SHUMI BOSE/KOOZI’ll confess that I am burdened, as many of us are, with misgivings around the idea of systems logic, machine learning and AI. I'm also intuitively wary of taxonomies and systemic classification. Yet along these same lines, there are synergies between your respective research on codes, language and learning.

MARIO CARPOLet me just offer this: when you say that you have an aversion to the logic of systems and things of a mathematical nature, so does the computation of artificial intelligence. Generative artificial intelligence is not a rule-based system. If you are a classical mathematician, you dislike generative artificial intelligence — or artificial intelligence in general — profoundly, because it is an irrational machine. It's stochastic; it's probabilistic. If you run it twice or even many times, you will never get the same result — it's like working with a madman, not a scientist. It's just crazy. It is irrational, wild; it is the savage machine, the expression of a savage mind. That’s actually not my expression; it's the title of this book (by Ginger Nolan, Savage Mind to Savage Machine: Racial Science and Twentieth-Century Design, 2021).

KOOZSuch examples, parlayed into so-called “racial science”, only underline the shadowy capacity of data-driven machine learning. Yet you both work quite creatively and even optimistically, about the attempts to rationalise ideas of language, code and modes of communication in architecture.

MANOLIS STAVRAKAKISYou know, it's funny that each of us has come here thinking about our own questions towards the subject. I read a recent piece that Mario shared on artificial intelligence and wanted to know how to comment on it. I was reminded of this piece by Carlo Ginzburg (and Anna Davin, titled Morelli, Freud and Sherlock Holmes: Clues and Scientific Method, 1980)

MCI know the text — it’s about the idea that you can detect the spirit of a work from a hidden lapsus, and one may trace it as Sherlock Holmes.

MSExactly. It came to my mind because AI, as you described it, creates layers of information to build an archetype; then, from the archetype, it makes all of these variations within a theme. I was thinking about the human ability to make similar jumps in thinking; that’s not a rational or a straight-forward process, it’s our default manner of connecting information. I was wondering if AI would make this jump of association, to overcome this idea of the layered, aggregated knowledge.

MCYeah, it is a leap in the dark, which is what the machine surprisingly does. One thing to bear in mind is that when we talk about generative AI — which is the topic that I’m discussing here — this is not related to classical artificial intelligence, which has been around since the late 1950s. That was a rule-based system, invented by mathematicians, computer scientists and engineers. That follows the mathematics we studied at school. In that system, two plus two is equal to four, ten to the power of two is one hundred and that's it.

The artificial intelligence that we're talking about is a completely different machine. It is an evolution of what we used to call machine learning, and a few years ago this technology started to work remarkably well for manipulating images, and then they extended it to the manipulation of text. This was followed by multi-modal text-to-image and image-to-text generation, which is what our students are now using all the time. But the spirit, if I might say, or rather the technical logic of the machine is something that would drive a mathematician or an engineer crazy, because what the machine does is completely irrational.

Take the example of image generation, as it is perhaps more familiar. It all starts with a collection of images; they call it a data set. It's a curated selection of images. Assuming that you are working with dogs, you make a collection of one thousand pictures of dogs. Now, what the machine does is that it finds the commonalities among all these images. The machine doesn't know that they are dogs, because you made the choice; only you know that they are dogs. The machine just tries to find out what all these images have in common; their essence, so to speak? If these are all dogs, the machine somehow arrives at the definition of the ideal dog. What makes a dog, a dog?

KOOZIt reminds me of nothing so much as the Platonic cave.

MCThis is Platonism; the ideal, the super-dog, the Platonic idea of dog. Now, there are differences. If you are an Aristotelian, you define the dog with words. If you are a renaissance Neoplatonist, you have an image of the perfect dog, the idea of ‘dog’ in the mind of God. The point is the system extrapolates this mysterious definition from the available data set, which is what we'll define under a certain nomenclature. In the Renaissance, this common quality or definition would be described as ‘a certain something’ they have in common: nescio quid occultum. But the machine will never spell it out, because this intelligence of the machine is called a latent space. To the students who ask if latent space is a space: yes, it is a mathematical space. It has a matrix of 1000 million vectors in every direction. It's unconceivable, unimaginable; pure machine space.

