Articulating Architecture // T.1-2-3 APPLIED THEORY

The project ‘Articulating Architecture’ investigates the potential of recent developments within the field of artificial intelligence to become an agent for complex topologies. The fast rising architectural developments of today show a lack of interest in expression of surfaces, patterns and intricacy. The use of artificial intelligence and machine learning could be trained with an input of cultural and historical data of publicly acclaimed aesthetics of complex architectural examples.


Based on the presence of a vast architectural heritage within Barcelona, we were interested in the enhancement of intricate surfaces and the value it can bring onto a buildings facades.

Key Research Question:
Can the use of AI in the search for complex topologies be the birth of a new aesthetic language?

Video – Click on image to play


Contemporary architecture undervalues the implementation of complex topology as a tool for cultural expression and urban identity. Throughout optimization of processes and the run on real estate for maximum profit rather than individualization, expression, wellbeing, culture, and intellect.

“ INFORMATION RICHNESS. The diversity and variability of the natural world is so pronounced, it has been described as the most information-rich environment people
will ever encounter. Whether natural or built, people tend to respond positively to information-rich and diverse environments […]’’
Stephen R. Kellert et al.

In the following, we are using the term ‘Complex Topology’ as an overall term for intricate surface expression. We are asking the question if AI can become a tool to inform surfaces with more expression. Whether we are about to enter a new age of applied tat or of thoughtfully integrated, intellectually substantial articulated architecture.

Figure 1: Aesthetics through Complex Topologies | Source: Mirna Pavlovic


As complexity surrounds us in natural ways, our buildings tend to portray the opposite. Left behind by Fordism, what has resulted has influenced our designs to be banal and our construction methods to be inefficient.  Contemporary architectural discourse lacks the same level of emphasis and investment in pursuing aesthetics for aesthetics’ sake relative to periods in the past evident in the complex topology of a baroque ballroom’s highly detailed and ornamental wall treatment. This type of expression, is made possible by huge social disparities at the time in the form of indentured, or forced labor, tremendously cheap material cost, and the cultural interest in communicating the ephemeral through physical constructs.  With new data driven methods and tools – such as 3D scanning, additive manufacturing, and the pantheon of different artificial intelligence technologies- design and production processes are becoming increasingly more integrated and available to the broader construction industry while making more diverse and customized design elements readily available at little or no additional cost.

With so much of the process of identifying design problems and then solutions we, as designers, are confronted with the existential questions of when or whether we have given too much control to the systems we create? Whether our systems are a form of genuine creativity in themselves and what the implications are for creatives? How is good design, creativity and culture defined in the age of AI?

Figure 2: Floorplan – Transfer learning of Baroque Topologies| Source: Stanislas Chaillou


Hallucinating Culture, a project developed by Hendrik Benz, Alberto Brown-Cruz, and Michael DiCarlo at IaaC, evolved from the interest in the latest developments of Artificial Intelligence for Art, Design, and Architecture. The project is a brief exploration and introduction into GAN’s as a novel tool for designers and maybe further a source for inspiration. The specific intention of the project here is you uncover or discover ways in which the same sort of tools can provide creative guidance or unexpected results. The researchers were looking for the unexpected. The method of this research is to collect data of existing architectural elements within the city of Barcelona and to feed this into a Generative Adversarial Network to train an AI with our own Dataset with the aim to come up with alternative articulations.

Figure 3: Final Output Columns – Hallucinating Culture | Source: Hallucinating Culture

The truncation value has to be set into the script as a number between around -1.0 and +1.0. If a value close to 0.0 is placed for the truncation, the output images will be more likely to look realistic (1), in our case close to the input images of the real Barcelona columns.

Figure 4: Overview Columns and Artifacts – Hallucinating Culture | Source: Hallucinating Culture

If a number closer to +/- 1.0 is placed, the outcome becomes more ‘interesting’, but probably many of the generated images will tend to look unrealistic.

Although there are more artifacts and unrealistic-looking results, many of the images are more interesting for their artistic possibilities and unusual combination of influences.
At times these more diverse images achieve a nostalgic, dreamlike, and painterly quality that I find very interesting.”
Derek Philip Au

A higher value then +/- 1.0 is possible to be placed as a value when generating images, but will likely result in the generation of weird artefacts and colour effects in the output images.


