axis—the analytical scale—runs from micro-scale bodily experience (Simmelian perception) to macro-scale systemic analysis (Harveyan political economy). The vertical axis—the ontological center of gravity—runs from human-centered phenomena (Jacobs’s street life) to distributed/nonhuman agencies (Latour’s actor-networks, DeLanda’s assemblages). For each city, we plotted the strength of the theory’s activation as surface height. The resulting epistemic landscapes suggest that the model uses a broadly similar interpretive framework for all three cities, but emphasizes different aspects in each city. Rome, Madrid, and Montréal are therefore not three entirely separate urban epistemologies, but rather appear within the same theoretical space, with the specific relevance of different theories shifting from one city to another. Rome shows high activation for Benjaminian historical layering and Lefebvrian spatial production, while Montréal peaks toward infrastructural and functional-zoning readings.
To map how strongly each theoretical framework activated for each city, we positioned the eight applied frameworks in a two-dimensional parameter space. The horizontal axis, analytical scale, runs from micro-scale bodily experience, anchored by Simmelian perception, to macro-scale systemic analysis, anchored by Harveyan political economy. The vertical axis, ontological center of gravity, runs from human-centered street life and lived appropriation, anchored by Simmel and Lefebvre, to distributed or nonhuman agencies, anchored by Latour’s actor-networks and DeLanda’s assemblages. For each city, we plotted the strength of theory activation as surface height. The resulting epistemic landscapes suggest that the model uses a broadly similar interpretive framework for all three cities, but emphasizes different aspects in each city. Rome, Madrid, and Montréal are therefore not three entirely separate urban epistemologies, but rather appear within the same theoretical space, with the specific relevance of different theories shifting from one city to another. Rome shows high activation for Benjaminian historical layering and Lefebvrian spatial production, Madrid shifts toward Simmelian intensity at the civic node and Harveyan/Latourian readings at the infrastructural hinge, while Montréal peaks toward infrastructural legibility, functional zoning, and atmospheric edge conditions.
The results produced several patterns of analytical significance. At the level of individual theories, the dependence on scale was the most consistent finding. Frameworks that focus on perception, like those of Simmel and Sloterdijk, produced their richest outputs at the pedestrian scale of street-level imagery, where body, atmosphere, and sensory texture are available for interpretation. Political-economic frameworks, like Harvey and Latour, generated their highest confidence readings at the district and metropolitan scale, where systemic logics of investment, circulation, and infrastructure can be inferred from satellite imagery. Failures were also informative, with Benjaminian historical analysis being weak in aerial views, where temporal layering is largely invisible, and Simmelian perception collapsed in infrastructure sites like transport interchanges, where the experiential register of the pedestrian encounter it requires is weaker. Visual medium functions as an analytical filter, and the camera’s position is already a theoretical choice.
The experiment did not produce radically distinct interpretive profiles for each urban context. On the contrary, the raw activation landscapes show a notable degree of similarity across the three cities, suggesting that the model applies a relatively stable interpretive structure under identical processing conditions. This shared structure is significant, because despite the historical, morphological, and cultural differences between Rome, Madrid, and Montréal, the model tends to organize urban scenes through a recurrent set of theoretical attractors. The delta-from-baseline visualization therefore provides a more precise reading of the results. Rather than demonstrating strong city-specific profiles, it reveals local modulations within a common interpretive landscape.
This implies that urban ontology, as enacted through these interpretive frameworks, is organized around urban tissue conditions, like modernist residential estates, or dense mixed-use streets, or infrastructural areas, rather than geographic specificity. It suggests that computational interpretation in this case is not reproducing cultural specific imaginaries but seems to track formal properties of urban morphology, like density or mix of functions, that appear in similar tissues across different cities. Rome appears differentiated by the persistence of theoretical activation in urban tissues that are historically layered, with Benjaminian and Lefebvrian frameworks remaining prominent. In Madrid, deviations are more visible around infrastructural and political-economic readings, where Harvey-inspired political economy, supported by Latourian and DeLandian readings of infrastructure and assemblage, becomes more adequate. Montréal shows a comparatively clearer activation around infrastructural legibility and functional zoning, with deviations that point toward a more regularized relation between residential, street, and infrastructural tissues. These differences are moderate rather than categorical, and they do not indicate three separate theoretical worlds, but three inflections of a shared model-level organization of urban knowledge.
