AR / Ralitza Petit / AI: Peter Pan’s Runaway Shadow, Digital Twins and an Intelligently Artificial Architecture of Irony

Artificial Intelligence is hot; and some architects are getting cold feet.

Those same architects would consider themselves socially engaged and culturally involved, and most definitely of their own time. While veritable contemporaneity often eludes architectural practice, and sometimes theory, the gap between cultural changes and architectural response has been attributed to the arduousness and length of the process of conception, design, and construction of buildings. The gap can further be described as the time lag between conceptual rhetoric of desired cultural effect and realized in physical materiality of architectural form; this gap — a discrepancy between the theoretical expectation and the physical reality of architecture — is perceptible but not insurmountable. Mostly, architectural form catches up with its desired rhetoric —eventually. The process of catching up is often revealed though irony.

The time lagged race from architectural theory to practice has been particularly pronounced with the ultra rapid infiltration of artificially intelligent tools in life, culture, society, environment — AI is seemingly in everything and everywhere. Such omnipresence is afforded by the rather loose definition of the term to encompass any digital operation appearing to be based in computer experience, or machine learning, which process is understood to be rooted in pattern recognition of source data; moreover, AI performance adapts and improves over time — in contrast to digital operations through previously explicit sequential computation.

The characteristically vast amount of source data and AI’s incredulous speed of data consideration have produced a new situation in relation to architecture’s notoriety of being slow to change. The unique problem for architecture is that this innovation has reached global acceptance and pervasiveness in an unusually short time and at an unusually large scale. In addition, the said innovation did find its place in buildings and theory at an unprecedented scale. Previously, it was only during major catastrophic moments in history, when destruction nearly obliterated the physicality of buildings. In such moments, the ideation of a societal change and its co-existing architecture were separated by a gap of such immense swiftness that the resulting juxtaposition of destroyed reality and dreamt up effect could be understood through irony.

Unlike in historically catastrophic situations, the AI-age is unusually productive and highly positive — yet, the gap between conceptualization and actualization, zeitgeist and buildings has appeared just as abruptly. This ensuing ironic engagement of AI with architecture is manifested in two ways: on the one hand, buildings manage to incorporate or find themselves endowed with the latest AI-ness — in the case of intelligent buildings; or on the other hand, the new technology—Artificial Intelligence—finds its way into the method of designing which is then termed generated. In both cases, an architectural theory of irony can conceptualize the connection between ultra-fast and highly flexible computation and the steadfastness of firmly grounded buildings.

As an intelligent building case of irony, one could consider The Edge Amsterdam, designed by PLP Architecture for Edge Realtors. The project began with the ambition to showcase the real estate group’s net-zero strategy, and upon its opening in 2014, the forty thousand square meter office building was, indeed, pronounced the most intelligent building in the world. Its twenty-eight thousand sensors, installed throughout the building, directly assign (artificial) intelligence by making continuous electronic measurements which in turn initiate continuous adjustments to the functionality of its mechanical systems. Such a process of observation and optimization of patterns of functionality is typical of what is considered an AI system. An example of a characteristic undergoing such AI optimization is energy consumption; and the success of reducing the amount used thanks to a continuous cycle of measurement and adjustment has contributed to the Edge Amsterdam’s rating as the world’s greenest building.^{^[1]} The apparatus of sensors, tracking and chasing ideal constellations of measured patterns, literally pins (artificial) intelligence onto the physicality of a building in much the same way that smartness is added to telephones, watches, toasters et cetera. The irony of this artificially acquired intelligence is that the endeavor sidesteps the materiality or the appearance of the Edge—its architecture—in order to continuously optimize external and internal environments.

Environmental sensors have existed for some time, and their precision and versatility has grown to measure values for: temperature, humidity, electrical conductivity, mineral composition, light, wind speed and direction, air quality, solar radiation, rainfall and underground water levels, pressure and quality amongst a plentitude of measurables. In fact, almost every aspect of the environment can be measured and monitored through sophisticated sensors which create data that rarely if ever reaches conceptual architectural design. Yet, these devices sooner or later make their presence known and very visible in buildings. Moreover, the expansiveness and speed of AI allows digitally enabled gadget creation and continuous regulation to outpace the conception of architectural form. The often-mandatory devices appear almost parasitically attached to buildings of any type, nonchalantly dismissing architectural morphology.

The irrelevant, irreverent and ironic bid of AI-ing architecture through sensors often falls flat due to the fact that most of these sensors are measuring parameters that could not meaningfully affect architectural form. Unlike a farmer who could measure the soil’s humidity and react in the direct benefit to crops, an architect is often aghast at the utility of possible measurements as none of them relate to a building’s Vitruvian obligation for firmitas, utilitas or venustas.[2] In fact, sometimes the most advanced AI-powered optimizations in buildings appear to function better without the building’s elements or the occupants’ actions. For example, forced air ventilation depends on the maintenance of a particular internal volume for the maintenance of pressure balance. An opened door — internal or external — immediately disrupts the carefully measured balance, followed by a disruption of temperature expectations, which in turn trigger the opening of more doors or windows while the entire optimization process becomes ironically superfluous.

