Digital Twins and Analog Kin

As an anthropologist doing fieldwork, I learned early that architecture and design are never just about buildings or things; they are also about the sites that shelter their shadows, about the wind that becomes a draft, about the things whose presence calls forth action: a stone, a gadget, a doorway. A building is not only a space of storytelling, acting, performance, living, dwelling, and livelihood as such: it is a monument of human-environment relationships. A thing can be smart or not, but it always engenders agency and co-creates new givens and phenomena.

Many architects and designers today develop eco-social concepts and solutions. They search for organic designs that work for living beings, analog and digital alike. They tinker with simulations and play with what LLMs can vibe-code for them. They also, increasingly, work with digital twins: continuously updated computational counterparts of buildings, districts, and grids, fed by sensors and tasked with predicting what their physical asset will do next. But this is not enough to balance the irreducible illusory and deceitful core of simulations, which can only approximate “the world … that is the case.”1 

The urgency of qualitative methods within architecture and design only intensifies as digital twins evolve into what the industry now calls “living” or “persistent worlds.” In these AI-native simulation environments, from Google DeepMind’s Genie 3 to NVIDIA’s Cosmos, autonomous agents populate, learn from, and reshape virtual spaces in real time, blurring the boundary between model and world. The old metaverse hype has given way to something technically more consequential: world models that generate environments on the fly, agent-driven architectures that never go offline, and generative AI embedded directly into Building Information Modeling (BIM) and urban planning workflows. But no matter how persistent and self-modifying these virtual worlds become, they remain inhabited by organisms made of carbon and water, living on a planet governed by soil, weather, and kin. 

Grounding predictive methods with substantial and existential data is not a retreat from computational power, but the condition of its responsible use. “Plans are worthless, planning is everything.”2  The faster and more autonomous these living worlds run, the more we need the slower, non-artificial intelligence that understands why and how certain patterns stabilize among actual bodies in actual places, and what the costs of that stabilization are. These are costs that no world model, however generative, will spontaneously account for. Ethnography is what does. Not as a counter-technology, but as the analog kin of the digital twin: a slower practice of being there, asking, listening, feeling, and asking again—what digital twins, by definition, cannot do.

What Digital Twins Actually Do

In 2014, the government of Singapore launched a project called Virtual Singapore. The National Research Foundation committed S$73 million and Dassault Systèmes deployed its 3DEXPERIENCity platform. Bernard Charlès, the firm’s CEO, told the press that cities are “some of the most complex ‘products’ created by humanity.”3 Products: this single word does the work. The project succeeded as a platform: Virtual Singapore was soon hailed as the world’s first digital twin of a country.4 Press releases describe what the twin does: optimize logistics, simulate emergencies, plan urban density. What the twin does not do is also on public record: for security reasons, Virtual Singapore is not connected to the open internet, and citizens cannot interact with the model that represents them.5 The twin therefore runs without those it claims to model. Cities, however, are not products of design: anthropology has long described them as sites of social experimentation, political self-invention, and ways of living together that exceed any builder's intention. To call a city a product is to confuse the polity with the platform that maps it.6 

This case is paradigmatic but not, as Dassault discovered, universally exportable. In the same year, the firm secured a parallel contract with Rennes Métropole to develop Virtual Rennes—the European pilot meant to translate the Singapore approach into a different political economy. Within a few years, the project stalled. An ethnographic study of the case found that the project repeatedly ran into the obduracy of municipal practice: its definitions had to be stretched to enroll local partners and bring the project into line with the daily business of European administration. By the end, even the basic question of whether Virtual Rennes constituted a digital twin remained contested: the Métropole maintained it did not, while Dassault insisted it did. This mode of governance is what Oskar Steiner has called “sandbox urbanism”: cities re-enacted as perpetual sites of technological renewal, never concluded.7 Where Singapore had absorbed the model on its own terms, Rennes registered the friction. The same software, in two political settings, produced two different objects. Nottingham, in the UK, working with Esri's ArcGIS Urban rather than Dassault, demonstrates a third variation: the twin as investment instrument, having secured the city council £4 billion in investment and redevelopment bonds.8 

What ethnography reads in these three cases is not a failure of representation. Rather, they are three different relations between representation and power. Singapore’s twin governs by withholding access, Rennes’s by proving incompatible with the institution it was meant to serve, and Nottingham’s by attracting capital. None of this can be explained by improving the model. It can only be described by being there, among the planners, the data engineers, the citizens whose absence is the model’s silent precondition. Ethnography is the analog kin of the digital twin. It does not aspire to model the city. Instead, it asks what the model has been told to count, who decided, and what the cost is of what gets counted as background noise: the ecological, the social, the creaturely.