Arguably most significant was the finding that theory activation correlated more strongly with different types of urban tissues than with the identity of the city. The modernist residential fabric of Laurentino 38 in Rome and the Jeanne-Mance sector in Montréal, for instance, had activation profiles more similar to each other than either did to the monumental civic node or the street fabric of their own city. The Ostiense rail hinge in Rome and the Turcot interchange in Montréal also activate similar theoretical frameworks despite their different geographic and cultural contexts. This implies that urban ontology, as enacted through these interpretive frameworks, is organized around tissue conditions rather than geographic specificity. It suggests that computational interpretation is not merely reproducing culturally specific imaginaries, but is tracking structural properties of urban form that cut across cultural particularity.
A fourth pattern concerned the consistent risk of narrative stabilization. Each theoretical model tended to produce coherent, internally consistent readings, even when the visual evidence was partial. This is a known tendency of large language models, trained toward fluent and confident narration. Roland Barthes’s distinction between the studium and the punctum offers a vocabulary for naming the cost of this: the model systematically maximizes studium—that is, the learnable, codifiable, culturally transmitted register of an image—while the punctum—whatever exceeds codification and resists any structured reading—remains structurally inaccessible.17 This must be explicitly addressed as a limit of the method, not worked around.
Inganno and Disinganno
Inganno (deception, illusion) and disinganno (the unmasking of illusion) are terms from the baroque vocabulary of appearance and its undoing, which are relevant here to describe what the experiments reveal about the default interpretative behavior of these AI models.18 Vision-language models can be made to produce differentiated readings of urban imagery when given a structured theoretical framework. The default behavior of these systems is to produce plausible, moderately informative, culturally averaged descriptions. When you ask a model to describe a street in Madrid it will give you something legible and generic, organized around the most statistically central features of its training distribution. If you ask it to describe the same street through the lens of Henri Lefebvre, you get a different output. Different lenses change what is noticed, foregrounded, and treated as analytically relevant. The difference is therefore one of interpretation, rather than merely being stylistic.
This suggests that the default interpretive behavior of AI systems itself follows a theoretical framework. It consists of a particular way of selecting and emphasizing features that have been trained to look like common sense by the few actors, mostly big US tech companies like Meta, Google, Open AI, Anthropic, and others, who get to define this in a highly concentrated global AI economy. The model’s default output does not describe the city as it is; it enacts a particular norm through the statistical regularities of its training. To unmask this through disinganno requires not just another description, but a sort of “poetic reason” (razón poética), which María Zambrano defined as a mode of knowing that does not mirror reality, but participates in its revelation, awakening the contingency of any single reading.19
Just as Evans argued about the non-transparency of architectural representation, the vision-language models used here do not work from cities but from representations of cities, and those representations are never independent from the cultural logic that organized the data from which the model was trained.20 Operationalizing alternative theoretical frameworks explicitly makes that logic visible and available for critical scrutiny. AI’s default output presents itself as natural description, but the forced multiplication of frameworks seeks to reveal the construction behind that appearance. This act of creating a compelling interpretive illusion and its subsequent unraveling allows us to see the illusion and the mechanism simultaneously.
Urban theory is disciplinarily balkanized; frameworks developed in different intellectual traditions rarely confront each other’s blind spots directly. A computational approach that operationalizes multiple frameworks and applies them to the same database creates a structured space for cross-theoretical comparison that does not currently exist. The disagreements it surfaces are not noise: they are maps of what different intellectual traditions have decided matters about cities, and making those maps explicit and comparable is a form of critical work in its own right.