Not surprisingly, the large number of articles on the most intelligent building in the world, the built epitome of blending the newest of technology with architecture, not once so much as mention the built project’s architecture. Without exception, each description of the Edge extolls the technologically savvy way of measuring and continuously adjusting characteristics pertaining to the micro climate or the population of the building, but very little is noted in respect to the formal order of space.^{^[3]} AI optimization was not attempted through architectural means — such as material transparency for the modulation of light, or structural porosity for the modulation of air flow, as mere examples; instead, the mechanical systems of modulation are disconnected from the body of architecture; the result: an environment in ironic parallel to the physical presence of architecture.

Notwithstanding its unremarkable architecture, The Edge has been proclaimed as nothing short of a “cultural revolution” on the basis of its unusual occupation style: each day desks and offices are redistributed. The human occupant, through their subjective presence in architecture, becomes key to an architectural conceptualization of space, the subject-object relationship that is in a permanent repositioning and tracking becomes the mode of spatial definition. In the age of digitalization, the continuously tracked relationship between subject and object is partially or wholly digitalized. Understanding the method of digitalization requires revisiting of the historical positioning of the human subject while comprehending architecture’s object.

Architecture as a spatial organization, a formal architectural order, ensuing from the directional relationship subject—object, is a theoretical stance placing the subject in the position of understanding architecture through vision. This geometrical relationship of the subject to the object is definable through optical laws of perspectival projection. The awareness of perspectival projection is often traced to the representation of architecture in painting during the Renaissance, notably described in Alberti’s treatise On Painting^{^[4]}. Even more importantly, it became possible for the reverse construction of perspectival projection to be used as a design tool — allowing the assumption of perspectival projection to effectively modulate the perception of physical three-dimensional space. For example, Borromini’s well-known visual “enlargement” of Palazzo Spada[5] is achieved by consciously using the optical illusion of perspectival projection as a design tool.[6] A gallery in the palazzo appears more than four-times longer that its actual length — a perception of a larger scale which is achieved by sloping the floor and varying the height of ceilings, column size and spacing. Spectacular as it may be, this illusion of an enlarged space only works if the viewing subject is standing within a specific optimal viewing position. Only from that position can one conjure a subjective opinion about the height and distance of the perspectivally foreshortened space based on previously encountered similar spaces and the awareness of his or her own height and distance from the implied flat composite image of a view. Thus, three-dimensional physical space is clearly dependent on the conception afforded by the two-dimensional image — through the construction of a perspective view. The human subject’s awareness of his or her own figure and proportions gives measure and reference to the spatial order – the architecture.

An extension of the conception of spatial order in digital terms became quickly necessitated by the advent and popularity of video games. The urgency of the resolving overlaps and intersections of interacting co-players necessitated an order of digital spatiality. Geometrical projection-based visual representation of digitally defined and ideated space opened the possibility to imagine interaction within a digitally invented space, as if it were an optical illusion of another world. The resounding expansion of digital worlds through online gaming came shortly after the online game World of Warcraft was introduced in 2004^{^[7]}; the population of active paying subscribers quickly reached twelve million players. What happened in the early 2000s, and overwhelmingly so in 2004 with the launch of World of Warcraft, is that online communication — for business or leisure — evolved into an idea of space due to the persistency of the gaming digital subject which continues to exist, and be seen by other logged-in players, after the subject’s human player is no longer connected to the digital world.

Even if digitally defined space were to be accepted as spatial by its theoretical definition though mathematical coordinates, the human subject could only be imagined within this Cartesian space through mathematical definition of virtuality— as a Virtuvian Man. This Virtuvian Man — not a misspelling but rather a virtual rendition of Leonardo’s Vitruvian Man, has been known by a few names already — an avatar, a char[acter], an emojicon, recently and most overwhelmingly: a skin. What makes the Virtuvian subject, and his or her corresponding digital environment equivalently anthropocentric — or avatarocentric — is that this subject is inscribed in mathematically measured terms, algorithmically defined and geometrically represented by coordinate-based terms in exactly the same way as the environment’s architecture. Gaming engines, i.e. coded computer software, often with the use of AI, in effect define Cartesian coordinates and then continuously check the veracity of this algebraic construction by converting (translating) the coordinates into a two-dimensional representation of perspectival projection which in turn is understood by the human eye as spatial.

While physical architectural order can be perceived and understood through a flat two-dimensional representation thereof, for example through perspective of a similar previously encountered projection as is the case in Palazzo Spada, the digital definition of space is already three-dimensionally conceived. In other words, digital space’s only mode of existence is in the form of mathematically formulated coordinates — that is, its conception is already three-dimensional by definition and two-dimensional projective representations on screens or VR devices are only a secondary representation, however indispensable in “accessing” the reality of digital space.

