(1) Introduction
Architecture and the development of the built environment, in general, continue to be disoriented, let alone to be coherent, when it comes to adopting technologies referred to as artificial intelligence – AI for short, and machine learning more correctly; they stand aside from the current that has been invading our lives and professions since the 1980s. Concurrently, news in brain biophysics and related computer science along with new computer technologies have emerged recently that render to offer to architects, other project stakeholders, and the public creative and communicative means and tools that the field has been missing throughout its history: the three- and more-dimensional spatiality and modeling together with diachrony lay at the heart of both (i) the new AI computational models and technques to be developed based on the new computational theory of mind and brain to provide superb capabilities for design, planning, and parameters review and assessment, (ii) state-of-the-art virtual reality technologies featuring unprecedented abilities to design architecture and the built environment as it deserves – a diachronic way in space, to create architecture from spaces, to understand it and communicate – and (iii) latest contributions of architectural theory to design practice, the theory of public space in particular. (This regard requires a precursory knowledge: Architecture emerges when a work of (physical) construction, communicating cultural, social, and material values with humans and society, is exposed in public space [1].)
And third, regardless of the novelties in biophysics, virtual and extended reality (VR/XR for short) technologies, and architectural theory, the AI practice, research and development (R&D for short) know and at a hectic pace further release strategies and techniques that have not yet been applied concerning architectural designing and planning so far – such as imitation learning, self-play, and transfer learning among others that could push the branch and the field forward – to the next level. Though easily overlooked as irrelevant at a superficial glance, AI-driven robotics is the realm where the main incentives for new AI developments for architecture and the development of the built environment are emerging. The opportunities, however, have met only a hesitant, if any, reaction in the respective professions so far, which aligns with the sidelining that architecture and the built environment, unnoticed, have been suffering for the past seventy years [2], losing the status and function as the summum templum [3], which they held and which represented their meaning and role within the societies, cultures, and economies throughout millennia.
A question arises: can the confluence of circumstances – opportunities and challenges in general, and, in particular, the newly revealed fundamental interconnectedness between architecture within the built environment framework, virtual reality technology, and mental, thus computational activity stemming from shared aspects of spatiality, spatial modeling, and diachrony – the trinity of three- and more-dimensional spatiality for short – be a prerequisite for a singularity in the evolution of the profession and the field?
The answer cannot be but a composition of – on the one hand – the computer scientist’s and developer’s ability to identify the problem niches and find cutting-edge solutions to fill them, and – on the other hand – the branch’s motivation and ability to adopt the new approaches and tools. A lack of mutual understanding of the needs, capabilities, and processes of one and the other expertise between architects on the one hand and AI applications developers and data scientists on the other appears to have contributed to the embarrassing results so far; and the gap continues to widen. Among other preconditions, correct understanding and grasping the aspects of authentic – poiétic [4] or poetic as will be asserted later in section (3) of the paper – architectural creativity on the one hand and parametricism in architectural designing on the other are essential.
Contributing to understanding the situation, uncovering the causes, overcoming the unfavorable trend, identifying the prerequisites for effective cooperation in the development of truly productive applications of AI in architecture and the built environment development, and bridging the existing gap between the professions in general is the ultimate ambition of this paper. To stand to the ambition, the paper must address architects, construction and real estate professionals with AI researchers and developers, and also tech investors. Each professional field has its language that fails to be understood comprehensively by the others. While trying to bridge the gaps, some text sections may seem redundant or inaccurate for one or the other profession: for such cases, the author begs for leniency. Timeliness, eventually, is another challenge that faces working in the field developing at an unprecedented, more than hectic pace – more than ten thousand new solutions and hundreds of millions of new users within a year. However, a field that, so far, has been lacking a critical overview more than needs to shed light on the status quo to lay the foundations for efficient further research and development.
Architecture (and the built environment development) on the brink of a revolution
The amount of R&D efforts on AI in architecture and the built environment development significantly lags behind the influx of applications in the economy, insurance, healthcare, the social field and politics, weapons systems development, the judiciary, environmental issues, and in general.
