When I learned about the concept of deep learning, it had never occurred to me to apply it to architecture.
The reason for this is that in a sense, the activities themselves that we carry out on a daily basis within our design office actually already consist of the make-up of deep learning. Deep learning implies the expression of the capability of problem-solving due to the accumulated “experience” through previously resolved problems; simply put, it does not sound so artificial or mechanistic. In that sense, we are not akin to the machine in that our learning process is deductive, but the machine is akin to us – we make decisions based on decisions we have already made – with the crucial difference that the machine has to be pre-programmed to learn that it is learning. This necessitates a structure of rules that dictate a didactic learning process.
In retrospect, every architectural work I have ever completed (or not) represents a single frame in the process of a deep learning library. This is fitting, because an architectural project, or one single “piece” of architecture, is not something that is ever complete. Naturally, I know that it is not perfect, or ever will be, but upon its never completing completion, it is rather a cross section of a continuous learning process. It becomes an object of endless observation and deduction – a part of its originator’s archive. This notion is much unlike the attitude of architects from the previous generation, who perceived each piece of architecture as a completed work of art for display, never to return to it. The significant aspect was that the work itself was autonomous, a separate entity from its art, and itself completed, rather than part of a larger process, of growth and expansion. The important thing to consider is the differential change or variation from one piece of work to the next. This string of consistent change marks the evolution of an architect over time, and the maturing of his character – his repertoire; each project, each frame, each procedure, part of the process of an architect’s evolution, becomes intertwined, and therefore part of another, overarching learning process – defining in an architect’s persona. In realizing this, I came to understand very clearly the age in which I am living as an architect – an age of technologically enhanced skill, artistry, performance and cognition – and how I differ from other architects.
The revolutionary aspect of deep learning lies in the fact that learning continues at a speed and volume inconceivable to that of the past, whether or not compensation can be obtained with a particular “skill” as the basis; without teaching any rules. In other words, the innovation behind deep learning consists of the algorithm’s ability to discard the limiting concept of rules when deemed necessary and productive. However, I do not think that this is a “revolutionary” learning method. Children interact with the world around them via understanding compensation as the founding reward of their learning, oblivious to any imposed rules that may influence their learning process. Being sneaky may equate not getting caught and getting caught may result in punishment; experience becomes a learning repertoire. The pre-adolescent compensation-led learning is simple, and very much related to the deep learning process. The unfortunate trigger that causes maturing away from this primitive learning method, to learning in an environment where rules are imposed, consists of the intermediary time that it takes for it to become necessary to adapt to a society that functions on the basis of fixed pre-determined, pre-meditated or consequential “rules”, or adherence to societal norms. Moreover, growing and maturing includes demoralization when these “rules” – or societal pressure – dictate that there is an “absolute value” that transcends, whether or not “compensation” is provided.
If the limitations of a fixed rule-based system are true, the circumstances that generated deep learning should be considered to be the fluidization of society and the downfall of the standard for an absolute value, rather than advances in technology and calculation speed of applications. There may be rules in society that are respectively fixed to a certain extent, but we have already started to realize that they are only relative, and not attributed to precise, absolute values.
From this perspective, I think that it is very strange to view deep learning applications as a threat to humans. People will always be the cause of these threats, in the end, the cause of bad things is the “artificial” construct of “society;” limitations that are imposed, trying to regulate a naturally occurring phenomenon for order.
The same can be seen in the world of architecture. Despite necessitating variables, nests were never designed via a rule-based system. It was also not representative of order; quite contrarily, it was freely and productively chaotic. Living organisms freely design nests on the basis of being “compensated” by the survival of their offspring. Nests are the original form of architecture. Under this lens, continued free learning was the essence of the culture of architecture. The output of this free “deep-learning” over time evolved from nesting to dwelling, and its inhabiting organism, to a human. The infinite degrees of freedom concerned with the freedom to learn are the make-up of the evolution of our discipline.
Rules started to become necessary when the need arose to strengthen order in the production of a fixed and organized society. This led to inscribing and defining by inscription (“set in stone”), the very definition of architecture in the writing of various architecture books, of which Vitruvius is notably representative. The system of rules cycled back in the same manner during Modernism; simply and solely, a new and modern set of rules similarly dictated and imposed. The fact that Le Corbusier focused on the “Five Points of Architecture” is further proof. The foremost rule of Modernism is that “Decoration is a Sin,” and this thesis dominated the 20th century. Naturally, a sin consists of a violation of the rules. A sin, would have been defying Le Corbusier. The normalization of opposition towards the proposal of something new or different is stagnant; inversely, a sin in the face of learning. Be that as it may, people in the world of architecture no longer talk about rules. The world that could be controlled by rules has since dissipated. What type of “compensation” system is now in place that replaces a suggestive rule-based system?
Starting in the 1980s, a trend emerged within which the “ability to sell architecture for a high price” fulfilled the role of “compensation;” “the Age of Compensation.” I call this the “transformation of architecture into art.” Similarly, to artists, the accession of a visually associated identity by architectural consistency as creators heightened the “value” of that creative individual, consequentially leading to an increase in the value of the work by that same individual. This system represented the maximum “compensation” for architecture, which dominated the past few decades under the pretext of starchitecture.
However, it is my belief that architects have begun to realize that this form of compensation does not satisfy, or produce fulfillment; including society as a whole. This compensation system was extremely effective in increasing the price of architecture, such as the sales price of condominiums. “Art” architecture that increases the price without any reason is generally systemically rejected. This type of approach is finally, but slowly coming to an end.
I am attempting to replace it with architecture that makes people around it joyful; something that is extremely ambiguous, where compensation elicits the opportunity for deep learning, by depending on it. Naturally, I want the people that use the architecture to express an emotionally positive response, but I also want the people that live around that building, or who simply pass by it daily to portray that emotional experience. I believe it is difficult that architecture may convey such emotional experience in the true sense of the word unless the craftsmen and the various other people involved in the construction of the building have a sense of fulfillment in the process. I don’t think myself and others who design can be said to attain this fulfillment unless personally engaging in a learning process during the designing of those structures – rather than simply repeating the same design over and over again, like a printer. Even if architects receive holistic compensation, I don’t think that you can be content unless it involves a learning process.
By this default, architecture is an emotional project. During the process, it is a journey; upon completion, an experience. An emotional experience cannot be dictated, or told how to operate or unfold; it has no rules. Thus, neither does learning, as a byproduct of the emotional experience. As far as computing is concerned, the joy of architecting cannot be experienced by it. Or experiencing the joy of inhabiting it, for that matter. No less show it. Although it is most useful in the process, and in-betweens of these very human experiences, there is no better compensation, or fulfillment to the emotional project of architecture than that joy, that may not be transferred to a machine unless it is scripted to feel joy; also a dictation. It is this susceptible excitement that we share that allows us to relate to each other in the production and experiencing of architecture as humanity’s longest lasting cultural project. Through the problem-solving nature that is architecting, in the process of architecting, deep-learning is profoundly and instinctively human.
Deep learning aided the realization that the value of the joy of all elements and experiences concerning architecture: the intrinsic humanism to architectural progress. Despite its infinite capacity to learn, a machine cannot and will not be emotionally invested in the architectural project. Its utility is perhaps best preserved and incorporated through computing decisions made by its operator. This way, the emotional investment is preserved, processed, and materialized by the architect. As far as artificial processes, they are modeled after our own, and our ability to comprehend the world and make decisions. For as good as anything artificial may be at reproducing the acting out of the repetitiveness of labor, it remains an incomplete reproduction of what we have always been so good at. We should be the authors of the world we build.