Hence, geometry, and particularly perspective, becomes the basis of the spatial comprehension which gives the possibility of conceptualization of algebraically defined objects in much the same way that the mechanism of real space perception is explained as an optical geometrical sketch of how the brain interprets the information captured by the eyes. In other words, the virtual gaming space is actually three-dimensional by definition since it exists only as a concept defined through three dimensional coordinates. Understanding and accessing this scientifically described three-dimensionality, however, necessitates a two-dimensional projection, constructed using the same rules as if it were showing physical architectural space. The Virtuvian and Vitruvian subjects, the skin and the human impose referencing control on their respective environments through the rules of projective geometry and with the premise of a subject’s scale and proportions.

The World of Warcraft character remained in that “world,” it also did not disappear upon a player’s exit from the game. However, with the proliferation of gaming worlds, a human could define multiple characters as multiple iterations and replications of a digital subject. Moreover, many of these characters can exist in more than one digital world. The character can further assume aspects of the appearance of their human and as alter-egos in the form of digital figurines; the characters can start to inhabit maps corresponding to real locations.^{^[8]} AI-generated alterations to the representations of the figurines adapt their appearance to reflect a real location’s specificity. These adjustments to representation are often independent of the subject’s perception but in direct response to the object of the digitally measured environment.

The absorption of real humans into representations, via digitally measured relationships to the environment — such as location, speed, proximities, frequency—has coincided with the relative loss of interest in traditional games — those with a predetermined quest to be fulfilled or a score to be bettered. The online games of the 2020s have moved away from the quest scenarios in favor of open-ended worlds. Popular titles, such as Fortnite, Minecraft and Roblox, which play out in open worlds, fall into a new category of “sandbox games” to signify both the absence of a preexisting goal and the possibility of modifying the environment. “The term “sandbox” is aptly derived from the concept of a child's sandbox—a play area where kids can build, destroy, and experiment without causing any real-world damage. Similarly, a digital sandbox allows experimentation and testing without repercussions outside its confined space.”^{^[9]}

In 2008, the creator of TheSims game—one of the two games that started the sandbox genre—Will Wright discussed gaming as a way to define “possibility space, build models of experiences, and collect reference knowledge called 'schema' that better enable us to successfully navigate through our reality.”^{^[10]}

The subject’s engagement with a sandbox environment’s object is simultaneously limited by the clear boundary of what constitutes the box, i.e. digital projective representation, but also freed by the countless multitude of digital objects to be created in the sand — digitally algebraically defined objects that can be perspectivally projected like the subject. The veracity of representation and credibility of persistence of the digitally described subject is made possible through the geometrical projection and shared time duration. In other words: the human subject is replicated into a subject who is mathematically defined in time. The replicate is further kept into existence and continuous, often autonomous, evolution by the many versions (various worlds) of an interconnected digital environment, continuously redefined into persistence. Some of these open digital worlds are referring to geographically determined locations, i.e., physical places on Earth that have been digitally projected as to become accessible to the digital subject.

Gradually, with the abandonment of the demand for a quest and the adoption of the open world sandbox model on the one hand, and the naturalization of skins into social media interactions — complete with the geographical positioning of the skin, also called a bitmoji, digital representation has transformed into open multiplication in the creation of a digital entity, a subject-skin.

The digital subject was conceived in a mode of avatarocentric control of the digital spatiality through the reliance on projective two-dimensional representations, such as perspective. With the digital subject’s continued replication, or spawning, to use the gaming term, and the multiple entries into digitalizations, and marked proliferation of digital two-dimensional representations of physical environment — tracked and pieced together through the many games and digital medias that are connected — the definition of the subject has become less precise. While visual representation of the subject, through projection such as perspective, was intended to conceptualize digital interactions as taking place in visualized space, overwhelming digital representation of physical space has allowed the digital subject to be scattered across physical and digital, leading to an ambiguous ironic positioning of the subject in relation to physical or digital space.[11] The human subject underwent a complete re-definition to be “spawned” in the form of an avatar — entirely in digital terms. Subsequently, the semi-autonomous digital replicas, the bitmoji subjects re-entered the physical world through actual geographical positioning referenced from the physical world. Meanwhile, the physical, solidly built, environment’s response to digitalization has been to either exist entirely as a digital model represented similarly to the subject, or alternatively — to incorporate some electronic means of referencing the physical object, some digital connectivity as a common reference system.

The fist appliance purported to have been connected to the internet (called ARPANET at the time) was a Coca-Cola vending machine located in a building on Carnegie-Mellon University’s campus — at some time during 1982. As the story goes, a graduate student unnerved by the vending machine’s location four minutes away from his office, and its irregular stocking, devised a way of checking on the availability of drinks without making the walk to the machine. After some creative wiring of the indicator lights and some programming, the computer science department fellow students were able to check the status of the vending machine from their respective computers. From the point of view of these students that surely appeared as a case of subject referencing object, a human subject describing, defining and ultimately controlling the inanimate object. Considering the manner of reference, however, the vending machine had to ‘communicate’ with the internet by sending the same kind of signals, or packets, in the same format, or protocols, as the human students did while exchanging thoughts via ARPANET-sent messages. That is the inanimate machine was behaving in a manner identical to the live humans. The objects in human environments able to communicate in a similar way will come to be described as participating in the Internet of Things (IoT) and by 2008–2009 close to thirteen billion such objects existed, which meant that more things than people were connected through IoT.

