The future of architecture in the context of Machine Learning: Why architecture as a social science needs to develop new knowledge systems.
Can design judgments in architecture be true or false, or are ethics in architecture a purely subjective matter for individuals to choose or are perhaps relative to the culture of the society in which one lives?
Architectural machine learning platforms like Google Delve, a new A.I architecture machine learning platform makes a tenuous promise to ‘empower urban development teams with cloud computation and machine learning to design better cities, faster, with less risk.’ They allow architecture to be designed and achieve planning without the input of a designer. This poses a potential quantum leap in the field of city design, but also poses ethical problems, and challenges the profession to rethink itself as a field of knowledge.
The first and most apparent aspect of design by the algorithm is that machine learning demands objectivity from a profession that has been almost entirely reluctant to give objectivity to its knowledge. Without the input of architectural knowledge, architectural design A.I is frightening. It will run purely on economic parameters as a default logic, potentially impacting housing standards for millions of people, especially in the developing world where this technology will have the biggest use case.
But is there a way to avoid a dystopia, where all housing is made by inalienable economic logic — as the key parameter for how we live our daily lives? It might be possible, but that is if we can successfully formalize the practice and process of making architecture's knowledge more objective. This way we then we can inform the algorithms of architectural A.I with ethical or social standards as the input.
This not only presents a challenge but also presents a rare opportunity to influence the development of an emerging field, and upscale important discoveries to overcome the problems of an asymmetrical distribution of knowledge across the world. Many municipalities have limited knowledge of social architecture but an enormously high need for a better quality of life that architecture can bring.
It has many critics: architectural A.I might omit some of the poetry from the architectural process, others fear it will destabilize the business model for architects. Both are true, but in reality, Architectural machine learning represents an unavoidable technological change that presents a zero-sum ethical position for professionals. We either contribute to architectural A.I and ensure its parameters are as well developed as possible, or we watch it expand exponentially without the right early input, meaning we are complicit in these problems. And it requires us to dust off the lessons from the likes of Christopher Alexander and build on this work.
If technological growth and population growth continue, influencing architectural A.I itself represents one of the largest opportunities of the architectural profession in all of its history. The sheer breadth of impact it will likely have in the future of the built environment is unparalleled. For the first time complex, architectural knowledge can be scaled laterally.
It would definitely require that we change the focus for the architecture profession. We need to think ourselves away from design as a form of physical production towards formal knowledge production. This largely means architecture's social research should become more important than architectural practice itself.
Architectural machine learning might have positive use cases, for example, it might act as a conduit for knowledge practitioners to indirectly inform large-scale developments that would otherwise be very difficult to change because of social or geographical disconnect from researchers, or the value engineering of the developer client.
Research frameworks in this area must be taken more seriously. So the question is: how can we consolidate architectural knowledge and make it universally and objectively useful, which can feed into Machine Learning to ensure quality control and the values other than economics in the built environment in the future.
Influencing architectural Machine Learning is really a question of ‘what matters’. Derek Parfit argues that unless we can show that objectivism is true, he believes, nothing matters. It is very difficult to influence modern society in a positive way without objectivity.
One way of approaching ‘what matters in architecture’ is by understanding humans from a neurological perspective, and what we need from society as a result of the way our minds work. For example, our cognitive capacity to determine the number of stable social relationships we can hold at any given time is limited, and we witness the implications of this in the size of our vernacular towns, which normally contain around 150 people. These vernacular settlement sizes have implications for communal space in the present day. Although our residential conditions have changed, human neurology itself has not transformed all that drastically in the last several thousand years.
Another example is in the social capacity of humans. How we hold a conversation, and how many people we can comfortably maintain stable conversation at any one time (four to five). This is related to our memory capacity, and ability to concentrate but also holds implications for the way we balance our living spaces in dense urban scenarios.
Imagine the last time you were in a smoking area and witnessed all of the conversations around you taking place. They are almost always in groups of four or five. Here you are witnessing the physical arrangement of our species’ memory capacity, and the nucleus of what we are calling social architecture. It’s this understanding that is the basis for objectivity in architectural knowledge.
By understanding the architecture in the mind first, we can begin to trace how missing spaces in our cities evolve into social issues, mental health issues, and how these will culminate into mainstream public health concerns if they are compounded. Today 1 in 5 people attend doctors’ surgeries for non-medical issues, and the majority of these issues can be traced to social isolation and are rooted in architecture, yet it is entirely unacknowledged as a root cause of many costly public health issues.
Arguments for objectivity
Historically there has been a negative reaction towards objectivity in architecture. The rapid progress in architectural algorithmic design, such as google’s Delve, creates a critical urgency to consolidate findings more effectively; Moore’s Law explains that technological breakthroughs happen every two years and will continue to happen, and delve will likely follow the same trajectory as all other sectors affected by the exponential curve of evolution technology provides. Therefore, the advancement in developing neural networks and machine learning technology would most likely lead to automated design transforming the architectural profession to the point that it is unrecognizable within the next 20 years. With it we can imagine our cities will be built and look completely different too.
This reality gives us an ethical responsibility to consolidate knowledge and articulate the objective parameters or ‘the self-evident normative truths’ we do know. It also pressures us to identify gaps in our understanding and incentivizes us to conduct serious research in areas in which we lack knowledge.
If there is no input into the parameters of architectural A.I am from the socially orientated architectural profession, there is a high potential for all algorithms to be purely economically driven. Mass-produced actions will create ‘batch mistakes’, and will particularly affect the emerging economies which are projected to leapfrog the developed economies by using these sorts of technologies in the next 30 years. Consequently, good work in this area could fundamentally alter and better the lives of hundreds of millions of people as the world evolves.
Universally useful knowledge.
One of the major missions of the architectural profession in the future must be to consolidate architectural knowledge and to make it universally useful. Social architecture knowledge must be verifiable to hold weight against traditional metrics like economic logic. ‘Universality’ means that once findings are approved for one purpose they can be utilized for another purpose or scheme. For example, findings might be crystallized to feed into A.I but the same information can be used to challenge regulations around the built environment in certain countries; it is not tied to a single-use.
Building an objective evidence base that is rooted in the social sciences, neuroscience, and behavioral psychology is an ethical response to unwavering technological progress. Without being steered by some universal grounding architectural knowledge there is a real risk that architectural machine learning will overlook the parameters that are most important to the human condition.
We lack a fundamental dataset here, and although consolidation attempts have been limited by professional attitudes towards this pursuit, it is highly important to create the right structures to allow this today.
A.B is a new multidisciplinary research group looking at sandboxing ideas at the outer limits of architecture. It gets its name from A.B testing (also known as split testing) which is the process of comparing two versions of the same thing in order to test which of the two variants is more effective. In this sense, our proposals may directly contradict the mainstream conditions of architecture with a view to enhance opportunities. Without the simultaneous testing of different pathways, there is a greater risk to the failure of the architectural industry. Members include social architecture practitioners from the likes of Assemble and project Foodhall.
A.B is thinking about creating research, peer review structures, and building communities of contributors to define the objective parameters from successful multiple emergent social architectures and pool findings.