In the field´s modest conditions, a gradually increasing flow of so-called AI applications has been entering the field of architecture since the 2010s, even before. After a decade, some of the applications integrate into the workflows of architectural teams; however, the results do not match the previously declared ambitions in terms of performance and quality. AI has not proven creative (not only) in terms of architecture; even the pioneers in the field are abandoning the projects the ambitious visions not achieved and reducing their efforts to pragmatic parametric tasks. On the other hand, AI-driven applications slowly-slowly start to spare the architect’s time and energy as his “new pencil” – making sketching faster – or substituting and widening the research of ideas and themes concepts. In addition, a complex AI-enhanced process has been introduced, from an expression of the client´s requests and expectations to the final design of architecture and construction solutions, as section (2) of this paper reports. All this, nonetheless, lags behind the vision of an (AI-driven) assistant, an apprentice of an architect that section (5) of the paper introduces as feasible, moreover realistic a prospect of the apprentice overcoming the master in terms of tirelessness, consistency, and accuracy.
To lay a path to overcome both the architecture’s and built environment development’s sidelining and the field’s failure to keep up in terms of adoption of the progressive technologies, the paper offers a comprehensive approach: first, an assessment of the feasibility of the past and recent visions and goals and an analysis of successes, flaws, and failures of the last decade. The so-much-wished true creativity of artificial intelligence remains at the heart of research: is its unattainability only a question of state-of-the-art approaches and technologies, or is it a principal misconception? Second, the deployed machine-learning networks and algorithms are reviewed more deeply. Section (5) of the paper concerns the potential of generative adversarial networks and supervised learning techniques that, so far, have been ruling AI-driven-designing attempts and proposes reinforcement learning strategies and imitation-based and self-learning techniques (all terms brought up in (2)) to take over. In terms of (non)availability and how to provide it, the respective training material is the third principal issue. And fourth, strategic fields of feasible and efficient deployment of AI in architecture and AEC (architecture, engineering, construction) are proposed and supported by arguments: “reverse prompting” and AI-fostered recherching, generative-patterns-based pre-design, design-development support by „advice whispering“, continuous and complex assessing of parameters of physical and quantitative aspects of an architecture and respective constructions, design reviews, and evaluations of solutions. Despite asserted incapability, let alone the true creativity of a learned machine, still a revolution may lay ahead of the architecture and the development of the built environment; the so-far overlooked potential of imitation-based learning and self-learn strategies may ignite it. Given the inherent computational demandingness due to the three- and more-dimensional nature of the tasks, a generally sought-for breakthrough in computational technologies, be it spatial computing, ternary semiconductor design and neural networks (introduced in section (4) of the paper), or another power- and performance-efficient model may prove to be a prerequisite for practical development and deployment of these strategies.
A breakthrough
Until recently, R&D in AI in architecture has been challenging architectural creativity and threatening to make the architect redundant. Substituting the human brain’s work with the performance of computer networks and various, ever-new-developing algorithms should have been the means. However, principal flaws related have not ceased to emerge.
A new theory of the human brain’s working mechanisms appeared recently to debunk the general working principles of today’s AI algorithms when it comes to substituting or competing with the brain’s or human Neocortex’ performance. Even then, however – further elaborated in section (3) of the paper – a computer remains incapable of consciousness and of true, poiétic creativity. In addition, such a “mission statement” fails to catch up with the natural potential of a breakthrough technology. If a technology is truly groundbreaking, it must perform unprecedented outcomes; if the new outcomes can compare to the existing ones, the technology is not a breakthrough. Had the father and son Stephenson’s Rocket not been a breakthrough, railway carriages would have continued to be pulled by ropes wound on stationary winches located along railway lines [5] – and the development of rail transport, and with it the upswing of the global economy, living standards, and culture, would have been delayed by decades. Innovation is a breakthrough only if it changes not only the quantity of output but also the quality of performance – as we have been witnessing in the case of the Internet, originally intended only to enable an exchange of documents between institutions, but profoundly and globally changing social values and relations, modes of communication, economics, culture, and environment in the end – and still, it is not an end [6]. Today, however, both AI developers and (the few) interested architects are only thinking about improving winches – instead of dreaming of the next generation Rocket. It is time to start dreaming instead of plain coding: as Rutger Bregman points out, But the real crisis of our times, of my generation, is not that we don’t have it good, or even that we might be worse off later on. No, the real crisis is that we can’t come up with anything better [7]. Can architects offer better? They must; through architecture different from what we know so far, be designed not by AI but by architects grasping the AI opportunity hand in hand with computer and AI scientists and developers – and supported by tech investors that, so far, have identified close to no opportunity in this field. However, this is a misunderstanding: architecture stands at the launch of any real estate development: as such, it delivers the starting points for the business that amounts to hundreds of quintillions of EUR globally. Such is the figure that represents the basis of comparison. A five percent yield of such a figure is still a quintillion. It is not just investors who should finally take notice of this opportunity [8]; the field and the architectural craft deserve the investments – and ultimately, society as the end customer deserves them, too.