And while the Coke machine at Carnegie Mellon may have been a single device with a straightforward communication vocabulary of ‘full’ or ‘empty’, the IoT would evolve and amass into an entire typology of ‘smart buildings’, smoothly allowing for the gradual loss of the hierarchical conception of subject referring to object, where the subject-object or subject-environment connection deeply embedded in the center of past architectural treatises becomes ambiguous: IoT allows ‘things’ — appliances, furniture, entire buildings and urban spaces to be interconnected in a similar way and with similar weight as people being interconnected, i.e., the internet is an equalizing inter-net where things and humans have equally defined access to an equally shared reality. The conceptual duality of coexistent digital and physical referent systems that equalize the description of subject and object through both representation system (computer generated perspective) and referent connectivity system (the Internet) aims at intentional ambiguity and renders the attempted ambiguity ironic and the subject-object physically and digitally referenced relationships as existing in parallel rather than in a complimentary or equivalent way.

For example, lest this loss of directionality or weight in the subject-object relationship remains lost on architecture, a popular sitcom capitalized on the comedic aspects of a now commonplace situation — conversing with a refrigerator. A highly discussed episode of the sitcom “Modern Family”^{^[12]} showed one of its main characters, Cameron, literally singing a duet with his smart fridge (named Brigette) — a situation followed by an infuriated partner, Mitchell, finally resorting to unplugging of the intelligent appliance in a jealous fit of competition with the aforementioned device. The memorable scene is an instance of the contrariness of this relationship in a building to a subject-object relationship constructed through the commanding gaze of the human subject, as in the perspective-derived anthropocentric tradition of architecture, or in the orthogonally derived juxtaposition and clear differentiation (but not mutual negation) of the modernist subject-object relationship.

Similarly, a scene in the animated film The Mitchells against the Machines alludes to the concept of the inversion of object and subject. The operating system of what appears to be every consumer object imaginable has gone rogue and turned predatory towards humans. Toys, toasters, laundry machines, refrigerators, devices of any kind have come to “life” and the live humans are in the center. While the premise of the film focuses on the control and power of an operating system, the image of centrality alludes to the repositioning of the subject in an observed spot. The subject is not projecting the gaze, the subject is being objectified, it is being studied, measured, contained. The former objects, on the other hand, are finally orbiting around the studied humans — in a reference to digital modeling and digital twinning, the human subjects are digitally repositioned in the focus of a controlling omni view of objectifying data collection devices and revealed as a twin model to be destroyed and reconfigured.

Finally, the premise of the IoT rests with an idea of geometrical representation of the Things connected, regardless of the parameters reported, as it is the perception of the Thing’s presence that is being manifested to the subject’s self-location on the outside of the object, i.e., measured parameters are being transmitted to the Internet in order to ascertain their visibility in connection to a human subject visible to a shared internet, as a system of geometrical projection.

The representation of a physical environment, the architectural object, becomes indispensable for the object of architecture’s entry into a relationship to the digital subject. Enter the concept of the Digital Twin.

A digital twin was first intended as an insurance policy for an object in crisis.

Building physical models of various scales is not a novel design practice. The use of such models is unidirectional — from model object to real object — and any change in a representational model is initiated from outside the object and actuated unidirectionally towards the eventual real design product. The original and the copy are clearly defined, whereas the model, even if historically preceding the real, is considered a replica. Moreover, the unidirectional connection is maintained only for the duration of the design process. A digital twin is conceptualized in an entirely different way, made evident by the adopted term. Twins are not replicas; one twin may be identical to another but neither one is the copy, nor is either one an original. Twins are their own selves. Considering a physical and a digital entity as twins assumes their equivalency.

While the actual term digital twin came in use in practice through a NASA report from 2010, the concept has been deployed by astronauts for at least the past fifty years. The digital twin’s usefulness, however, was highlighted during the narrow averting of disaster during the Apollo 13 mission. The characteristics of the simulators of the aircraft modules, which qualified these particular simulators as the first digital twins are as follows: the physical object twinned is out of reach (for example in outer space, but can also be otherwise inaccessible) but also continuously connected.^{^[13]} Further, digital twins are adaptable and responsive to which two characteristics are relatively easy to achieve in a digital model; the lunar aircraft’s more than a dozen models and simulators were physically changed and re-engineered multiple times during the three day duration of the crisis.

While a sequential record of the events is feasible, it was the possibility to maintain a continuous two-way connection between an aircraft’s changes far away in space and the multiple re-engineered responses of the former duplicates housed in a building in Texas, on Earth, that makes the duplicates format of existence comparable to digital twinning. The remote connection between aircraft and control center required both devices and coding to make the continuous re-referencing possible — digital twins construct a relationship as an entity in itself; further, the object of the link as a persisting element becomes essential for the continuous intelligent (AI) existence of both the digital twin and the twinned reality. If indeed, it was a Coca-Cola vending machine that introduced the IoT, it is not surprising then that the possibility to conceptualize a link as an entity, a function, a service such as vending, outside of both subject and object, that would allow the IoT to facilitate digital twinning.