We conducted a series of discussions in regards to exploring more in depth the role of AI in art, architecture, and design in the context of both educational and professional environments.

Figure 5: Evolutionary stage of artificial intelligence | Source: Authors


Aldo Sollazzo

Aldo Sollazzo is Director of the Master in Robotics and Advanced Construction at Iaac and Director of his interdisciplinary firm Noumena here in Barcelona.

Interviewer: Beauty is difficult to define, and shifts across demographics. We can use technologies like artificial intelligence and machine learning systems to develop a new class of tool to augment the creative and aesthetic decision making processes in an effort to democratise the design landscape and produce a more discriminating collective body of work. What are your thoughts?

Aldo Sollazzo: I believe that these new tools, of course, are opening up a lot of opportunities on how we can develop a new aesthetic language, and I believe it’s the connection of multiple processes that are happening nowadays. And we are still not fully aware. And that’s why I believe it’s really important, and I am having this conversation with multiple voices now, because we are still trying to understand how to implement these instruments properly and what the output is, is not evident to anyone yet. So, the interesting part is that there is a shift towards a new aesthetic, but I believe there is a shift towards the new decision making process that can trigger these aesthetics. So from one side machines, and their own intelligence, and from the other side, all these computational processes of informing design decisions as an input are coming from the physical world, from simulation, from material properties. And finally, what we are also exploring in the master program, a fabrication protocol that can bring back these explorations to the real world.
So just to sum up, I believe it’s not this effort towards the definition of a new aesthetic, it is rather how we can make decisions together with a machine. Not to use a machine as an instrument, but to rely on the intelligence of a machine in a parallel moment, where decisions are taken at the same level together with them. What we are doing is that our design constraints are now defined in a different model or a different aspect of the design. Which probably is not by defining the form but is by defining the process behind it. By defining the farm of creativity with the machine. Through this process, the machine is learning a series of inputs that we are introducing. So I guess there is our part of design decisions. And we really need to rely on the creativity of the machine or our understanding how these forms can generate associations one with the other.

Interviewer: Where do you see problems with the implementation of AI in a design?

Aldo Sollazzo: So I believe it’s inevitable that machines will start to make decisions for us, or with us. What I believe, it’s always important that we are aware of how we define the metrics to evaluate the outcome, otherwise, it becomes a black box for us, where you might not fully understand the result. I think those are decisions that should inform us in making more sustainable designs, develop higher aesthetics that could rely on material properties or whatever the values are, that we are seeking to generate.

Interviewer: Can the use of AI become the birth of a new aesthetic language?

Aldo Sollazzo: I believe there is a moment where we are defining a new aesthetic, this aesthetic is the result of many different variables that we are exploring as decision makers. I believe that we are all contributing individually right now and as a collective and it is a moment of understanding. These technologies and the increase of availability of neural networks can not extend more than 7-8 years ago. The publication of new papers on these topics is increasing and increasing.

Interviewer: Where do you currently have a form of artificial intelligence or machine learning applied in your projects, for example in your firm Noumena?

Aldo Sollazzo: At Noumena we are implementing the use of computer vision to understand spatial dynamics. We are using these tools to underline the association and correlation between dynamic and static components. For example recognising images, images represent 80% of all unstructured data. The majority of the data we produce is unstructured. It cannot be defined clearly in a set of categories, it can be hard to label the content of this data. Machine learning and neural networks are good in understanding what an image represents. They are very good at structuring these huge sets of data that we are producing, which is not meaningful information at first. We work in micro farming, and analytics, retail store distribution and also of course, in advanced construction. We are using the machine as a tool where it gives us a benefit opposed to human capabilities, especially when it comes to data-processing and then to act in collaboration with the human to select and make decisions.

Interviewer: Part of our thesis is aiming to implement AI into the creative decision making process. Some voices see the use of AI critically. How do you feel as an architect and designer about giving up control over the design to an AI?