But theoretical modules are always reductions. They extract measurable indicators from frameworks that are not, at their foundations, built from indicators. The models can tell something about what Simmel would attend to, but they cannot replicate the movement of his thought, the dialectical relationship between concepts, or the historical argument that gives his framework a distinctive and relevant critical force. Within these limits, what this work does is make interpretive criteria explicit and traceable. When analytical moves become visible, they can be questioned, contested, and revised. Explicitness is not the same as depth, but it is a condition for accountability—which is not trivial in a field increasingly shaped by systems whose analytical assumptions remain opaque.
Urban Imagination and the Politics of Seeing
One of the recurring arguments in critical digital urbanism is that AI systems have foreclosed urban imagination.21 By learning from the existing distribution of urban imagery, weighted toward particular cities, aesthetics, and modes of representation, these systems reproduce a closed world of possibilities. It is a combinatorial logic that recycles existing forms. The problem, however, is not only that the systems are biased, but rather that their biases present themselves as common sense, and their closures as the natural limits of the possible. Simondon warned that technical objects tend toward a “closed world” of operational optimization, losing the openness of the pre-individual milieu from which they emerged.22 The combinatorial logic of generative models is precisely this: a recycling of existing forms that forgets the transductive potential of technical invention. Against this, Zambrano’s poetic reason insists that knowledge must remain an awakening, a despertar, rather than a classification, and that the city’s resistance to capture is the very condition for its continued imaginative life.23
By processing the same images through eight theoretical lenses simultaneously, AI can generate productive frictions in places where frameworks disagree: where what one tradition foregrounds another suppresses; where the image looks fundamentally different under different analytical criteria. These frictions do not guarantee novel insights, but they make visible the contingency of any single reading. To demonstrate that urban images do not yield fixed meanings—that interpretation varies systematically with analytical framework, spatial scale, and image modality—is to reopen the question of what cities are. This is a modest computational enactment of what Nelson Goodman called “worldmaking,” namely, recognizing that different symbolic systems actively construct different versions of the world, and that no single version exhausts what is there.24
AI should not be used to imagine urban futures. But what it can do is something more modest, and perhaps more productive: use its own limitations as analytical material, and to let the gaps, failures, and implausible coherences in the outputs speak to something about the frameworks being simulated, the models doing the simulating, and the cultural assumptions circulating through them. Whether that justifies the epistemological risks involved is a question these experiments raise but cannot answer. Yet in conditions of genuine uncertainty, the precondition for new urban imaginaries is not a more powerful algorithm, but a more honest and uncomfortable account of the frameworks through which we see the city, including the systems we are increasingly asking to see it for us.
The primary works on which this essay draws are Henri Lefebvre, The Production of Space, trans. Donald Nicholson-Smith (Oxford: Blackwell, 1991) and Writings on Cities, trans. and ed. Eleonore Kofman and Elizabeth Lebas (Oxford: Blackwell, 1996).
Georg Simmel, “The Metropolis and Mental Life,” in The Sociology of Georg Simmel, trans. and ed. Kurt H. Wolff (New York: Free Press, 1950), 409–24.
Karen Barad, Meeting the Universe Halfway: Quantum Physics and the Entanglement of Matter and Meaning (Durham, NC: Duke University Press, 2007).
“Yo soy yo y mi circunstancia.” José Ortega y Gasset, Meditations on Quixote, trans. Evelyn Rugg and Diego Marín (New York: Norton, 1961), originally published as Meditaciones del Quijote (Madrid: Residencia de Estudiantes, 1914).
Georges Canguilhem provides a similar account of how the pathological, like the urban, is not discovered but rather produced through the very procedures that claim to observe it. The Normal and the Pathological, trans. Carolyn R. Fawcett (New York: Zone Books, 1991), originally published as Le normal et le pathologique (Paris: Presses Universitaires de France, 1943).
Robin Evans, “Translations from Drawing to Building,” AA Files 12 (1986): 3–18, reprinted in Translations from Drawing to Building and Other Essays (London: Architectural Association, 1997).
Gilbert Simondon, On the Mode of Existence of Technical Objects, trans. Cécile Malaspina and John Rogove (Minneapolis: Univocal, 2017), originally published as Du mode d’existence des objets techniques (Paris: Aubier, 1958).