How is the paper organized
To unleash the potential – investment opportunity included – an essential interconnectedness between the three realms of the trinity of three- and more-dimensional spatiality and diachrony needs to be understood and the resulting singularity must be grasped in theory and deployed in practical use. Till then, however, even with “only” the approaches and tools that are already at hand though remaining not mainstream, there are still significant steps of improvement happening or to be undertaken. Today already, AI in architecture is asserted not only to generate images that help to articulate expression of the poiésis of the architecture within design concepts (or to customize them) or to shorten the way from a simple sketch to materiality-rich rendering. The trailblazers already coin the AI design process replacing the traditional design process from the client´s vague idea to final design with CAD/BIM (computer-aided design/building-information management) model integrated. Also, providing predictive simulations and models starts to be a business-as-usual today; section (2) and the second and fourth subheads of section (3) of the paper zoom in on all the state-of-the-art. In addition, processing generative patterns to a pre-design, a solution as close to the set parameters as the stock of generative patterns allows, final adapting the pre-designs and tailoring the final, specific solution, or continuous and complex advising through the design process, designs´ reviewing and solutions´ evaluating, or optimization criteria specifying – these are the strategies, practicals tool subsequently that render on the horizon currently and in sections (4) and (5) of the paper.
To allow for this, new classes of the existing algorithms and strategies of imitation-based learning and self-learning zooming in on the design-development processes instead of the results (to be) achieved must enter the field of AI in architecture, perhaps supported by the new mind and brain theory-based computational models, XR, or ternary semiconductor design and neural networks (introduced in section (4)).
Revealed within the research of the state-of-the-art, three focal perspectives of the critical review are introduced to launch the (3) section: authentic, poiétic creativity preceding and transcending parameterization and algorithmization, second, novel, in architectural designing not yet applied learning strategies and training approaches, and third, the trinity of three- and more-dimensional spatiality, spatial modeling, and diachrony that interweave architecture, recently developed virtual reality technology, and the new theory of human thinking and intelligence together with other novel computing approaches. The virtual twins‘ technology, its benefits and prospects are reviewed in [9]; this paper takes them for granted and only puts them into the context of the architecture – XR technology – and computational-model trinity of three- and more-dimensional spatiality and diachrony. The third section of the paper revisits existing deployments of AI in architectural designing and in general, introduces novel AI-learning strategies to be examined, and delves into starting points of true creativity – consciousness, reasoning, and poiésis. The next (4) section delves in the prospects of the new mind and brain (or Neocortex respectively) theory by Hawkins and Numenta in terms of computing model, the contributions the of state-of-the-art XR technology, ternary semiconductor design and neural networks, and nowel approaches in terms of architectural designing and built environment planning.
Fundamental as incentive benchmarks for the „new wave“ of AI in architecture and built environment development, the latest achievements in AI-driven robotics are introduced: Figure-1 starting from section (2) throughout the paper, and Phoenix by Sanctuary AI in section (5).
Section (5) discusses the misunderstandings and limitations of recent attempts and expectations regarding the use of AI in in the branch together with proposals for new R&D. The „fantastic achievements of AI promising to make man redundant“ discussed recently in the profession and beyond are debunked and, in contrast, other, so far not considered approaches and perspectives are rendered: in elaboration of the latest knowledge of human Neocortex working, in parametric pre-design and human-in-the-loop approach, in new classes of the existing algorithms and strategies that strive to mimick (first) the human-in-the-loop and (then) the processes learned, in continuous assessment of diverse outputs of creative architectural and urban design concepts, and in approaching to sustainable development (or, better, to the comprehensive resilience) tasks. The final (6) section concludes the today-available and to-be-developed productive applications of AI in architecture and the built environment development, the computing strategies and networks development „to-dos“, and the perspectives and a „mission statement“ of the field.
Introduction figure: www.dezeen.com
Michal Sourek