Aldo Sollazzo: No, I mean, I’m curious. I like to discover what is possible. I guess it is a little bit a representation of my personal commitment to learn and visualize what I wasn’t recognising before. Also the work of Noumena is connected to the philosophy of Emanuel Kant and Platon and the concept behind it is the ability to “see beyond” the object in front of you and start extrapolating and understanding its meaning. To discover what the machine can see, what the machine can interpret from a set of instruments which you provide to the machine, with you as the decision maker. Because that’s your task now. It becomes a beautiful way to orchestrate and distribute the creative process between humans and machines. And I don’t feel left aside, I feel I’m part of this process. I still feel part of this process, we can really start to appreciate what the machine can produce and generate.

Interviewer: In the beginning while looking into the topic of ai for architecture and the interest in the process of pattern recognition for complex architectural elements we asked ourselves if this use of ai could lead to a new aesthetical language once fully implemented into the design processes.

Aldo Sollazzo: If you look at Gottfried Semper, he’s talking about the primitive hut. Architecture more or less started with the first definition of a hut. And in this primitive hut there were already all the structural components that we still use to build today. And the interesting part is that also the topic of ornament has always been a little bit historically connected to the morphology of a hut, the architrave, the column, the idea of a growing tree, that’s why we have got the capital. The interesting part is that ornament in any case, represents a message that represents a memory and represents and also communicates value. But the interesting part, I guess, is to explore which kind of message a machine produced ornament could deliver.


Zeynep Aksöz

Zeynep Aksöz is an architect and computational designer. She is Research Associate and lecturer at University of Applied Arts Vienna, Assistant Professor at TU Vienna. The following was our discussion revolving around creativity with AI.

Interviewer: Our first experimentation with AI and neural nets were the columns we generated for ‘Hallucinating Cultures’. The aim was to use these tools, novel for us, to start the process of imagining new designs based on the cultural heritage of Barcelona. Where does AI stand within design? Can AI actually be creative?

Zeynep Aksöz: Can I be creative? Like an alternative intelligence is it radically changing the system or not right? I think the question is not the right one to ask, as the question is trying to make this tool more human rather to approach it as, what it is. I see AI as a tool. I don’t think that creativity can be accredited to it. There are so many theories of creativity. Creativity in itself is an unknown environment. So you cannot claim AI to be creative that simply.

It can revolutionise the industry, it can revolutionise how we think about designing, but it is also not super novel. AI was here in the 60s, AI was here in the 90s, the construction engineers are continuously using AI and continuously using neural nets. So just because we architects have discovered it, doesn’t mean that it is novel. Machine learning is the new processing, like 10 years ago, if you would use agent based systems and processing, it was everywhere, everybody was using it and it was kind of bringing a new aesthetic language.

Maybe even more interesting, if you look into the roles; what is the role of the AI in the process? And what is the role of the designer? Or what is the role of the architect?

Interviewer: Creativity is with the designer or the architect and AI is a tool?

Zeynep Aksöz: Creativity is where you bring something novel into a monotonous environment. Creativity can be described as something which is not monotonous. But if you look into the behaviours, who is monotonous and who is not monotonous, then you will see, the AI is always doing the same thing, it has the same behaviour. If you are using StyleGan if you’re using any machine learning whatsoever.

Except reinforcement learning, because then you can achieve some creative behaviour by kind of rewarding the neural net. So you can achieve some novel behaviours that can maybe be explained as ‘creative’. But as long as you’re using basic translation algorithms, the creativity remains with the curator, but not with the system.

Interviewer: How do you see the use of AI and its increasing popularity? Where do you see benefits in its use? Where do you see limitations or implications?

Zeynep Aksöz: I think everybody’s expecting that AI has a very alien aesthetic, right? ‘This is super alien. This cannot be thought about from a human’. But I don’t think it should be like that. This is purely a tool. The aesthetics are defined by ourselves. I believe in the next industry 5.0 the aesthetics are not going to be defined through AI, but they will be more defined through energy. This huge question of climate change. This is actually the aesthetic language that we should be looking for. That’s going to be brought by us by different sensing tools, different energy collection tools, and therefore aesthetics will be driven by environmental performance of the building. A novel aesthetic language, not the language AI has generated, but maybe optimised by it.