So we don't know exactly what it is, but we know what it does, because now when you show a new picture to the system, which has assimilated this idea of the dog, the machine will tell you if it's a good match. It's a 92% match with the idea of ‘dog’ that I have in mind. So it's an analytic question: is this a dog? Answer, it's a good match. But then the part which is more interesting for us — the generative part — comes when you ask the machine to create a fake dog, a dog which does not exist. The machine imitates the data set. It is creating a fake dog which looks like a dog insofar as it is similar to all the dogs known to the system. It's an imitation machine, a machine that imitates the data set which we have fed into it. The technical term is that the machine is trained. A machine trained on dogs generates fake dogs.

The problem, if you look at it as a classical art historian, is that the machine is imitating. Imitation is a mystery, because we don't know why this dog is similar to all the other dogs. There is no mathematical definition. Why does something look alike? The classical tradition is full of metaphors trying to explain similitude. The typical example of similitude or resemblance or looking alike is the similarity between a father and a son. Mind you, not a mother and the daughter; it’s only the Father and the Son — because this is the classical tradition. So let’s say that you can identify a resemblance between the Father and the Son. But why? How can you tell? Do they have the same nose? They are not twins. Do they have the same lips? Do they have the same eyes? You can't tell. You can just feel it. It's an aura, it's a feeling, it's a fragrance, it's a taste, it's a manner, it's a style you recognise.

KOOZYou’re pointing to a crucial difference between intelligence — which is discerning — and imitation. Perhaps we should call it artificial imitation, as the promise of artificial intelligence suggests something different than the premise of iterative mimicry.

Yet systemisation does seem to be generative in your research, Emmanouil, studying the work of the architect Michael Ventris and his work on deciphering the lost language, Linear B. As I understand it, it deals with the possibility of architectural grids both to delineate a certain linguistic logic, while also generating a structural disposition or mapping of its script.

MSOf course, it's very complicated when you have to decipher an ancient script, right? So what I was trying to argue was how Ventris used the exercise of mapping buildings exactly as an exercise that allowed him to create spatial relationships within the undeciphered text of Linear B. Because he was trained as an architect, it was easy for him to introduce relations among the signs of the script by creating an abstract map in the form of the grid. In that sense, the mapping of the buildings was used as a preliminary exercise in Ventris’ mind on how to map relations of abstract entities on a piece of paper. In other words, Ventris began constructing and mapping internal relationships among the signs, which allowed him to understand which were their common elements. This then allowed him to make the jump in thought — the leap in the dark — to decode the script. So devising that system was a tool for him.

MC.. which he could use as an architect, because he knew how we do formal analysis. Today, he could have probably used shape grammars in the same way.

MSExactly. There is also the Aristotelian idea of the relationship between form and function, in that you can define a unit according to those two characteristics; this was the mode to decipher the characters themselves. But Mario, you triggered my mind here: using the analogy of Linear B, when you are deciphering an ancient script, the more documents of the script that you find, the larger your sample for statistical analysis. The more material you have, the easier it is for you to associate elements with each other. As you were saying before, with generative AI, it starts with a bulk of images, let’s say one thousand dogs. The system defines a dog; the super dog, let's say. But what if I go back and provide an input of another thousand dogs? Will the super dog be redefined?

MCOh, yes! That’s how it works. The super dog is the commonality among all the pictures, so if you expand the data set, it will change; that definition of a data set changes all the time.

MSThat's the interesting thing; the more you feed it, the quintessence of what it's trying to define will change. And there is no end to that, right?

KOOZIndeed, the word quintessence, which specifically defines or relates to the undefinable fifth or quintile element.

MCExactly. It is the spiritual fifth element, which we can’t define and don't know what it is. It's a mystery.

KOOZWhat Ventris achieved with Linear B is endlessly intriguing — precisely because it operates not only these two levels, systematising but also mapping to decipher a script. But why is this notion exciting? What does it generate in terms of knowledge creation, computational intelligence and in terms of design?

MCIt's a good question. Why does generative AI matter in design? Honestly, I don't think it matters very much. It may change tomorrow, but for the time being, what it does is practically nothing, even considering the early euphoria of students when the technology came out. After six months, they realised it's basically a video game and moved on.

However, there are practical implications for people who use the technology to replace Photoshop, for instance, to make renderings cheaper, faster and so on. Some expertise is already being replaced — but these are anecdotal changes. They don't deal with the spirit of our profession. There are some offices that have understood how the machine learns to imitate a style, based on the data set which you feed into the machine. If you feed the machine a data set with all the drawings made by a particular architect, the machine can learn a style and replicate it forever; we can have Zaha Hadid forever — which of course, no architect interested in design would accept.

