I answered to the call for papers in an interesting initiative promoted by the PhD students at the faculty of Architecture in Genoa: the theme was something very dear to me — Forms and changes in architectural thinking through tools, paradigms and technologies — and I proposed a little something on the use of Machine Learning in designing spaces that can learn and react based on the feedback received from their end users.
My abstract was accepted and I developed the idea further during a symposium held on July 3rd, 2025 in the beautiful context of the Benvenuto Hall in Genoa. If you’ve never been there, you can’t believe what it is to be able to speak in an old chapel turned classroom.
It was very exciting to see other people’s research and to participate as a field practitioner. Papers are due in August, and a collection of the proceedings will come out subsequently.
Here’s a rough translation of my abstract and, below, you can flick through my slides.
The idea that architectural space might evolve over time to meet the needs of its inhabitants is a desire that has accompanied architectural design since its very beginnings. Already in vernacular architecture, in the mobile devices of nomadic tents, the seasonal transformations of rural dwellings, or the theatrical mechanisms of antiquity, we can glimpse a tension towards a responsive architecture, capable of altering its configuration in response to environment or use. Yet it is in the modern era, with the Industrial Revolution and the emergence of technology as a design language, that this dream begins to take more conscious shape.
In the second half of the twentieth century, this perspective intensified. Projects such as Peter Cook’s Plug-in City (1964) and Cedric Price’s Generator (1976), developed with Julia and John Frazer, conceive modular and reconfigurable environments governed by computational logics. At the same time, Christopher Alexander’s pattern language describes architecture as a living and dialogic system, while Richard Saul Wurman and Nicholas Negroponte propose spatial models as programmable, information-rich platforms, paving the way for interaction between architecture and intelligent systems.
These visions anticipate scenarios made tangible today by machine learning, which enables the development of environments capable of collecting and interpreting real-time data and autonomously responding to changing conditions. Space thus becomes behaviour: not merely form or structure, but a system that can learn, adapt, and act, projecting design into an evolutionary dimension driven by feedback loops and continuous transformation.
In recent years, the notion of the built environment has undergone profound change, embracing the possibility of behaving in adaptive, sensitive and contextual ways, including through experimental scenarios already in operation, from Philip Beesley (2010) to The Living (2014). Research and experimentation, such as that of the MIT Responsive Environments Group, confirm the emergence of a design grammar in which architectural space becomes an active and co-evolutionary system.
The analysis of field-tested design experiences, coupled with the current potential of artificial intelligence, allows us to outline an operational taxonomy of architectural adaptivity strategies. This classification, rooted in the intersection of practice and technological innovation, distinguishes three main approaches: reactive, predictive and generative. Reactive strategies follow input–output logic, where space responds to direct stimuli with predefined actions, like lights or ventilation activating when a person passes by. Predictive strategies employ machine learning models to identify usage patterns and anticipate needs, adapting environmental conditions to users’ habits. Generative strategies deploy more complex systems — such as neural networks and evolutionary algorithms — to produce spatial configurations in real time, turning space into a co-agent of the design process.
The introduction of artificial intelligence into design ultimately raises new questions around the identity of the designer. The architect is no longer solely the author of forms, but a curator of behaviours and relationships. They become a designer of frameworks: establishing rules, defining conditions for evolution, selecting datasets, and training models. In this sense, the act of design becomes metadesign, a design of the design process itself, operating upstream of generative mechanisms.
This figure recalls the thinking of Cedric Price: “Technology is the answer, but what was the question?” Progress, on its own, is not enough. What is needed is a new ethic and a new aesthetic of design responsibility, in which the very notion of “functioning” becomes an architectural value, an integral part of spatial imagination.
Machine learning on one side and deep learning on the other thus open the way to new conceptions of design as a continuous and performative process. The genesis of form no longer coincides with a crystallised initial moment; rather, it unfolds over time, adapts, and is renegotiated. The architect’s critical autonomy lies in defining behaviours, constraints and evolutionary potentials, not in sealing the formal outcome.
This approach proposes a departure from the dichotomy between space and interaction, form and function, author and user. Architecture becomes a living, dialogic system, in constant conversation with its surroundings, bodies, and data. A new generation of spaces that learn: not merely containers of life, but learners from life.
No Comments