Artificial Intelligence is not a magic bullet for Architectural Design

Artificial Intelligence is omnipresent in contemporary media, scientific research and especially social media, where it’s easy to find extreme voices declaring how AI is going to change our life and work, disrupting how we did things for centuries for the better, or even destroying human life. Architecture is not alien to the discourse, and even if the interest of architects in Artificial Intelligence is not new, it is possible to observe increased attention towards Machine Learning and all the subfamilies of approaches we call Artificial Intelligence.

I have often read about how artificial intelligence will revolutionise architecture, even taking the place of architects. These opinions, extreme and sometimes enriched with commercial meanings, do not consider the complexity in applying these methods to our profession, a field by itself based on complexity.

To get a professional view on the topic, we interviewed ten digital champions of architecture, asking them to describe their experience and share feedback on the use of advanced machine learning methods inside our industry. Our research presented in the paper The role of Artificial Intelligence in architectural design: a conversation with designers and researchers is summarised in this article.

Artificial Intelligence in Architecture

While is it true that all those technologies we call AI are currently changing our society through their extensive use in IT and scientific research, the hype on Machine Learning methods makes it difficult to distinguish effective contributions from suggestions. As many people know, the field of Artificial Intelligence was born in the fifties, enriching itself over 60 years with many methods and approaches, from the simplest Rules-Based Systems to Machine Learning, Shallow and Deep Neural Networks, Natural Language Processing, Speech Processing, Image Recognition and many others. There are two primary reasons behind the contemporary bloom of Machine Learning. This is mainly because of fast graphics processing units (GPUs) which have significantly increased our computational capability and made now it possible to apply such novel Machine Learning techniques to an extensive set of problems, and of course, because of the proliferation of Data.

Data is the new oil, they say, but oil is stinky, dark, muddy, and it takes a complex system of knowledge, technology, and standardisation to extract and transform it, making fuel out of it.

With the coming of Big Data, a quantitative change in the data we are now able to collect, we have understood that to extract knowledge out of these messy collections of information, we needed to exploit automation, of which Artificial Intelligence represents the currently maximum possibilities. Then, as to transform oil in fuel we need a complex technical apparatus, the results of machine learning are not possible without a culture of data, that includes specialist knowledge and technologies. To this, we add that architecture is a complex and viscous activity: despite the wide interest in digital and information technology, only a few practices have embraced the full potential of advanced digital methods and programming.

Then to investigate the spread and contributes of artificial intelligence in architectural design we interviewed ten digital champions of AEC, professionals with great experience in digital design: Daniel Davis, former Director of Research of WeWork, Aurélie de Boissieu, London Head of BIM Grimshaw Architects, Xavier de Kestelier, director of Hassell, Al Fisher, Head of computational development of Buro Happold engineering, Irene Gallou, Head of specialist modelling group of Foster+Partners, Harry Ibbs, Europe design technology director of Gensler, Andreas Klok Pedersen, partner at BIG – Bjarke Ingels Group, Arthur Mamou-Mani, director of Mamou-Mani studio, Edoardo Tibuzzi, director of AKTII and Pablo Zamorano, head of computational design of Heatherwick studio.

At the end of the interviews, we asked them to order the enabling technologies of industry 4.0 with digital techniques characteristic of our field, based on the usefulness they will show in architecture. The interviewees, therefore, gave a vote from seven to one to the technologies of Internet of Things, Augmented Reality, Virtual Reality, Digital Manufacturing, Machine Learning, BIM and "Other computational methods" category which includes other methods not mentioned among others in the group. We asked the respondents to make a prediction about the next 10 years based on their impressions, taking for granted the benefits already showed by these technologies in contemporary practice within their experiences, with obvious reference to Building Information Modelling.

As is clear from the figure, Machine Learning is the technology that got the highest score with 53 out of 70 achievable points, followed by digital manufacturing with 51, third "other computational methods" with 47, then IoT with 38, BIM and Augmented Reality with 37, last Virtual Reality with 30. It is interesting to note that Machine Learning and "other computational methods" both obtained the first place in the personal rankings of the designers four times, as well as happened twice for BIM and once for Digital Manufacturing. It is therefore clear that based on the experiences and expectations of the interviewees, machine learning and its derivations will play a role within the architectural practice, a contribution that, for many of the interviewees, will be decisive in ten years.

According to our respondents’ answers and reflections, there is a confidence that artificial intelligence will be highly useful for professional practice over the next ten years. It is important to note that the interviewees also see utility for architecture in technologies such as digital manufacturing or other computational methods, of which artificial intelligence could become an engine.

The relationship between Artificial Intelligence and professional practice is currently at a recognised pioneering status: few studies and companies with IT skills, familiarity with data, experience, and parallel computing capabilities, are exploring the applications of advanced artificial intelligence to architecture. Within these offices, some most advanced methods of machine learning and deep learning have shown significant results in limited areas, in terms of data analysis and optimisation, also proposing effective spatial configurations when the boundary conditions are duly defined.

Data, which plays a central role within each machine learning strategy, must be a lot and consistently organised. These algorithms use thousands or even billions of examples to learn autonomously, and the amount of data is decisive for the result's quality. Among the first data used with these strategies, we may easily find those produced in years of professional experience, especially simulations, together with data created to address certain problems, and data collected on the field. That of data is one of the most sensitive subjects from an ethical point of view: more and more actors collect data, from technology giants to governments, and it is important to question their validity upstream of any artificial intelligence process. Architect's responsibility is therefore enhanced: while the architect can question the machine's outputs, he must also be able to understand the quality of the inputs with which he trains the machine.

As Daniel Davis testify inside our interview: ?I don't think machine learning should be seen as a magic bullet, it's a difficult and complex technology to use, and only really works in certain situations?. Given the complexity of architecture, our interviewees exclude that artificial intelligence will have disruptive effects in our field in the short-term, the role that AI will probably assume in the immediate future will be that of an assistant, extremely specialised and able to solve very specific problems.

Even if it is not yet quantifiable how much the introduction of these hyper-specialised artificial entities will erode from the architect's practise, we can assume that our profession, within some of the most innovative studies, will further approach computer science, statistics and science: proficiencies needed to obtain information and solutions capable of speeding up the project and improving the final architectures. The contribution of software houses within the phenomenon of artificial intelligence in architecture remains to be investigated. It is probable that, with proven usefulness and diffusion of AI within the most innovative architectural firms, software houses will be decisive in the dissemination of new tools developed ad hoc for our sector.

To know more about this research, you can read and download the full paper: The role of Artificial Intelligence in architectural design, conversation with designers and researchers, written by Giuseppe Gallo, Giovanni Francesco Tuzzolino and Fulvio Wirz.

The full interview with the Architects is published with Giuseppe Gallo’s doctoral thesis.

Acknowledgements

We warmly thank the architectural offices and designers who have agreed to participate in our series of interviews, our thanks also go to Francesco Grigoli for his technical comments and suggestions.