Harun Farocki, “Phantom Images,” Public 29 (2004): 12–22. Farocki coined the term operationale Bilder to describe images produced not for human contemplation but to serve as input for automated systems of recognition, classification, and decision-making.
Edward W. Soja, Thirdspace: Journeys to Los Angeles and Other Real-and-Imagined Places (Oxford: Blackwell, 1996). For the feminist geographic critique, see Doreen Massey, Space, Place and Gender (Cambridge: Polity, 1994) and For Space (London: Sage, 2005).
An embedding is a high-dimensional numerical vector that compresses an image or text into a coordinate within a latent semantic space, where proximity indicates statistical rather than ontological similarity. Darío Negueruela del Castillo and Iacopo Neri, “CLIP and the City: Addressing the Artificial Encoding of Cities in Multimodal Foundation Deep Learning Models,” in On Architecture (2023): Conference Proceedings (Strand, 2023), 100–109.
Darío Negueruela del Castillo et al., “World Gist: Implicit Urban Imaginaries in Foundation Models,” in Architecture in the AI Era for Research, Practice, and Pedagogy, ed. A. Agkathidis et al. (Singapore: Springer Nature Singapore, 2026), 63–72.
Darío Negueruela del Castillo, Iacopo Neri, and Andrea Alfarano, “City in a Bottle” (unpublished manuscript, 2024–25). Jane Jacobs, The Death and Life of Great American Cities (New York: Random House, 1961). Walter Benjamin, The Arcades Project, trans. Howard Eiland and Kevin McLaughlin (Cambridge, MA: Belknap Press, 1999); “On Some Motifs in Baudelaire,” in Illuminations, ed. Hannah Arendt, trans. Harry Zohn (New York: Schocken, 1969), 155–200.
Mehdi Cherti et al., “Reproducible Scaling Laws for Contrastive Language-Image Learning,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (Los Alamitos, CA: IEEE Computer Society, 2023), 2818–29.
OpenAI, GPT-4o System Card (San Francisco: OpenAI, 2024), ➝.
Haotian Liu et al., “Visual Instruction Tuning,” in Advances in Neural Information Processing Systems, vol. 36 (Red Hook, NY: Curran Associates, 2023), 34892–916.
The choice of scale is never epistemologically innocent. Franco Farinelli has argued that “cartographic reason” (ragione cartografica) does not reflect territory passively but actively constitutes the spatial ontology it claims merely to depict. The satellite view, in this sense, enacts a different city than the street-level gaze, not because one is closer to reality, but because each mobilizes a distinct cartographic regime with its own center of gravity. Franco Farinelli, La crisi della ragione cartografica (Turin: Einaudi, 2009). Laura Kurgan also argues in Close Up at a Distance: Mapping, Technology, and Politics (New York: Zone, 2013), that ssatellite imagery and spatial data are not transparent windows onto territory but are enacted through specific chains of technical, military, and political practice.
Roland Barthes, Camera Lucida: Reflections on Photography, trans. Richard Howard (New York: Hill and Wang, 1981). Barthes’s studium names the culturally codified and learnable dimension of a photograph; the punctum names whatever exceeds this codification and strikes the viewer with an idiosyncratic force that cannot be transferred or taught.
Andrew Horn, “Andrea Pozzo and the Jesuit ‘Theatres’ of the Seventeenth Century,” Journal of Jesuit Studies 6 (2019), 213–48.
María Zambrano, Claros del bosque (Madrid: Siruela, 1977).
Evans, “Translations from Drawing to Building.”
Shannon Mattern, “A City Is Not a Computer,” Places Journal (February 2017), ➝; Rob Kitchin, “The Real-Time City? Big Data and Smart Urbanism,” GeoJournal 79, no. 1 (2014): 1–14.
Simondon, On the Mode of Existence, 59–61.
Zambrano, Claros del bosque, 89–92.
Nelson Goodman, Ways of Worldmaking (Indianapolis: Hackett, 1978).