Edouard Cabay

Edouard Cabay is an architect, designer, and educator. Edouard Cabay graduated from the Architectural Association in 2005 to work for Foreign Office Architects in London, Anorak in Brussels and finally for Cloud 9 in Barcelona, where he occupied the position of head-office. Currently he teaches at the Diploma School of the Architectural Association as a unity master for Diploma 18, and at IAAC as OTF Co-Director, MAA Senior Faculty Member, and MaCT Studio Faculty Member. The following is a short discussion we had regarding his teachings, complexity in designs, and control.

Interviewer: While teaching in OTF, the projects that have come out of it usually consist of curvy, complex surfaces and there seems to be no sign of flat walls. The curvature does have structural qualities and cooling characteristics, however could you build more on that?

Edouard Cabay: I’m not a supporter of complexity. I don’t believe that we should introduce complexity for the sake of itself. Complexity is an incredibly beautiful domain in the creative world. Complexity is not new, but we have seen a recent development of a lot of tools or fabrication processes that were not available before, mainly through computation. That enables both the design and to build things that before were very difficult to make. What’s so beautiful is that there is a relationship to nature, because nature is complex, which is something that definitely has a clear relationship to the computational world. I think it’s important to understand what complexity is good for. I like it as a creative field, it’s bringing things that we didn’t even know were possible, it brings something that goes beyond the human imagination and the human faculties. I think that this is something that is very interesting, because it’s changed, to design something that you cannot imagine yourself that I think is something that artists and many creators have always been after.
Why did I start with this kind of slight reaction towards this word complexity? The reason is that I think that some people feel that it’s the answer for everything. In other words, then forget the richness of our culture and of our past. That I think is a mistake. I think that surface can be as beautiful as a curved surface, but it depends on how you evaluate it depends on the context, it depends on many factors.
Regarding OTF, there is something very beautiful in 3D Printing, which is that we work with materials that are liquid and then they become solid. As a consequence, the way that the material dries up, it retracts and it changes shape. If you get a straight line to retract, that is where something interesting happens, because you cannot get a straight line to retract because it cracks. But if you get a non-straight line, it will retract but it’s going to lose length and therefore change in shape. This 3D printing project is about understanding the attributes of a curve. We make straight walls out of curves, and then it takes a different shape. Alex and I share these thoughts of the performance, of the structure, of the climate, some functionality that reinforces the aesthetic as a consequence.

Matias del Campo

Matias del Campo is an architect, designer, and educator. He is associate professor at the Taubman College of Architecture and Urban Planning. He conducts research on advanced design methods in architecture, primarily through the application of Artificial Intelligence techniques in collaboration with Michigan Robotics and the Computer Science department. The following is a discussion we had with him regarding his work and thoughts on AI in relation to design.

Interviewer: In one of your lectures, you mentioned an experiment of the AI research lab of Facebook, in which two chatbots were programmed to talk with each other. After a while the two chatbots were creating their own language. Do you think AI can be creative, and if so how could this be implemented in architecture?

Matias del Campo: That is still an ongoing discussion. I don’t think it’s solved yet. And it is really interesting how much it changed in the last couple of years. That Facebook experiment was in 2017, it’s not that long ago. I heard about it and immediately thought; ‘Okay, we are there, AI’s are creative ,they made their own language, problem solved.’ The more I think about it, the less I’m convinced that that was already a creative moment, for a couple of reasons. Number one; We don’t know what they’re talking about. Right? Because it’s their own language. So we don’t know whether it makes any sense. Is it really better or faster to discuss it this way? Or is it just gabble? This is just garbage? We assume that they are discussing economic issues, but we don’t know. Then the other one is, what if it’s just a bug in the code? Or maybe creativity is a bug, who knows? There’s several issues that especially this experiment, is not really the best example for creativity. The better example, in my opinion, is AlphaGo. In the game of AlphaGo and Lee Sedol, where AlphaGo made the famous in which the commentators started saying; ‘Well, that is stupid. I mean, that doesn’t make any sense. No human would do that. Maybe the programme has a bug or maybe it’s not working correctly.’ Lee Sedol was looking at that movement closely and then you start seeing his face, change to pure horror. 10 minutes after that, the commentator said; ‘Wait a second. That’s actually a creative move.’ And that was actually an interesting sentence because he literally used the word creative. Without overthinking it, it was almost a natural response from him, as if he would be reacting to a human doing that move. So this is still a discussion, whether this is really a creative move, or whether the AI is so well trained, that it just calculated that if he makes that move, his chances of winning are higher. In that case it would be a purely pragmatic decision and would not have anything to do with creativity.