Of course, there are architects who actually build the same building forever; in the case of Mies van der Rohe, as he actually was proud of doing so. As he said, you cannot invent a new architecture every day. But you don't need artificial intelligence to repeat a building by Mies, you just need just descriptive geometry and a photocopier. But two or three years ago, a team at Coop Himmelblau published what they were doing. Into a machine learning system, they fed all the sketches drawn by the principal Wolf Prix — since he was a child, I presume. The idea is that the machine has actually learned the peculiar, unique style of Wolf Prix — which, we must admit, exists. Wolf Prix has a recognizable style, so the machine will repeat the style of Wolf Prix forever, long after he’s gone.

KOOZLet's pause on that bombshell of a possibility. Emmanouil, back to the question of why all of this matters today.

MSFor me, there is only one word that matters, and that word is intuition — as per the thesis I completed with the late Mark Cousins, at the AA. Earlier in this conversation, we described the process where Ventris applied the knowledge that he had acquired as an architect into another field, namely decipherment, right? Well, whatever knowledge he was applying in decipherment — that he was not fully aware of applying in that category — we call it intuition.

MCOr tacit knowledge.

MSOr tacit knowledge, exactly, and it’s an important concept in archaeology, especially. In my ‘viva’, Mark Wigley engaged with my claim that if everything in the thesis that we called architectural was identified as intuition, why not claim that intuition is architectural? That is, of course, a very Wigley-Derridian way of reading the relation between architecture and intuition…

MCThis is post modernism. If you are a modernist, you don't like intuition: you want a handbook with the rules all written down. You’ll have a manual on reinforced concrete to calculate an I-beam, leaving intuition to a minimum. Equally, you would also not like imitation. When you were studying architecture, did you ever hear the term imitation? If a teacher said, you're imitating someone, was it ever meant as a compliment?

KOOZAt that point, I believe originality of concept was still the goal.

MCNormally, if you're imitating you're bad. Imitation was either forbidden or a derogatory term; at the time when I was a student, it meant plagiarism. This is the modernist idea of imitation, identical replication: it is inherently bad, and if you do it, you're a criminal, because you're stealing. You take something from someone else's work, you claim it as your own: it's copyright infringement.

Now in the classical tradition — which is what modernism eliminated and cancelled for many good reasons — imitation is not cut and paste. It's inspiration, assimilation, transformation and translation. Meaning that if you are imitating in the classical tradition, the original and the copy have a certain something in common, but you cannot tell what precisely. If you can easily say what they have in common, then you're a plagiarist, a charlatan, a moron — you're an ape, the term which was used during the Renaissance. If you take something from someone else's work you put into your work, you're stealing and you're an ape — because you're imitating without knowing what you do. This is what bad artists do. But good artists get inspiration from a model. Inspiration is assimilated; it feeds you, it becomes part of you. Then you create something new, and you can tell — there is something in your work which derives from something else, but precisely what comes from where is intuitive.

KOOZSo why are these ideas particularly valuable in architecture right now? The words we’ve used today — intuition, inspiration, influence — are not as readily associated with so-called artificial intelligence as others: invasive, insidious and invidious, for example.

MSI can answer. I think that one of the main things that architectural education does to you is to make you feel comfortable with not knowing what you're doing. How many times did our tutors tell us that something is not right, without really explaining what is right or not? In terms of design, One of the things that architectural education does is to actually teach us how to be comfortable with being intuitive, which means not knowing exactly what we're doing.

MCBut there is a generational factor. You were trained by postmodernist wackos, whereas I was trained by good old modernists. Take Christopher Alexander’s Notes on the Synthesis of Form (Harvard, 1964). There is no intuition, no imitation, there is nothing. There is a data set, even if we didn’t use that name. You make a plot, you make a diagram and then step-by-step, you generate — through perfect rationality and deduction — the form which will fulfill the requirement of the problem. It's a rule-based system, and a dumb computational machine could do it faster than we do.

MSMario, I will disappoint you: in my Greek architectural training, postmodernism never arrived! But I’ll add something else. Ventris was educated at the AA within a pure modernist system, which I've analysed in my PhD — and it's exactly as you say. It's deductive problem solving, as clear as mathematics. Yet if you look at Ventris’ work in decipherment, there is a moment where he says:“I can keep analysing forever, but unless an intuition comes to me, I will never be able to solve the problem.”