The question that follows is epistemic. If the twin’s force lies not in its accuracy but in what it forecloses, then what kind of knowledge does it produce? What disposition is required to live with it?

Operational Realism: How Twins Deceive Precisely

The double-slit experiment is the simplest test in quantum optics and the most stubborn. Run a single photon at a screen with two narrow openings, and a pattern of interference builds up over time, photon by photon, as if each had passed through both slits at once. Try to observe which slit the photon takes, and the interference vanishes. When Martin Warnke and I did fieldwork for this, what we saw was not a contest between the real and the virtual, but something stranger: the simulation could be far more precise than the experiment, and that very precision was where the trouble began.9 The computational model resolved what the apparatus could only approximate, and in resolving it, generated discrete monads of matter where matter itself refused to be discrete. We called the disposition needed to inhabit this gap “operational realism”: neither the naive acceptance of simulation as a transparent window, nor the skepticism that refuses to engage with what simulations actually do.10 Precision and deception lie unusually close in computational models.

At the scale of a single building, the same pattern recurs. Commercial real estate firms now use industry platforms, including Bentley's iTwin, to model retrofit options across entire building portfolios. Vendor materials and consultancy reports describe substantial reductions in design-and-planning cycle time and double-digit improvements in identified energy savings over conventional analysis.11 The numbers are persuasive. But what the dashboard does not show is what the model has been told to count and what it has been permitted to ignore. A systematic review of building digital twin deployments in operations and maintenance finds that real-time occupant feedback is a consistent absence: the dominant applications are component monitoring, anomaly detection, and predictive maintenance.12 What the people in the building might say or feel remains, in most of these systems, beside the point. Where the double-slit produced discrete monads of matter, the building twin produces discrete monads of inhabitation: modeled occupants whose comfort is the function the system can compute, before any actual person has been asked what comfort means. This deception is not malicious; it is built in. The twin works because it forecloses what would unsettle its forecasts.

A class of media forms can be called epistemic media: media geared not toward communication but toward turning data into knowledge. They depend on a specifically digital premise: that all phenomena are equally information. This flattening opens new cross-sectional perspectives while structurally leaving outside the account whatever does not reduce to a signal, among them the reflexivity that would make these descriptions accountable. The paradigmatic precedent of this is double-entry bookkeeping: a medium that offered a precise account of the state of affairs while erasing the knowing of self-knowing.13 Digital twins operate by an analogous logic, with greater reach and far less transparency. Inside the design disciplines, the fantasy of a total computational model—one that would hold all geometric, material, thermodynamic, and economic data for a building and its context—has haunted Building Information Modeling since the 1960s. Yet the fantasy has never been separable from its nightmare.14 

Federico Campagna gives this configuration a cosmological name. In his account, the present is governed by a metaphysical architecture he calls “Technic,” whose first axiom is that everything that legitimately exists can be captured, measured, and rendered productive through descriptive language and its computational extensions.15 Technic does not merely describe the world; it enframes it. A forest becomes a stockpile of timber, a person a stockpile of labor, a building a parametric score optimized for carbon and occupancy. The shift Campagna advocates is not a retreat from computation but a modification of what he calls “reality-settings.” He names this composure “Magic”: not mysticism, but an axiomatics whose first principle is the ineffable, what cannot be fully described and refuses total productivization. Where Warnke and I argued for operational realism from inside the quantum laboratory, Campagna formulates the same exit cosmologically, from outside. The two vocabularies converge on a single requirement: to inhabit the digital without surrendering the real to its grammar. 