Since the connection between physical and digital twins does not need to be rigidly predetermined, it is a connection that is based on learning between the two entities in what would be described as artificially intelligent way of observation, however limited or expanded the similar situations could be found. A smart building, like the Edge, then is performing similarly to sandbox online game, like Fortnite. Both the building and the game claim as their main purpose accommodating people, for work or entertainment, both claim ingenuity and innovation based on a similar digital system of geometrical projection as a twining relationship and mathematical positioning. Architectural experimentation in the early 2000s relied heavily on the use of sensors to provoke spontaneous responsive change in buildings. Building elements connected to a particular monitoring sensor performed a function predetermined by the algorithm embedded in that system — most often the actuated alterations concerned the actual geometry of the building in question. It was only through the use of sensors through AI-powered twinning that alterations did not need to be actuated: an action became based on similar situations. Most importantly, and most ironically, in both the building and the game, functionality of the work place or the gaming experience is not compromised vastly because of the individual compliance of participation — whether an employee opts out of tracking or a gamer takes a break — the building/gaming place persevere in physicality and as digital twins, and the missing participants in work/game are smoothened over by the best guess of the digital intelligently generated compensation. Ironically, this flexible indeterminacy is achieved by the fastest and most precise of contemporary tools — the artificially intelligent ones.

Non-ironic architecture, on the other hand, is an object of determinacy and certitude; it is the antithesis of an object in crisis — it solemnly acknowledges its own heaviness rather than hide or question it. Non-ironic architecture is monumental, it echoes and reflects a subjective presence without re-instantiating it — instead of replicating, it reaffirms. Light and shadow in such monumental architecture become the measurement and validation of presence and solidity. The human subject in the context of monumental architecture, for example in a Louis Kahn building, is grounded; his or her shadow stretching across the surfaces and following the outlines of heavy architectural form. Analogously to an avatar establishing reference measure in gaming space, relying on the assumption of a player’s knowledge of perspective of one’s own dimensions. A Kahnian shadow^{^[14]} establishes the boundary and reference between human and architecture through the explicit projection of shadows on the solidity of a building. In both cases, the projective mechanisms are used to reconfirm credibility and stability.

Shadows have been used to indicate the passing of time, they have also been cultivated to articulate space through distance, scale and form and to sharpen and intensify atmospheres of place. There is also a particular case of a shadow in crisis and the ensuing ironic reading of a spatial subject-object relationship as seen with the fictional persona of Peter Pan, especially in the latter one’s capacity as an antecedent of the avatar in digital space.

First introduced in 1902 by the Scottish novelist and playwright J. M. Barrie, Peter Pan is most famously endowed with eternal youth and an ability to fly. While perpetual childhood has been discussed at length in literary, psychological and philosophical studies, it is the ability to fly that sets Peter Pan’s presence inside a house that is at odds with the house’s inhabitants. Buildings being grounded, and architecture conceptualized through the experience of a similarly grounded human subject, Peter’s independence from gravity during interactions with other children in the house displaces the stability of the subject-object relationship. The shadow of the flying child is expectedly disconnected from his body — while in flight. It stands as a reference to the surrounding space and a marker of the solidity of the walls and the distance between the body and the environment the length between subject and object. The naturalness of a body-shadow connection, however, is overturned when Peter touches the floor, a chair or a windowsill while his shadow does not connect to his body and alludes to a disembodied experience — a detachment of the subject reference marker to the object. The continued presence of a shadow, nonetheless, attempts to establish this marker of the subject as an entity in and of itself — while the idea of a subject without a visible reference to the object, a creature without a shadow is not novel —- Barrie must have been well-versed in his native Celtic lore describing demons as creatures without a shadow, or a reflection for that matter, as an allusion that those non-corporeal beings might exist only within the mind of the beholder. Returning to an avatar’s reliance on perspective as a tool in ascribing subjectivity to the digital object by simulating Borromini’s technique of immersion, it is noteworthy that the relationship between Peter Pan and his shadow only enters the realm of irony by virtue of that referential object momentarily taking the subject’s role and space — thus allowing for the shadow to become both a measure of space and a creator of an alternative space — a digital twin so to speak. Curiously, the excerpt introducing the free will of Peter Pan’s shadow reassures of its “normality” as a mere projection and simultaneously normalizes the possibility of relating to that projection as a physical entity in its own right — the shadow can be “folded” and “put away in a drawer”:

The independent movement of a shadow which is both made accessible to perception by adhering to building surfaces and inaccessible by defying the geometrical principle of light projection as well as gravity’s force on the subject casting the shadow. The runaway state of Peter Pan’s shadow conjures up alternative modes of inhabiting architecture, as well as alternative ways of subject projection.