KOOZIndeed, the connections when you see Ventris’ famous architectural section — with the scratchy little pencil drawings of the Linear B plates alongside — the suggestions made by his syllabic grid can only be intuitive. Even in Christopher Alexander’s Pattern Language, I found some liberty in the act of defining diverse, often informal spatial conditions as generative typologies.

MCAlexander eventually mitigated his position and lost the rigour of his youth. But Notes on the Synthesis of Form was about eliminating the leap in the dark — or at least reducing to the minimum. You formalise the premises of the data set to such an extent that form is generated almost by logical inevitability: that was the dream of modernity. Form follows function or whatever, but it follows something: there are premises, there is a deduction, there is a mathematical method, and you get the result which is verifiable and repeatable.

Of course, architects never really believed it; they just pretended they did. We were told that imitation was bad, as I said: it was a forbidden word. To say that someone was imitating was to say get out of here, you will never be an architect. Another term which was censored was style, because it came from the legacy of the classical tradition, which we wanted to eliminate for many good reasons. Imitation and style were forbidden — yet we all knew that style existed, only we could not discuss it. We all got inspiration from models that we imitated, but we could not conceptualise it as such. Imitation became something that we did, but we couldn't discuss it critically, because it was prohibited. It was practice without theory and we did it in secret.

KOOZYou have set me up quite well here. We're talking about what was either proscribed or prescribed, in architectural training, as a way to minimise risk or creative freedom. Where you stand now, and what do we do with these parallel understandings of creative influence versus imitation?

MCWe are dealing with a machine which is embodying the irrationality that we once tried to eliminate from our profession? Generative AI is a machine which automates imitation — but not imitation in the modernist way, meaning the copy-paste plagiarism without citation or identical replication. Generative AI is bringing back the idea of imitation, which was embedded in the classical tradition as inspiration, assimilation, and transmutation. Why should a machine come to remind us of what we always did?

Of course, the machine doesn't have the intelligence to do it critically, but it is only doing something that we had forgotten; something we don’t even have the critical capacity to discuss, because we lost the terminology. There is a book by the literary critic Harold Bloom called The Anxiety of Influence: an entire book on influence where the term imitation is never mentioned once. Influence is good but imitation is bad, because under the canon and the dogma of modernism, imitation has a stigma as something that should not be done.

KOOZI should note that we’ve talking primarily about images in generative AI, rather than text. In your roles as educator, how does this sit in terms of what you receive from students?

MSYou see, because my whole research is on Ventris, it's the analogical way for me. I can tell you that I'm a strong believer in the human capacity to be intuitive… I'm a strong believer in the possibility of being able to create something full of faults. I believe in human intuition, in the leaps of thought that a human can make. I could almost be the counter-AI person; I want the mistakes, I want the faults. I was wondering if a student brought me an AI-generated image, would I be able to recognise it as not made by a human? Especially for history and theory, it makes a difference between English and Greek, because AI in English is way more equipped than it is in Greek, right?

KOOZIt happens all the time. So when you're confronted with a piece of work that is AI generated; how do you respond?

MSI’ve never encountered it. What do you do?

MCFor the time being, the commonly used systems to generate images are so primitive that it's quite easy to tell when a picture has been generated by AI. With text, it's difficult to tell; I use generative AI myself — for editorial purposes, for translation — and there is stuff that it does well. It's been fifty years since Umberto Eco wrote a famous piece claiming that he could tell by reading a text if it's been composed on a typewriter or using a word processor. Every tool we use feeds back on the stuff we make, and so sometimes that tool leaves a trace.

KOOZCrucially, is it a problem?

MCI don't see a problem with it. Meaning that it's a productivity tool, basically. For instance, for technical translation, it already works better and cheaper than I could otherwise manage. Technical translation is a shitty job. If a machine can do it, that’s better.

KOOZThat tracks; productivity and purpose present a similar dichotomy as imitation and intuition. Rather than dismiss AI-generated imagery out of hand, we can concede that it can begin a process of critically reading, deconstructing or intuiting, as a stepping stone within the course of influence and inspiration.