Because the biggest problem is that we, as humans, tend to interpret results that come out from neural networks as creative. Occasionally, I do it myself, like; ‘Oh, that’s interesting, I can use that. That is something new, we can use it in our project.’ This doesn’t mean that the AI is creative, or the neural network is creative, it means that I am able to creatively interpret it.
So there’s always this problem that like, what is really creativity? First, we have to answer that question, what do you define as creativity? And then you can say, yes, this is creative, or it’s not creative. And I have the feeling that a lot of things that are happening now in the arts and music, also in architecture, in terms of grappling with the concept of creativity and neural networks, most of the time it’s only the ability of us to interpret something as interesting.
That’s something that neural networks cannot do yet. For example, what we as architects do very often is we take a happy accident like; ‘oh, wow, that’s cool. I can use that, I can change it this way, and then use it as part of my project.’ Right? That’s a creative thing to do. But AI cannot do that yet. Recognizing a mistake is an advantage. Although Sandra, my partner, already said that that can be solved too. We just need to train a neural network to understand the mistake as an advantage. You have to feed it like a couple of 100,000 examples of where a human accidentally found something interesting and applied it in a positive way. It has happened so many times in science, too. We haven’t finished that conversation yet.

Interviewer: Interesting thoughts. How do you see the question of authorship when handling these vast amounts of

Matias del Campo: These are all questions we’re in the middle of. First of all, there is the person who coded the neural network itself. Who did the algorithm? The way he sets up the algorithm will also influence what the results are going to be. So it’s really important to understand that that’s part of the authorship. Then there is the user, the person that is applying that neural network to any sort of machine learning process. By selecting a specific neural network, you have also a part of the authorship, but you don’t have the full authorship. Look at the example of this famous painting Portrait of Edmond Belamy, which was sold as the first AI portrait at Christie’s for about half a million dollars a couple of years ago. And the artists basically said; ‘Yes, it was done by an AI.’ Well, of course, they selected the database of paintings, they chose the images for the database, meaning they introduce the bias towards Western paintings and portraits from the 15th to the 19th century. There are very specific decisions here, made by the artists to create a result. They’re pushing the network to create a specific result.

So, the authorship is not fully with AI either. I think it’s a shared authorship. It’s shared between the end user, the programmer. And of course, if you have a database of 10,000-15,000 paintings, then probably part of the authorship is also with the painters who made the original paintings the network’s using. The aspect of authorship is probably going to change a lot in the 21st century. It’s going to go away from the 19th century idea of this old genius who’s the author of a very specific book. I can see it coming, that for example, authors are more and more relying on automated processes that are already available online today. Text services like Grammarly use machine learning already to understand and support writers. Does it mean I now have 100% authorship over that line of text? Probably not. The whole aspect and the idea of authorship is shifting in the 21st century and I think it’s time. All the things are like 19th century inventions, because the way our data is supporting our creative work is totally changing.

Interviewer: Right now we are still talking about the use of artificial neural networks for 2-dimensional generation of images or text. How are you using these tools in your practice or teaching in your course to create 3-dimensional architecture?

Matias del Campo: We have been actually working for over two years now in terms of going into 3D with neural networks. The algorithms are getting better, however we have found specific problems within the data sets for these neural networks. More specifically in the process of creating a data set.

We started totally creative, how can we use the neural networks creatively? By which you get to the point where you’re like; ‘This is as far as we can get with images. How do we get into 3d?’ Then you start with 2d to 3d work? But we’re in a discipline, where you are inherently 3d. So we need to work completely in 3d. And that’s what we’ve been working on already for the last two years. We’ve already published papers on entire 3d workflows, with neural networks, using 3d graph convolutional neural networks, for example. And I had a conversation recently with Patrick Schumacher, who said that the whole work we’re doing right now reminds him so much of what they were doing 20 years ago with Maya, trying to figure out how to work with it in architecture.