In Peter Pan’s case, the marker of body in space appears to be only temporarily detachable, foldable and even trappable. The shadow, as a pure signification of a relationship between a subject and an object, is behaving as an entity which alternates between referencing the space and referencing the subject, or behaving as a copy and behaving as an independent runaway subject. The identification of a relationship between a subject and an object as an entity allows the constructive use of ironic tension between intention and reality. Identifying this relationship as an entity outside the subject or the object defines an entry into the concept of the digital twin.

While the simulators at hand during the Apollo 13 crisis can be retrospectively theorized as digital twins, it was only in 2010 that the idea took shape and was named. During a University of Michigan conference, the NASA scientist John Vickers presented a “Technology Roadmap” that would introduce the idea and implementation of something called a digital twin. In that report, Vickers defines the concept as “a set of virtual information constructs that fully describe a potential or actual physical manufactured product from the micro atomic level to the macro geometrical level”. Digital models had existed for several decades before 2010. What differs in the concept of a digital twin’s implementation is the third component: original, copy, and link. The link between the physical and the digital twins is continuously maintained and, moreover, it is a two-way link that permits the digital twin to actuate changes in the physical one. Most of the time this link is created as an AI entity.

Returning to the example of The Edge Amsterdam to re-examine the subject-object relationship in the context of digital twins.

As discussed, the Edge sensors continuously measure and monitor any parameter prone to change — temperature, humidity, occupancy, light, air flow, water usage and local climate conditions, functionality, paper stock in printers and even the readiness of coffee machines and assignments of desks while the occupants are free to move around and change their workday spot with the help of a dedicated telephone app:

The relationship between building and occupants, object and subject, is designed to achieve extreme precision of coordination by replicating physical reality into a digitally defined “identical model”, a Digital Twin; the impossibility of a precise relationship or an accurate replication becomes clear when occupancy produces ironic incoherencies: to facilitate the link between building and digital twin, the building’s client and main tenant, Deloitte Netherlands, distributed smartphones with a pre-installed Mapiq app to all employees. The dedicated app can be used before any activity is undertaken — apart from finding a suitable workday spot, it can also locate colleagues, read a meeting’s location from one’s online calendar and suggest the route to that location, check the schedule of trains for after the meeting, track progress in the on-site gym, order food and otherwise continuously micro-coordinate life in the building for over 2800 employees. Rather than architecture revealing and ordering the space of the human subject, the movement and the view of the latter are engaged and guided through a digital device system that references the building and the humans in a deliberately non-architectural manner. According to a case study produced by researchers from the Architecture Department at the University of Cambridge,^{^[18]} not everything in the building can function by sensor command alone. The clash between the intended effect of optimal occupancy — digitally defined and optimized by AI learning from existing databases of building usage — and the reality of spontaneous humans at work can be described in ironic terms. For example, since the existing number of desks available to be assigned within the building are only half of the number of employees — presuming meetings, home-office days, vacations, etc. — some days, most often Fridays, the mass of “surplus” employees arrive looking for a desk only to find themselves crowded in non-working areas like the cafe, gym, and lobby. While the full convivial cafes might enhance the experience of the building, the unintended distribution is clearly at odds with the precisely monitored optimized spaces exactly because of the precision of the intended occupation. Moreover, in another problem in this complex orchestration of architectural experiences through digital devices is that the sensors embedded in the building can only communicate with ‘willing’ smartphones; “Deloitte’s employees can select, and often do, to not be tracked by their Mapiq apps.”^{^[19]} Consequently, the responsive features of the building are not able to be accurately adjusted to the reality of physical humans. Ironically, the perception of digital markers that the building seeks would be most accurate at the times when the building is entirely empty as tracking noncompliance would be theoretically impossible only at those times. The physical and the digital realities, the building and its digital twin, bypass each other even if conceived and conceptualized as identical.

Digitalization has allowed replication processes to achieve a speed and accuracy unattainable in previous times. The change of terminology from replication to twinning,” however, is not a casual one. The digital copies are likened to identical twins inasmuch as these become entities of their own — the physical original and the digital equivalent. Unlike identical copies of the industrial age, digital copies are identical only in definition and not in material composition. The precision of replication is greater but the copies are entities of their own existing only through mathematics and perceivable only through geometry. It is then natural that AI methods would attempt to lose the circle of replication and base further mathematical definition, i.e., further entity creation, on the basis of perceptible visualization only — the AI approach is not unlike a projective geometry system. Yet, perspective and axonometry assumed either a single, mono vision construction of perspectives space with a single point focus or the omni vision parallel projection of axonometrical space with a focus on infinity. AI projection foci are dispersed and multiplicitous, but definable.