On Being Networked: Animism Without Animacy

What this disposition looks like in the daily condition of bodies wired into networks begins to show in the industry language used to describe them. Industry markets digital twins in the vocabulary of life. Bentley Systems calls its iTwin Platform “the lifeblood and [the] central nervous system” of infrastructure.16 Autodesk Tandem promises “descriptive twins” that “auto-tune” building systems through AI.17 NVIDIA Omniverse, built on the Universal Scene Description standard, lets a Siemens factory model, an Autodesk building design, and an Ansys simulation coexist in a single, photorealistic virtual world rendered through what NVIDIA calls “physically accurate, real-time ray tracing.”18 WillowTwin, the Sydney-headquartered AI-driven platform whose real-estate clientele includes Brookfield Properties and Microsoft, markets a continuous loop of operational optimization, sustainability management, and real-time fault detection across whole portfolios.19 This language performs animacy; the loop processes what can be measured. The digital twin's environment is neurotonic: not as the twin registers it, but as it acts on the bodies wired into it, modulating mood, attention, and property through the rhythms of its own clocking, without any single moment at which it could be said to act. This is where the medium tips into illusion. The unmeasurable is treated as not-yet-measured, the living as resource, the relational as extractable. From data harvesting through colonial dispossession to the necro-economies of contemporary capitalism, the depths of illusion run on a single principle: extraction indifferent to its object; of matter, labor, value, and life alike.

Architects and designers deploy this condition when they install an “occupant behavior” model: a temporal architecture in which the building’s response time is no longer governed by the social rhythms of its inhabitants, but by the polling rates of the sensors that observe them. Sonia Fizek, writing about automated play systems, calls this configuration “playing at a distance”: the system acts; the human watches.20 Transferred to building management, the concept holds—and reveals what it costs. In video games, the player chooses to step back. Here, no such choice is made. The system plays the building, and the occupant is moved like a game piece, with the tie between dwelling and decision severed.

Physics gave me an earlier form of the same condition. Over more than a decade, I have done ethnographic fieldwork at the Supercomputing Centre of Forschungszentrum Jülich, among the informatic physicists who write the simulations; at the quantum optics group at the University of Vienna; and at the ATLAS Experiment of the Large Hadron Collider at CERN, where I have watched machine learning come to drive the practice of discovery itself. At CERN, Monte Carlo simulations distinguish particle signal from spectral noise, and increasingly, machine-learning trigger systems decide, in microseconds and before any human observation, which collision events count as worth keeping. This is the data regime I have described as the “Big Data Game.”21 The physicist here, like the occupant of a building managed by a digital twin, does not choose the moment of step-back; the apparatus chooses it. Across all three sites, physicists described their instruments as alive: detectors “see,” particles “behave,” calibration routines “listen.” I call this “ontological opportunism”: animist vocabulary as a way of inhabiting a measuring infrastructure that would otherwise refuse all metaphor.22  

At CERN, this stance produced knowledge. The digital twin imports the same vocabulary into architecture, but without the epistemic problem that generated it. The logic of photography is that the image of the living certifies its having-already-passed. The building twin inverts this logic: it releases the inanimate into a simulation of presence, producing creatures that are neither dead nor alive, neither absent nor truly there. Ethnography has always had instruments for this threshold: reading formations that are hybrid or ambiguous by necessity, attunement to constellations, discovering conjunctures; holding suspension in view without resolving it.23 This is what the practice carries and passes on, not through formalization, but through being there. It only needs to describe what the model cannot see in itself. 

Regardless, the twin is called alive. It is described as learning, adapting, sensing. The language performs animacy while foreclosing its conditions: the unpredictable presence of bodies, the negotiations of threshold use, the encounter of a spider and a human causing unforeseen emotions and affects, the eco-social life of corridors that no behavioral model captures because it was never designed to. Technical innovation has long been understood as realized through the framework of magical ideas: feats once considered magical are rendered mundane by computation while displacing the genuinely animate dimensions of social existence.24 The foreclosure of these animate, relational dimensions is why the metaphor of kinship matters. In the anthropology of exchange, kinship systems regulate social time: through the obligation to give, receive, and reciprocate, they structure who owes what to whom, across which temporal horizons, and under what prohibitions.25 Digital lineage systems track the provenance of every datum across platforms and pipelines. They structure social time at the level of software, governing which information reaches whom, when, and with what valence. Infrastructure, too, carries a theory of the gift: it gives, obliges, and binds. The material and the semiotic are poles of one spectrum, not separable layers. To design a building without understanding how these lineages already organize the social time of its future inhabitants is to build for Technic’s optimizable unit, not for the beings who will actually dwell there.