Interviewer: Do you see parallels between the upcoming of Parametricism and the recently booming interest and research in the use of AI for architecture?

Matias del Campo: I see the difference rather than there is a parallel, and that this work on AI is not a style. Patrick Schumacher definitely proposed Parametricism to be a style. And the reason why I don’t think that, what we’re working on now is a style. Because neural networks, artificial neural networks, and AI, have already become a big part of our everyday life. It’s not like a specific design tool that keeps you creating pretty curvy projects. That you can define as a style, because they have specific features which can be defining a style. AI by itself doesn’t do just one shape, it doesn’t do just one language. There’s a variety of different languages. It all depends on the data you feed it and will give you different results.
It already exists in most of politics, economy, science, art, painting, music, etc. Some of those cases are subtle and some are very visible. Parametricisim has not had that much of an impact in our lives in that sense, yet. So I think it’s not a style, it’s really something that is changing society at large. And that we as architects definitely should understand. Because it’s going to change our discipline a lot. And what I always tell people is, if we don’t pick it up and figure it out, other people will do it for us. And this is kind of like the worst developer architecture you can imagine, when talking about beauty and cultural value in architecture.

Because what’s going to happen, and it’s already happening now, is that automated planning is gonna come. I see that a lot of people are going to use it. Because it’s going to be convenient, it’s going to be precise, it’s going to give you very reasonable planning solutions. But I think it might be dividing the architecture disciplines into everyday practitioners using AI just to crunch out very quickly buildings. You will have a group of people who are interested in sort of like, what does it mean for the discipline? How does it contribute to us as a culture? And then there’s going to be a part who’s going to stay with using manual tools. And these are going to be luxury brands, going to design your house by hand. I’m looking forward to that; luxury brands are going to design your house with a pen and a piece of paper. But it’s going to cost you a lot.

Interviewer: In your paper ‘A Question of Style’ you are describing the connection between AI and the use of the term ‘Style’. Can you briefly explain the connection?

Matias del Campo: The paper starts with this conversation between Muthesius, who is this german architect from around 1900, who basically said that architecture has to get rid of the term style entirely. Because they wanted to differentiate themselves from the 19th century who were obsessed with style. The whole 19th century was trying to imitate historical styles in architecture and created the NeoGothic and NeoClassical, and NeoBaroque. And the 20th century tried to differentiate themselves by getting rid of the term ‘style’. And Muthesius was actually proposed to use the term ‘type’. And when he talks about type, he talks about primitive geometric figures, like cubes, spheres, pyramids, and compositions of those. I mean this already sounds like a postmodern nightmare to me. But it’s, nonetheless, it’s something that they set up very early on to do. And then, in the paper, I try to interrogate why the scientists who came up with style transfer, to use the term style. And the reason is a very banal one. They’re not architects. They’re not art historians. They were just computer scientists and for them, the term style is very simple. And there’s painters who have specific styles like van Gogh. They love van Gogh, because it works well with style transfer. So for them style was really useful. On a very superficial level. It was not like the architectural discussions where we really try to understand that style might have more implications than just being a decoration of a building, or a decorative style.

But that there’s political implications, that there’s economic implications, that there’s this whole discussion of Adolf Loos, basically saying that ornament is bad, because people have to carve them away their entire life and wasting their life carving ornaments instead of doing something better. And today, all of those things don’t apply anymore. Because if you want to do an ornamental piece, you’re not using the lifetime of a person enslaved to create this ornament by hand, you use a robot. And that’s it.