The process of creating digital copies of the environment, i.e. to objects, extends to defining digital copies of humans. Many copies. Both subjects and objects are twinned, duplicated and interlinked through multitudes of sensors and algorithms. Unlike the IoT which equalizes subjects and objects through attributing “smartness” by a process of access and connection, digital twinning equalizes by attributing intelligence by a process of referencing, controlling, adjusting, correcting and monitoring any potential difference. Continuously. The Edge Amsterdam is touted as the world’s smartest and most intelligent building in part thanks to its more than twenty-eight thousand sensors controlling the capacity of rooms, tables, parking spots, bathroom usage and cleaning staff, occupants’ location and personal habits down to individual humidity preferences, solar energy usage among other controlled values. Perfect twinning of all imaginable criteria is the ambition. The emergent actual disparity between twins and duplication processes, especially the disparity between intent and result as in irony, occurs when the individuals supposed to be observed and tracked in order to enhance the occupied building by literally becoming a part of a symbiotic organism refuse to adhere or participate^{^[20]}. Hence, smart buildings can open to constructively actuated juxtapositions of the scattered subject to the architectural environment, the ironic relationships of simulated discreetness or incompleteness of a subject to a similarly unevenly defined environment.

With the conception and actualization of “reals” as human subjects and “copies” as inanimate digital subjects in interaction, succeeded by things and beings interacting over the IoT—which happens to also be the Internet of humans—thus animate and inanimate internet-connectees are equally digitized, the equalizing smoothness of AI proliferation is different from the scaling of the Industrial Revolution, rather than being expansive by means of multiplication, the digital AI revolution is densifying within creative boundaries of existing architecture by means of proliferation.

A Provisional Conclusion with respect to the irony in architects’ fear of technologically anachronistic architecture: “Technology is the answer but what was the question?”^{^[21]}

If AI is used for optimization in digital twins, especially in large models like smart buildings and smart cities, then generative AI would have to anticipate rather than react and optimize.

On the back of the pre-histories of digital twins interconnected with their originals through the IoT, ambiguity has befallen architecture’s form since 2022 when ChatGPT and other visual generative AI tools were introduced — seemingly at everyone’s disposal. Ironically, what would normally be data flowing from physical objects or subjects to be used in optimization functionality tested on digital twins, has become data accessible “midstream” and utilized for the free generation of not optimized but fantasized representation of essentially a digital twin.

The availability of easily accessible methods of digital visual generation, even creation, is an availability that has reached the top of the Gartner hype curve^{^[22]} and it appears that every aspect of life is eager to claim some AI-ness; the speed, ease and sheer sleekness of particularly of AI-generated images has been nothing short of amazing. Nevertheless, it is the AI images’ two-dimensional nature which makes them relevant to architecture, and it is the artificially or rather superficially intelligent way of surfacing with immense precision and speed, without the help or the need of depth or even translation of any three-dimensionality defined mathematics as with previous digital images. AI generated images are perspectives that are not revealed through construction, movement that is not expressed through vectors and they are disconnected from both sides of digital twining from subject and object to achieve ultimate lightness.

If indeed, architecture relied on the connection between mind and eye and the conception of space occurred in a mechanical geometrical manner, which explicates the three-dimensional order onto a two-dimensional plane, then computing is different because digital space is already three-dimensionally represented through coordinates, algebraically represented in matrices. The possibility for a human to “access” computational or algebraically expressed space is only through a graphic representation. The appearance of perspective is, on the one hand, the mental expectation of a geometrically constructed monocular perfect perspective and, on the other hand, the algebraic matrix divided by a z as a depth coefficient which provides a way of “understanding” and touching of the two conceptions.

Moreover, it appears that the very instantaneity of transition from conceptualized responsive architecture, which constituted an architectural conceptual search for responsive form, to a near-total abandonment of formal ideation and giving in to the seduction of singing with a fridge, might have also suspended architecture’s rhetoric. Such an ironic suspension of architecture is precipitated and actuated by the disheveling of its subject’s integrity in a more profound way than ever before — rather than a Frankensteinian recomposing of the entity or the idea of the subject from parts, that subject — in parts and as a whole — is simultaneously multiplicated and scattered.

The subject is no longer observing but being observed while data is “extracted” from the human, formerly acting as a subject, by the object. Each of the data extractions, in turn gives the possibility of a digitally defined subject — definitions are multiplied as many times as the number of data set types. The digital twins of the human subject are multiplicated; they are also scattered to respond to the temporal or positional optimizations initiated by the digitally twinned object.

Thus, in the digital era relationship between subject and object, the multiplicated scattered state of the subject(s) is being referenced by a multitude of twinned objects. Such a reiterated subject-object digital continuum is dependent on the optimizing processes of functional pattern detection, i.e. artificial intelligence. The equalizing—optimizing—agency of artificially intelligent processes linking twinned subject(s) and object(s) operate within a boundary “around” all acceptable sources which are digitally referenced and can be used to generate new combinations, AI-generated form. In the context of AI’s manner of operation from and within a boundary of possibilities, it is important to notice here the contribution of Alan Turing who created the very concept of artificial intelligence, or in his words, machines capable of thinking — at the time only as a theoretical conjecture. Turing devised a method for ascertaining the intelligence of machines, where the ultimate test consisted of the ability to understand and generate human speech, the Turing Test.^{^[23]} The test, conceived as a playful conversation scenario, was originally called the “imitation game” in reference to the computers’ pattern-based decision making — significantly in the discussion of a boundary around possible data, around the same year, 1948–51, Turing co-programmed an example of machine learning — Turochamp[24] which can be considered the first digital game…a chess game. While the program proved too complex to be run successfully by the computational devices of the time, it is noteworthy that this early form of an AI method was applied to a form of gaming and solutions were sought in a finite pool of possible moves, a sandbox open to internal progressive complexification.