Hanging Around: Extraction and the Practice of Ethnography

A digital twin can extract as readily as it can ground; the difference lies in what is placed beside it. The same predictive maintenance that reduces a building’s energy use can be calibrated to extract the maximum surplus from the bodies inside it. What is at stake is not accuracy but whose social time the twin organizes, and to whose benefit. In architectural practice, digital twins now sit at the center of BIM workflows, daylighting analysis, HVAC tuning, carbon accounting, and what the industry markets as “occupant-centric” design. Their dashboards are precise; the bodies they index are not. Without fieldwork, the twin becomes a positivist speculation in code, calibrated on assumptions about how people actually use thresholds, corridors, and shared rooms that, more often than not, prove wrong.

“Hanging around” has been theorized as the ethnographic counterpart to the dérive: an open, drifting, sensory attentiveness.26 The dérive, developed by the Situationist International, designates an unplanned, receptive traversal of urban space in which the researcher surrenders to the city’s own logic of attraction and repulsion, allowing the unofficial, the residual, and the atmospheric to become visible. Anglophone urban research has converged on the same intuition: drifting has been rearticulated as “hanging-out-knowing,” a process of “becoming with” the city through ephemeral encounters that produces no representation but reorganizes the knower.27 Architecture pedagogy has also begun to take this seriously, assigning the dérive alongside drawing and modeling as a research method for the built environment: a discipline of disorientation that works simultaneously on the senses and on inherited assumptions about how space is organized and used.28 In my own teaching, I see that students take to it readily. No dashboard can replicate such practiced exposure to the city’s unofficial rhythms. It requires being among caretakers and tenants, parasites and pets, planners and protesters, old and young, among that which is considered humane and that which must be disgracefully called inhumane. 

Thick description—Clifford Geertz’s term for accounts that record not bare acts but dense webs of meaning, intention, and context that make them legible, narrating gesture together with everything needed to read it—ties twins to the ground: janitorial rhythms, informal uses of thresholds and lobbies, the social life of corridors, conflicts around signage and access.29 These findings impose realistic limits on positivist speculation; they are what no dashboard can produce. They matter beyond the spreadsheet. Ethnography brings back into view what dashboards filter out by design, what happens upstream of the consumer’s choice: soil and water, supply chains, the energy footprint of the model itself. These concerns should no longer be reducible to profit; they are central to any planetary perspective.

Architectural ethnography is not a new import from anthropology. Modern architects from Bruno Taut to Hassan Fathy drew on extra-European building practices as epistemological resources, seeking a counter-flow of knowledge the discipline had largely suppressed.30 The knowledge those architects drew from extra-European building practices was still primarily a resource for the discipline’s own renewal—epistemological extraction, however well-intentioned. Since then, however, an “ethnographic turn in architecture” was opened by ethnographic studies of design offices, treating design as distributed across humans, models, and software,31 and a recent edited volume consolidates the field as neither anthropology of architecture nor ethnography for architects, but a crossing of the two, oriented toward lived space.32  

This turn is not only descriptive but generative. In some recent design studios I have witnessed, architects sent students out to live with a region’s building practices and bring back what standardized construction filters out: the acoustic grain of a room, the more-than-human tenants of a wall, the place-knowledge that building norms were never written to hold.33 Read this way, ethnography is not only an audit of what the twin excludes but a resource for an architecture built otherwise: less normed, more attentive to the creaturely and the sonic, exactly the background noise the model discards. The slow turn toward lived space marks the beginning of a different logic, one that must answer to those who inhabit the buildings, not only to those who design them. From the anthropology of dwelling one can argue that any model claiming to represent a building before or instead of its inhabitation enacts an epistemological inversion.34 

Designers and design students already operate as para-ethnographers: in their everyday practice they decode and recode culture in order to shape lifeworlds.35 What they are missing is not more data, but a reflexive method that names and deepens what they already do. Hence the appearance of ethnography and anthropology as method and theory in design curricula. Ethnography, in this account, is not an external audit of design practice but its analog kin: a way of seeing that designers already partially embody, and that can be made methodologically explicit.

