KE: Thinking of a project you worked on that used generative or artificial intelligence tools, what was it like compared to more traditional ways of working? What was your motivation for using the AI tool?

CS: Moving from a traditional way of designing to a generative and AI approach is like moving to a meta-design, using this term referred around 1970. It means firstly and foremost designing the design.
This involves how to identify the characters each architect intends to give to their possible architectural outcomes. Following the tradition of the great masters of architecture, design is not reduced to problem-solving but works in the construction of a possible transformation of the past into the future. Each architect uses different logics of transformation, producing uniqueness in his design process.
Designing the design means work in identifying and structuring operationally a multiplicity of past-future transformation logics for each project, in such a way as to make them available to be used at the appropriate time. These transformations are managed inside a topological paradigm able, in progress, to control the generation of the architectural event in all its possible variations. The subjectivity as poetic of the architect is therefore essential in building the generative structure to be used. The recognizability of each output will be strongly characterized by the poetic of the designer.
The motivations for moving toward generative design and AI are essentially two. The first one is to achieve the possibility to manage in progress the quality identity of the project results. Normally every project has to walk through a time in which, from the first sketches to the final result, the architectural quality and the correspondence of the results to the architect's poetics must progressively be increased.
So moving to generative design was, for me, and also for my students, to be able to progressively achieve the quality sought, knowing how to maintain this quality in subsequent projects. In other words, the process of searching for quality and recognizability of each one's poetics should not start from zero for each project, but from the results achieved in the previous experiences. This is fundamental not only for quality by itself, which grows project after project, but also for the time spent to achieve it.
The second motivation is based on a desire that we feel in each project. When developing an idea for the next steps of design choices, we are forced to forget the discarded alternatives that could also be critical to possible outcomes. Managing an idea and keeping all the possible variations works, for me, as one of the strongest motivations of the generative design process.

KE: Can you describe the labour that went into completing the project, in terms of person-hours, skills needed, etc.? Please be as specific as possible.

CS: Building a generative project for architecture is a long work that strongly involves each one's subjective and cultural identity. First of all it is necessary to be able to inquire in architectural design and strongly motivated in identifying possible logics connected to the progressive transformation from an idea to the executive realization of it into a project. This involves being able to operationally explicate the design transformation logics, as a professor of architectural design might try to do in teaching design to his/her students. In addition, since these transformational logics are explicable as algorithms, it is necessary to have the ability to independently write and directly manage algorithms and their control structures. Using pre-packaged tools or delegating to other digital experts the algorithms design greatly lowers the intensity and effectiveness of the generative system you are creating, simplifying the explication of your poetics and identity as an architect, downgrading it into an anonymous level.
My first generative software required many months of work in the mid-1980s. It was a software capable of generating detailed 3D models of Italian medieval towns, variations each one different but all recognizable by their strong identity. Each variation identified the idea precisely through the plurality of outcomes. As in all realizations that involve the definition of a poetic identity, subjective creativity arises from the interpretation of the past. In this case the transformation algorithms were born from my subjective interpretation of the paintings of Giotto and Simone Martini, whose ability is to allude to a dynamic three-dimensionality towards the fourth dimension. This approach I had developed in previous years in my book "The Non-Euclidean Image".
In terms of the relationship between time spent and project realization, surely the first project is not indicative. But the time spent designing transformation algorithms is not time lost because these geometric transformation algorithms can be used also in subsequent projects. They are in fact what, in other AI projects, are the data used for Machine Learning. In this case, they are themselves operational algorithms. Today, Argenia, my generative AI software, has grown a lot. The progressive transformation algorithms are many and different and they can be used in sequence and in parallel way. They were born from my different choices in many different projects and design moments performed in the last forty years following my progressive interpretations of my main masters, from Piero della Francesca to Borromini, from Gaudi to Picasso, and so on.
In 1990 Marco Somalvico, an AI pioneer and great friend of mine, identified my generative software as AI. Over time it has become a complex labyrinth of transformation algorithms in which each new architectural idea, structured as a topological paradigm navigates using transformation algorithms both in parallel and in succession. Navigating this sea of operational data, the paradigm itself grows in complexity to handle the construction of complex 3D events. Each "navigation" through the labyrinth of algorithms is different since it arises at a different time. This implies that each result is unique and unrepeatable but still able to tell poetic visions.