From the Aristotelian perspective, it’s harder to reward AI-generated text in the same way, as that would seek to elide the process of expression and articulation. In design — partly because there is such a rich heritage of influence, imitation and iteration — it seems acceptable.

MCBut If you can use it intelligently, why shouldn't you? I've read papers where I can tell that they were mostly compiled by a chat bot, because some students don't know how to use the tools. But if someone is smart, they can use the tool for what it can do: to find ideas, to edit writing, to provide data — which most of the time remains completely unreliable.

Every data set is an accumulation of precedent — precedent, mind you, which has been chosen and curated as pertinent to the matter at hand. But precedent is the language we speak. Every word we use means something because it has already been used. Everything we say is to some extent plagiarism, because if a word has not already been validated by use, it would mean nothing.

KOOZManolis, you really don't think you've seen anything as a tutor that has been artificially generated?

MSNo, no, I haven't. I'm telling you for sure that I have not seen anything AI generated — neither in history and theory nor in design. I would have recognised it intuitively, at least I hope I would, ha ha…

KOOZI love you for this answer.

MSBut Mario, you mentioned that an intelligent person could use the AI as a tool, right? For me, this is the biggest question mark. I don't want to be so pessimistic, in a way; an intelligent person will be able to find their way by using any form of technological equipment as a tool. The problem remains for the rest of the people who do not have the critical distance and do not treat it as a tool, but rather as a kind of bulk of knowledge, a bulk intelligence. I think this is a big concern.

KOOZThat's a good point. Stepping out of the architectural profession and its related discourse, such modes of creation, production and decision-making are increasingly shaping the world around us. What does it mean when the systems that are producing the built environment and its politics are based on evolving data sets, with their biases and machine-led irrationality?

MCWhy would I want to borrow someone else's intelligence — never mind if it is artificial — to imitate someone else's work? That seems counter-intuitive. People who want to use the machine that way are morons; morons have always existed, and they will always exist.

What the machine may help us do is to conceptualise the underlying operation. In this case, the way that artificial intelligence works reminds us that everything we invent is derived from precedent. It is extrapolated from a data set which is made of history, tradition or whatever precedent is, whatever there is is out there. Nothing is created out of nothing. Everything is generated out of something that is already out there, which feeds our imagination or whatever you prefer to call it. The machine — with its own technical logic — reminds us that our duty as intelligent human beings is to come to terms with the logic of this process, which is exactly what the machine cannot do. The machine imitates. We should ask ourselves, what does this imitation portend? Are there cases where imitation is something we need? Are there cases where imitation is something we definitely wouldn't need?

KOOZIt seems you're confident, Mario, that before we surrender the organisation of the world to an artificial intelligence, we will gain a sense of criticality in terms of understanding its operations and limitations.

MCWe will not gain it; we have to develop it. In this case, paradoxically, the technical logic of the machine compels and obliges us to deal with these critical categories which have always existed, in which we have forgotten for hundreds of years. It is a certain paradoxical irony that it is a machine that obliges us to come to terms with what we always did.

MSBut you're saying something more here, Mario. You mentioned that for the new generation, the word style does not correspond to the 19th-century art historical term, but rather to another notion, maybe that of software. This spurred another question. It seems that generative AI will actually change the meaning of the words and references themselves. If this is still going on a hundred years from now, the word style would not resemble the art historical term at all.

MCI’ll answer briefly to the idea that technology is changing the meaning of the term style. No, this is not, in my experience, what is happening. The technology is just bringing back a term which had been cancelled by twentieth century modernism. It means the same as it meant hundreds years ago. Style in style transfer software is what Gottfried Semper had in mind. This is the idea of style as invented by Vasari in the sixteenth century. Modernism obliterated that term, and now it's coming back. It's exactly the same concept. And I cannot complain if there is a technology which is proving what I always said: there is something called style. It’s actually vindicating my lifetime position!

MSThat is very clear in your discourse, I would say, overall. But what I'm saying is more about the context. There’s the analogy that today, if you ask a child what a chicken is, they would refer to the image of a drumstick or the plastic wrapped package from the supermarket rather than the animal. You know?

Now, the image that first comes to my mind when I say the word chicken is my great-grandmother's chicken that was in her backyard, in which she would take care of and cook for me herself. So now I’m thinking about the corresponding image of the younger generation. For sure, we all know that the chicken in the plastic package and the one in the backyard are the same thing, and that's what you are arguing. But it doesn't bring forward the same image, or it doesn't bring forward the same context. It's a completely different context. It does alter the meaning.