As Matias has pointed out the different implications that can be brought on by artificial intelligence, there are many questions at hand regarding creativity , authorship, and style. Authorship has been circulating in the discipline for a long time, as professors and other critics continue to discusss who truely is the author in design projects. Is it the intern? Is it the owner of the firm? As we continue to have these discussions, upon including AI to the design process, things become even more complex to deduce. Its interesting to even think about how an artificial intelligence could become the author of a design, or at least have partial rights. With creativity, how can we measure whether an AI has it or not? As mentioned in the previous interviews, we think that something is creative because we are interpreting it in a creative way. Since a designers/ architects involvement is still relative when working with current machine learning processes, the interpretation of the designer/ architect is what defines the creativity of the machine and to what extent the machine learning process could be taken within the current stage of AI that we are in.

Figure 6: Question of the shared Authorship | Source: Authors


Case Studies

To recognize the emergence of a new style, aesthetic language or even a single element as novel one must acknowledge and understand the system of creativity driving it and how “creativity” itself is defined.
Culturally, “creativity” is generally understood in one of two modes, as either coming from some sort of ethereal or divine inspiration, planted into the chosen mind of an artist or scientist, the other understands creativity is often the result of “happy accidents” or, serendipity, having the ability to recognize a mistake or unexpected outcome as a viable or even preferred solution to the given problem or even more recognising that a proposed solution for a problem is perhaps more suited for a different problem entirely, one such example being a distinctly human or anti-robotic trait, directly opposing rules of automation, scripting and the sort of linear logic typically associated with parametric modeling.


This Vessel does not exist

The project “This vessel does not exist”, by software developer turned ceramicist Derek Philip Au, uses a styleGAN or generative adversarial network, created by Phillip Wang and Nvidia, to explore the potential of this ready made AI package to generate “imagined” examples of novel ceramic vases. The output of Au’s system runs the gamut of expected forms, glaze colors and textures, as well as approximations of figural patterns, all of which come from the robust data set from which the Ai is learning, some 150,000+ images in total. Beyond simply reinventing images vessels which could readily be identified and perceived as such, it is possible to tease out using different techniques “less obvious” results, representation of objects not necessarily possible or recognizable. It is in these moments when the system, designed, ostensibly, to produce one thing gives us something we didn’t ask for that we take pause. These outputs tended to be more “interesting”, if we remove from the task the investment in the output, and define interesting as something that evokes an unexpected emotional response and cause for pause.

Figure 7: This Vessel Does Not Exist | Source: Derek Philip Au

Plato’s Columns

Matias del Campo’s project, “Plato’s Columns”, is about “Celebrating the Glitch” as he says. The “glitch” being the result of specific predefined parameters like extrusion temperature and speed in a robotic 3d printing process that are given to a degree of variable “randomness” which result in a degree of emergence and variation in the final part. Del Campo is exploring the relationship between uncertainty and creativity. The difference here from the previous examples is the level of technology involved in constructing the space for divergence, all the “intelligence” is left to the physical properties of a material and the whims of the operator controlling with parameters to randomize as opposed to designing that can intuit desirable outcomes not hardcoded.

Figure 8: Plato’s Columns | Source Matias del Campo et al.

Vientos de Alisisos

Edouard Cabay’s series of semi autonomous robotic installations “Machinic protocolls” is a more playful example of the opportunity do have a genuine dialogue with technological systems and machines. In particular, the piece titled “Vientos de Alisisos” in which a pen attached to a sail is blown about a space circled by fan with proximity detectors. The system is defined, the parameters static and without invented glitches, the result and it’s variability is rather analogue, relying largely on the way air is being pushed about.

Figure 9: Machinic Protocols | Source: Edouard Cabay et al.



In an experimental approach we wanted to investigate peoples preferences for complex topologies and asked for their tendencies with the use of Instagram polls.

Figure 10: Instagram – Hallucinating Culture | Source: Authors

Question of Style

When asked, we find that a consistent majority of respondents to the polls we conducted choose intricate detailing, complexity in form, and images or objects that could be classified as unexpected, novel, or grotesque. Our polls, which were directed toward  designers and civilians alike, first asked the respondents to choose between images of similar building elements, side by side, one from a more recent era in design with hard edges and smooth surfaces whereas the others came from more ornate or opulent eras of architecture. The questions asked were structured only to understand what the viewer “prefers” or finds more “interesting”, devoid of any association with performance, costs or context, thus giving insight into the way humans naturally interpret and process form. Given the millions of years of human evolution that give rise to the robust systems of pattern recognition and deep neurological associations with certain forms found in nature, it should come as no surprise that our poll recipients have a more emotional response to architecture reflecting the natural world more than objects in direct opposition.