Because the source data possibilities are finite in quantity — regardless of the vastness of the boundaries around the available pool of data — an eventual saturation of AI derived outcomes can be theoretically reached. In anticipation of such saturation, AI processes lead to a smoothening of all results and variability being sought in “unusual” combinations from within. This anticipatory saturation of AI normality appears to have been taken up by a culture of collabs — superseding the “both/and” credo of postmodernism from fragments to an amalgamation of formerly singular entities — superheroes like Batman and Superman, Godzilla and King Kong begin to feature together in films, former fashion adversaries like Gucci and Balenciaga team up to rebrand in collective and collectible items. The collaborations have even become more desirable than the “single origin” entities — accordingly, it appears that architecture is steadily attempting a formal collab with computing.^{^[25]}

The irony of such a collab, however, is revealed as not only are the subject and object in a relationship with continuously switching directionality and questioned ends, but the plentitude of physical sensors, or AI-ready data, which should produce this “collab” conspicuously do not include any architectural generation sourced from intrinsically architectural data and [A]Intelligent buildings’ architecture intelligently continues to react to the ever stable “sensor” for gravity showing 9.8m/s2 and [A]Intelligently generated architectural design continues to be dependent on 2D representation and geometrical projection.

Finally, the object itself, the architectural environment, which has most recently been through the consequences of rhetorical purification (modernism), recomposition (postmodernism), dematerialization (parametricism) appears to be in a relationship with a “naturally artificial” subject through an AI-induced digital smoothening. The smooth AI-generated images often appear “retro” and nostalgic, as a romantic vision of an easy life from the past which is co-linear with the promised easiness of the present, this outcome is a ready-for consumption imagery. But if irony is an act rather than a significance, then architecture which is “activated” through AI is about to come into being through constructive irony.

The scattered self, measured and tracked, defined continuously with precision — not only geometrically described but defined by measurable parameters — and the change measured in time is related to a fragmented and layered environment. The intelligence of learning from observation, as opposed to experience, is only as thorough as the view, hence the visual nature of artificial intelligence as well. All the information or thoughts that fall outside the view are outside of consideration, all generated “artificial” knowledge” is added to the pile of examples.

AI generative tools — creating something new in textual, graphical, spatial form — rely on the super-fast examination of super vast databases of past examples (to “learn” from) but the multiple duplication, replication, multiplication, re-equalization and twinning already described is creating a pool of examples that are already quite similar to each other. In this relative homogeneity, architecture is left with the ironic task of finding order by searching for hierarchies and proportions in the sameness. Enactments of artificial intelligence rely on a mimicry of a process rather than product — i.e. AI tools “let the data do the programming” ; and while irony stems from a disparity of intention and effect in the process of duplication, it is the mixing of the different kinds of duplicates during in the digital twinning, and their subsequent accessibility through the IoT that allows a “smoothing over” and the legitimization of instances of newness as products of generative AI that are non-traceable simulations. Is it possible that an AI cave cannot be exited?[26]

The parallel existence of technology and buildings, as in the intelligent building approach, has precluded a meaningful co-relation of the AI optimized environment of a building and its structure and image. For example, the mechanical systems in any contemporary building that can be, and are, controlled by sensors to make that building intelligent and can be continuously optimized and adjusted through artificially intelligent solutions are not necessarily a part of that same intelligent building’s architecture. The creation of digital doubles has produced an entirely new layer of devices associated with a building — quite independent of the form, style or architecture of said building. In such case, architecture might not have to question its own conception and order in response to the AI world but rather consciously disassociate from all the IoT measuring equipment and rethink itself?

In the case of intelligent buildings, artificial intelligence glides past the object of architecture, concerning itself with concurrent optimization of experience that may not be architectural. The object of architecture and (artificial) intelligence are sidestepping each other through the creation of a digital twin. The creation of the “artificial” architecture of the digital twin, defined by tracked values, arguably removes physical architecture, in body and concept, by the tracking process.

Conversely, but no less ironical, in the case of architectural design through intelligently generative “artificial” space, which is accessible in representation only, the necessary step of 2^d projective geometry such as perspective for the re-instantiation of a digital-space-dependent avatar subject, arguably removes the human subject outside of the time dependent tracing process of digital subject-to-subject relationships. In both cases, the AI-ntelligent implementations in buildings and AI-generated architectural design, the dialectic of irony in architecture is expressed in that the sophistication of the digital twin and the capability of artificial intelligence appear to be focused in a direction that does not directly alter architectural conception or perception — yet.