One of the most important developments in the area of Architecture, Engineering, and Construction (AEC) will be the way designers relate to climate change and the notion of climate adaptation as a primary response to the environmental pressures from climate change. Before we move into solutions, there is a need to understand and critically reflect on what adaptation, within the realm of design, really means and how it is apparent in nature, which is a constant inspiration for environmental design innovation.
A critical reflection always begins with an attempt to clarify and define important aspects of a problem. There are currently three major misconceptions in the AEC industry, nested within three important processes of design: Parametric Design Optimisation, Ecological Design and Evolution, Climate Adaptation.
PARAMETRIC DESIGN OPTIMISATION, DESIGNING WITH PARTIAL INFORMATION AS THE NORM
Parametric Design (PD) is defined as ‘a process based on algorithmic thinking that enables the expression of parameters and rules that, together, define, encode and clarify the relationships between design intent and design response’ [1]. The word parameter is at the core of parametric design, but the ‘relationship between design intent and design response’ is as crucial. It is this relationship that parametric design is supposed to uncover to inform the design and allow it to be optimised iteratively and in a quantitative manner.
Unfortunately, in practice, this relationship is rarely adequately explored. The misconception here is that we think this partial understanding is enough to properly inform the design process in a ‘better less than nothing’ approach. Practitioners claim to have ‘parametrically optimised’ their designs even if a handful of scenarios, a fraction of the possible design space, are assessed.
The industry is being short-sighted here in a very literal sense;
while it can ‘see’ the most minute of design spaces, it is unable to explore any significantly sized design space, at least under most performance-based evaluations.
MACHINE-LEARNED REGENERATIVE
Bellman’s notion of the curse of dimensionality [3], the exponential increase of design alternatives in high-dimensional spaces, is not something unique to AEC but has had a tremendous effect on the quantity and quality of our design spaces and processes. Even a simple box model with minimal computational requirements, often used as a proxy in the AEC, suffers from this curse (Figure 21). The exponential increase of design alternatives, computational time, and resources required to conduct comprehensive parametric optimisation studies is one side of the problem. The fact that this leads us to design spaces with incredibly sparse parameters is another much more important and underestimated problem. It seems that the curse of dimensionality limits not just our design spaces, but our creativity and imagination as well.
ECOLOGICAL DESIGN AND EVOLUTION, NOT AS CLOSE AS WE THOUGHT
So how are we to overcome all this? Various Ecological Design approaches have been widely used in AEC for many years in trying to tackle this very problem [4]. Perhaps the most popular has been Evolutionary Optimisation (EO), and specifically Genetic Algorithms (GA), which are often employed to find optimal solutions to difficult problems. The core idea is to start with a few simple rules (parameters) and allow for complex systems or designs to emerge. To reduce the dimensionality of the problem and the number of alternatives needed to be assessed, the optimisation process then focuses, and indeed filters, the best performing alternatives instead of exploring the design space itself.
playing with façade
the curse of dimensionality
orientation (8) * wwr (7) * sill height (2) * glz specs (3) * wall specs (3) * shdg lenght (3) * shdg angle (3) * shdg orientation (2)
climate
* location (6)
design alternatives
6048
36288
time
50 hours
300 hours
indoors
* hvac (4) 145142 1200 hours
or 50 days
massing
* shapes (4) 580608 4000 hours
or 200 days
Figure 21
The basic mathematics of parametric design optimisation. The explosion of design alternatives, even under simple parameter combinations, is a significant bottleneck for conducting truly parametric optimisation studies.
However, what if nature did not evolve in this way? What if the beginning of life was much more complex and beautiful than we thought? What if there were more designs then than there are now?
I was fortunate to find an unexpected inspiration on this topic in
‘Wonderful Life: The Burgess Shale and the Nature of History’, an incredible view into the [hi]story of life by the great evolutionary biologist Stephen Jay Gould [5]. Gould completely deconstructs and then reassembles the notion of evolution, changing everything we once thought we know about it, including the ecological design principles we habitually apply in our generative design processes.
It turns out, Gould tells us, that at the beginning of life, deep in the Cambrian sea almost 500 million years ago, there was more design diversity (more unique design archetypes) than exists today! Perhaps not surprisingly, nature is a better parametric design optimiser than us. It is crucial to point out that this indicates that the richness and diversity of nature do does not seem to be the result of ‘millions of years of evolution and natural selection’, a popular catchall phrase. Instead, it was the condition of open exploration and experimentation from the start that was crucial for the development of the wonderful story of life as we know it.
CLIMATE ADAPTATION, SOLUTIONS BEGIN WITH NEW DEFINITIONS
This brings us to our final misconception, adaptation. Inspired by Nature once again, we have selected ‘adaptation’ and ‘resilience’
as the central paradigms of ecological strategies that are currently being developed and implemented to combat climate change, perhaps the single most important issue of our generation. The notions of ‘adaptation’ and ‘resilience’ are both related to the flexibility of designs, policies, and strategies, which allow our urban environments to return to their previous states after sudden shocks, climatic or otherwise.