Figure 11: Graph showing preference of design based on Instagram polls


AI generated Columns

In our subsequent polls we asked respondents to choose their preferred images of either Ai generated vases from the “this vessel does not exist” project or columns from the “hallucinating culture” project. In these polls the choices were both generated by the same GAN system with the difference being the “truncation value”, this value acting as a sort of chaos tuning knob, a lower value will yield a result closer to a replica of an input whereas the higher the value the further your output wades into the murky waters of the unexpected.

Figure 12: Graphs showing preference of AI output column designs based on Instagram polls

AI generated Vessels

The images used truncation values of .6 or 1.2, and all came from the same system of inputs, training protocols and controls, yet still there was a distinct tendency to choose the options with a higher truncation value, that is the option with a higher demonstrable metric of “chaos”, even in the cases where the difference was not so obvious to the naked eye, implying the notion that there is an intrinsic element to humans’ aesthetic valuation system that responds positively to even the most subtle complexities.

Figure 13: Graphs showing preference of AI output vessel designs based on Instagram polls


The forms and patterns found in nature, at a glance may seem chaotic and random, are in truth the result of a delicate balance of competing forces, a choreography of parameters ebbing and flowing in an infinite ballet. This competition and dance of inputs and outputs behaves like the systems a designer may construct, whether a fully automated Ai or parametric game of analog and digital inputs, without specific intention but the capacity to “make decisions” the way they are in nature; collectively, and emergent.

Paper – Click on image to view PDF


Credits (click on text to follow link):

Video: Articulating Architecture – Teaser: 

Figure 1: Inquie?tum | Source: Mirna Pavlovic (Photographer)
Figure 2: Floorplan – Transfer learning of Baroque Topologies| Source: Stanislas Chaillou 
Figure 3: Final Output Columns – Hallucinating Culture | Source: Authors
Figure 4: Overview Columns and Artifacts – Hallucinating Culture | Source: Authors
Figure 5: Evolutionary stage of artificial intelligence | Source: Authors
Figure 6: Question of the shared Authorship | Source: Authors
Figure 7: This Vessel Does Not Exist | Source: Derek Philip Au
Figure 8: Plato’s Columns | Source Matias del Campo et al.
Figure 9: Machinic Protocols | Source: Edouard Cabay et al.
Figure 10: Instagram – Hallucinating Culture | Source: Authors
Figure 10: Graph showing preference of design based on Instagram polls | Source: Authors

Figure 11: Graphs showing preference of AI output column designs based on Instagram polls | Source: Authors
Figure 12: Graphs showing preference of AI output vessel designs based on Instagram polls | Source: Authors



. Salingaros, N. A. 2003. “The Sensory Value of Ornament“ Communication and Cognition.
. Salingaros, N. A., and K. G. Masden, 2006, “Architecture: Biological Form and Artificial Intelligence”
. Kellert ed. al,. 2008. “Biophilic Design: The Theory, Science, and Practice of Bringing Buildings to Life”,
. Roberto Naboni, Ingrid Paoletti, 2014, “Advanced Customization in Architectural Design and Construction”,Springer
. Edward Relph, 2008, “Place and Placelessness”
. Poorang A.E. Piroozfar, Frank T. Piller, 2013, Routledge, “Mass Customisation and Personalisation in Architecture and Construction”
. Tom Verebes, 2015, “Mass-Customised Cities”, John Wiley & Sons
. Wes McGee, Monica Ponce de Leon Mar, 2014, “Robotic Fabrication in Architecture, Art and Design”, Springer Science & Business Media
. Branko Kolarevic, José Pinto Duarte, 2018, “Mass Customization and Design Democratization” , Routledge


Articulating Architecture // T.1-2-3  is a project of IAAC, Institute for Advanced Architecture of Catalonia developed at the Master in Robotics and Advanced Construction in 2021 by:

Students: Hendrik Benz, Michael DiCarlo, Aslinur Taskin

Faculty: Mathilde Marengo