KOOZBefore we wrap up, I obviously asked AI to generate some questions for each of you; let’s try one each. How might Ventris’ dual identity — as an architect and amateur linguist or de-cipherer — affect our understanding of interdisciplinary practice?

MSI can answer that one with a simple response: it is the paradigm. As Mario used before, this relationship is the quintessence of interdisciplinary practice and not only that, the paradigm of human nature, of what a human actually can understand and transcribe.

KOOZ… as seconded by the filmmaker Werner Herzog, who states that Ventris’ deciphering of Linear B is one of the pinnacles of human achievement.

Mario, your AI-generated question asks about the most surprising or exciting ways in which classical approaches influence contemporary practice. I will embellish by referring to a recent essay on the ten books of Vitruvius which discusses — as in your book Architecture in the Age of Printing (MIT Press, 1998) — the addition of images to text was also an intuitive, transformative, even plagiaristic process.

MCIn a nutshell, having been trained as a classicist, the fact that technology is bringing back some categories of the classical tradition only vindicates their importance. However, let's always be aware that precedent in architecture means tradition and history, and when architects refer to precedent as tradition, there is always a reason to be wary or alarmed — because in the history of architecture, tradition is about the organisation of people, which means nations, places and races.

KOOZOf course. Tradition reinforces structures of power.

MCWe claim architectural history as a sign of identity: these are structures to indicate who we are. It is a fact that in the history of architecture, reference to precedent has always been a tool of discrimination. We build frontier posts as we claim locations. This is what every civilisation has done; this is what architectural history always did.

A dataset is by definition, an exclusionary tool: we put something in order to kick someone out. This is the way every reference to precedent always plays out in architectural history. We cannot avoid that, but we have to be aware. As educators, we have to stake our position. We cannot avoid precedent; we have to find a way to deal with it, because it's out there. This is why we should teach history even when it goes through the mediation of a computational tool; either we deal with it or a machine does. But either way, if we do not know how to position ourselves vis a vis precedents, the idiots or the fascists will do that in our stead.

KOOZI don't have anything to add to that.

MSYou left us speechless, Mario. Shumi, thank you so much for the idea of putting us together.

MCThank you very much.

KOOZIt has been a real pleasure.

*The students’ drawings presented here come from "Athens Erased", an exercise that took place as part of the elective course of the Special Topics in Architectural Space and Communication 7th semester, titled The City: Aesthetic and Political Theories of City Architecture at the School of Architecture at the School of Architecture in the National and Technical University of Athens, taught by Manolis Stavrakakis during 2024-25. Taking on the 19 th Biennale of Venice’s theme of the artificial versus the natural, the students were guided through a series of weekly tasks to erase one part of the city of Athens and substitute it with an uninhabited landscape. The landscape was lived by the students’ imagination. While the erasure of part of Athens came as an intuitive response to their daily experience of the city. The students were asked to create a drawing that rendered this condition of erasure and juxtaposition between the city’s fabric and the landscape.

Bios

Mario Carpo is an architectural historian and critic, currently the Reyner Banham Professor of Architectural History and Theory at the Bartlett, University College London and Gao Feng Visiting Professor at the College of Architecture and Urban Planning of Tongji University, Shanghai. His research and publications focus on the history of early modern architecture and on the theory and criticism of contemporary design and technology.

Manolis Stavrakakis is an Assistant Professor of Architecture at the National Technical University of Athens. He studied architecture at the National Technical University of Athens and the Graduate School of Architecture, Planning and Preservation, Columbia University, and completed his PhD at the Architectural Association in London under the supervision of Mark Cousins and Spyros Papapetros. He has been practising as an architect and teaching in Athens and London.

Shumi Bose is chief editor at KoozArch. She is an educator, curator and editor in the field of architecture and architectural history. Shumi is a Senior Lecturer in architectural history at Central Saint Martins and also teaches at the Royal College of Art, the Architectural Association and the School of Architecture at Syracuse University in London. She has curated widely, including exhibitions at the Venice Biennale of Architecture, the Victoria and Albert Museum and the Royal Institute of British Architects. In 2020 she founded Holdspace, a digital platform for extracurricular discussions in architectural education, and currently serves as trustee for the Architecture Foundation.