KE: Would you say working with AI involved more labour than a traditional product, about the same, or less? In other words, did AI help you as a creator save time or energy by doing any of the work for you?

CS: Today, the time it takes me to generate a design with my AI software is extremely short: 2-3 minutes to generate an STL file of a 3D model of about a million faces that I can directly print with a 3D printer. And each result reflects my architectural idea in all its facets. This has helped me amplify my creative potential, handling each variation as if it was the final result and telling the story of my idea without having to reduce it only to a single design outcome achieved by loosing all alternatives.

KE: Can you describe any data used in the training set(s) to prepare the AI system?

CS: As I indicated in previous answers, the training set are transformation algorithms that reflect my design and architectural history and are upgraded each time I use the system.

KE: Did your team reach out to get permission to use any underlying works used in the training set?

CS: All my software are original. The first version was written by me in the 1980s, and it does not use existing tools or external data.

KE: Does your team feel any legal risk around the publication or use of AI-generated works?

CS: No, because the AI used is my own original and personal software. It's like having a staff of hundreds of virtual architects (free and very fast) that are my alter-ego designers, knowing perfectly the various facets of my architectural poetics and producing results that in their multiplicity tell the complexity of my design vision. Possible variations among which I can make the final choice that will always be one of the possible expressions of my poetics.

KE: Did clients, users or audience members play any role in shaping the development of the product or helping in any way?

CS: Certainly. Each client or user asks architecture to respond to his/her subjective functional, symbolic and aesthetic needs. The design response might be complex in order to respond adequately even to requests that have not yet formulated. With my software I tried to respond also to the complexity requests for trying to satisfy the multiplicity of subjective needs. The more the requests grow and are complex, the more I have opportunities to increase new transformation algorithms. Each individual variation of results reflects the subjective multiplicity of demands, as unique and unrepeatable as each human being. Considering the double uniqueness of user / single variation, generative design has been of great help in developing the recognizability of the project idea/code and its strong project identity.
This happened also with my students who followed the same process of construction of a subjective system of events generation, both in architecture and in design. They developed their projects experimenting in process their design identity helping them toward their professional growth and they still thank me for it today.

KE: How do you think audiences feel about the output? Are they comfortable with products that had AI input? Are they fooled about the origin of the creative output? Confused? Delighted?

CS: The public certainly doesn't feel fooled. People appreciate the option of choosing the most congenial one between different results. Some architects or designers are sometimes confused because it's not easy to follow this path, mainly because it's not easy to "tell" one's poetics with algorithms. Even my students stay in difficulties, in the first part of the course, in abandoning the deductive logic, more controllable, for embracing the interpretive one. But the results, in the end, comfort them.

KE: What do you think might be some impacts on the industry if more creators start using automation / AI tools to create products?

CS: The immediate impact arises in the industrial components of architecture and in the design objects, making possible the industrial production of objects always different but with a strong identity and recognizability of the designer.
After a century characterized by the author's death, the human creativity finally regains the visibility of its role thanks to AI making the industry able to regain the strength of the product's uniqueness. But we are only at the beginning. It’s a singularity time!

KE: You discuss your creative role when using generative algorithms as selecting from “possible variations among which you can make the final choice.” In your view, how does this “curatorial” labour differ from more traditional architectural design work?

CS: It is not exactly "curatorial." On the occasion of each project, my generative process is highly creative. It works by configuring itself in the software structure and by developing new open variations. In parallel, I design a paradigmatic/topological structure that operates as a control of the project in fieri. This is not a formal idea but a complex structure able to perform relationships and characters. The final choice is a tuning between client and designer.

KE: To what extent can the output of the generative system “surprise” you? Is it predictable in the sense that it reflects your personal poetic vision and quality, or is there any unpredictability in the results, and if so, how does that surprise fit with your embrace of the generative process?

CS: Unpredictability surprises because variation, while belonging to my poetics, structures an event that expands the predictable range. In this sense, it is always a pleasant surprise when the formal representation of the creative idea, originally abstract, becomes more complex.

KE: Thank you very much for sharing your insights.