At first, glance, designing environments that can withstand shocks seems to be a logical choice on our part, as, after all, we do like this kind of stability. However, Nature once again does not operate in such a manner. Nature has never been simply flexible but has always had the capacity of differentiation and radical transformation under adversity and contingency. In this sense, Nature is much closer to what neuroscientists would call plasticity; that is, the explosive potential for differentiation and functional change [6]. If we hope to indeed reverse the course of environmental degradation with strategies that will make for truly resilient cities, we need to turn to plasticity as our core principle of design.
This would mean that our designs need to not only be able to adapt and resist climatic pressures but be able to differentiate themselves under varied environmental conditions. We need to learn to design for all contingencies at once without sacrificing performance. This means that designs, and by association our design processes, can no longer be passively subjected to predefined heuristics but need to become active subjects of creative transformations. The potential for ‘resistance,’ ‘negation,’
and ultimately ‘explosion’ that plasticity offers will bring new transformative possibilities for the future of design [7].
BRINGING IT ALL TOGETHER; FROM DATA TO INTELLIGENCE-DRIVEN DESIGN
The big question, of course, is how to design for all contingencies at once without sacrificing performance. How can we escape our preconceptions, disrupt our design thinking, practices, and workflows? How can we learn to be curious and playful? Finally, how can we design buildings and cities that can not only survive but thrive under many conditions at once? My claim is that Machine Learning (ML) can be the framework under which we can finally integrate the above ideas in our design processes. Thought out and implemented thoughtfully, avoiding the hype and all the noise, ML can change the nature of the relationship between learning and design, allowing us to do the parametric design on generative scales!
Implementing ML requires us to change various aspects of design and optimisation processes. We first need to develop AI-capable design environments that can handle and export large volumes of different types of data (tabular, image, semantic, etc.). We need to connect these environments to the many different ML frameworks in an efficient and transparent way. Finally, we need to follow the AI industry’s example in developing this new constellation of design software in an open-source and accessible manner. Only in this way can we hope for the dissemination of knowledge and ML-infused workflows in the AEC industry at the pace at which we need them.
All this should give us the chance to start our designs from the point of diversity, just as life did all those millions of years ago.
It should allow us to finally shift our focus from the almost obsessive, goal-driven optimisation to the more playful design exploration, a process of learning from the design space itself. If we are lucky, one day this might even give us the chance to be surprised by our designs, to broaden our thinking and design practices to include the notions of plasticity and differentiation, leading to truly resilient designs that are gravely needed under the pressure of climate change. After all, isn’t the goal to have a positive impact on our world?
REFERENCES
[1] Jabi Wassim (2013), Parametric Design for Architecture. London: Laurence King, 2013 [2] H. Snellen, Probebuchstaben zur Bestimmung der Sehscharfe, Utrecht 1862.
[3] Richard Ernest Bellman (1961). Adaptive control processes: a guided tour. Princeton University Press
[4] Luisa Caldas, Leslie Norford, A genetic Algorithm Tool for Design Optimisation, Media and Design Process [ACADIA ‘99] Salt Lake City, (29-31 October 1999): 260-271
[5] Gould, Stephen Jay. Wonderful Life: Burgess Shale and the Nature of History. New Ed edition. London: Vintage, 2000.
[6] Doidge, Norman (2007). The Brain That Changes Itself: Stories of Personal Triumph from the frontiers of brain science . New York: Viking. ISBN 978-0-670-03830-5
[7] Catherine Malabou, Plasticity at the Dusk of Writing, Columbia University Press, 2010, DOI:
10.7312/mala14524
The recent explosion of digital tools for environmental analysis has made it easier than ever to ‘do less bad’. Measuring the damage we do today and reducing that of tomorrow is not yet trivial, but there are countless software packages and metrics available to help. Regenerative design, however, holds designers to a higher standard, and actively improving human and environmental health requires more than an arsenal of metrics. Improving health requires a vision greater than any single number, and oversimplification poses a danger of treating the newly available tools as implements of technocracy. Few architects enter the field of design to be technocrats; most are motivated by creativity. The design process can be a turbulent sea to navigate, and a project’s destination is ultimately determined through discussion. Digital tools play a key role in describing regenerative strategies within these design discussions.
The last several decades have seen a gradual transformation in the digital tools themselves, from tools which were designed specifically for evaluating HVAC systems towards tools intended to assess the performance of an entire building. The family of plugins in Ladybug Tools builds further upon this development and acts as a node between the realms of building science, thermal comfort analysis, and architectural design [1]. Architects can now retrieve a great variety of daylighting and thermal comfort metrics at the level of spatial precision necessary for making design decisions. This astounding degree of analytical flexibility has resulted in a subsequent Cambrian explosion in forms of digital analysis. Designers can build a systematic framework specific to the questions of each project.
The result is that architects now have access to rich bodies of information composed of multiple metrics when making design decisions. This opens possibilities for regenerative strategies, but it also makes it necessary to choose which metrics are most important. It will always be easier to optimise isolated measures as an effort to ‘do less bad’ than to map the constellation of metrics which prove a proposal is actively contributing to human or environmental health.