Opening Digital Black Boxes

Ethnography makes a world of simulations observable, describable, and designable. Not animated, but alive: a world that “kicks back.”36 It narrates human-environment stories of every kind, reflecting on who acts, who is acted on, and what is at stake. It works in the gaps between categories: in thresholds, leakages, transitions, in everything that refuses to be one thing or another. It gives room for what digital twins by definition cannot count: energies and spirits, atmospheres and cosmologies, tricksters and tenants, pets and pests, place-wisdom and cosmopolitanism, popular culture and its others, creaturely citizens and those who are not, cyborgs and queers and those who refuse those categories, and the otherwise inconvenient. It acknowledges the plurality of worldviews and the planetary condition that humans are always-with others, organisms, materials, infrastructure. And ideally it does so from an ethics of collaboration rather than extraction: not studying people but working alongside them, accountable to those it studies.

Cultural systems discipline body and mind, but people slip scripts through constant transformation. In design and architecture, that means pairing parametrics with publics, coupling dashboards to diaries, measuring not just kilowatt-hours but kin-work. Media, from messengers to matter-waves, play a non-reducible role in human-environment relations. They are political instruments; they shape our affordances; they raster, filter, transmit. The task now is not to retreat from AI but to make it an ally rather than an oracle, a tool of care rather than extraction. 

Digital twins are simulacra in the strict sense: operative chains that produce—from building to city to inhabitant—the very concepts they subsequently claim to represent.37 What such chains exclude is what ethnography learns by attunement and reading traces. In the ruins and citadels of modernity, it sits with those in the corridor whose use was never modeled, the inhabitant whose presence the trigger filtered out, the threshold whose meaning was not asked. These appear beyond the edge of what the model could resolve, and most clearly at the moment the dashboard’s certainty briefly fails. Knowledge of buildings proceeds through such gaps, not despite them.

What lies outside the operative chain of digital twins is what Friedrich Kittler retrieved from Greek antiquity under the name orchēsis: music, poetry, and dance as a single embodied practice of knowledge, older than the separation of thinking from feeling.38 To practice orchēsis under digital conditions is to do what ancient Greeks might have called magic in its non-occult sense: to make perceptible what the instrument cannot show, to take into view what is, by construction, held back from sight. This work cannot be expected of every architect or designer alone. Instead, design practices and architectural offices need to bring in people trained in ethnography, from the humanities, whose discipline is precisely this attention to what the dashboard cannot count. 

As LLMs draft the briefs, generative simulation proposes the massings, and what is now called “vibe-coding” replaces the slow learning of how a system actually fails. Retaining attention to what the dashboard cannot count is a structural urgency, not nostalgia. The classical components of architectural and design training are thinning out in the everyday formation of architects and designers: drawing by hand, site visits, sustained observation, written reflections of how a building is actually used. What was once implicit in the discipline is no longer reliably transmitted or expected. And what is not transmitted is also not retained: practitioners themselves are working at increasing remove from the physical reality of what they design. The hand is no longer in the loop as it was. 

Every medium teaches a way of being. Writing taught memory to live outside the mouth. Print taught the eye to read alone. The digital twin teaches us to mistake the measurable for the real. But no medium is destiny. Each is pharmakon, poison and remedy at once. Its valence depends on who composes it, who owns it, who profits from it, and at whose expense.

Ethnography is not against new tools. It is for what new tools forget: the heat of a corridor, the politics of a threshold, the slow agreement by which strangers come to share a wall, the small garden they tend in common with its pets and its rats, the named and the unnamed, where presence is the only entitlement. Ethnography holds these in view while the model does its work so that the two can be composed into a new whole. The ethnographer in architecture as well as in design is not a romantic objector to the Technic but a practitioner of what remains when the model has done its work and the world still exceeds it.