ARTIFICIALLY INTELLIGENT ARCHITECTURE
Edited by: Brian Kelly & Frank Jacobus
ORO Editions / Winter '23
Role: Chapter Author
"Patterns of Interiority" is a book chapter within Artificially Intelligent Architecture, edited by Brian Kelly & Frank Jacobus.
Introduction
Artificially generated imagery has the capacity to open up novel possibilities for design and architecture infused and intensified by linguistics, images, sounds, experiences, time, and tactility. Our critical, disciplined mind has been conditioned by our experiences and education, and while it gives us the ability to analyze and evaluate, the conscious mind has programmed limitations that we cannot escape. Limiting beliefs about what kind of architecture is acceptable, valuable, and critical, restrain our creative process and restrict our freedom to produce innovative concepts. Artificial image-making offers a new space for creation bolstered by an act of synthetic collage, a reconstruction of existing material into new juxtapositions and orientations. Multiplicities and complexities of material content amalgamate to form a collage, much like emergence of artificially generated imagery. Collage, adopted by architecture, was first explored through the literary production of poetry, and it makes sense that prompt generation relies on language to birth an image.1 This creative act has come full circle. The generation of imagery through machine model learning is an act of collage, a synthesis of information and material suggestive of temporality, hapticity, and experiential qualities. Artificially generated images support and expand theories of interiority through episodic and collaged image generation, imperfect representational artifacts, surrealist acts of interiorization, and exploring our most private fantasies via misalignments of form and content.
1. Shields, Jennifer. Collage and Architecture (New York: Routledge, 2014), ix
Synthesis + Genesis
Combining existing architectural DNA, new architectural species are birthed by artificial intelligence. This is not a new mode of creation, but a new venue to engage the act of collage. Juhani Pallasma describes in Hapticity and Time: Notes on Fragile Architecture the profundity of a layered, collaged, and tactile architecture that opposes the supremacy of a singular, visual image.1 “Every artistic work, be it literary, musical, or visual is bound to be a juxtaposition of images, emotions, and ambiences in order to construct an articulated and engaging spatio-temporal experience.”2 Everything is collage- even pristine new buildings rely on past technical, intellectual, and theoretical foundations laid before them. Collage as a form of representation is highly feminine and sensuous; it allows us to experience a heightened sense of interiority, intimacy, and temporality. The incompleteness and imperfections allow for more subjective experiences, encouraging viewers’ own interpretations and interactions. Although with careful prompt manipulation, one can create immaterial abstractness, formal purity, whiteness, and flatness, a more fascinating mode of representation is the artificial image demonstrating the effects of time, tactility, and human occupation. These qualities are inherent to the datasets within midjourney, and can be further drawn out with precise language. Collage is not only an end product but also an action and process; time is a critical component of the act as evidenced in artificial image generation. Episodes of artificial creation deploy time as an agent to shape the image. Each image spawns inside the Midjourney platform, demonstrating a serial and episodic procedure reminiscent of collage.
2. Pallasmaa, Juhani. “Matter, Hapticity and Time Material Imagination and the Voice of Matter.” Building Material, no. 20 (2016): 171–89.
Tangible Imperfections
Raw images co-authored with artificial intelligence, without much stylistic intervention, exhibit uncanny warps and defects that renegotiate our understanding of architectural space. Materials are articulated imperfectly, with blemishes, defects, and stains that characterize an image which cannot be achieved through conventional rendering software. The uncanniness captured in these representational artifacts provoke new mysterious, curious, or mischievous readings of space. Using language that exudes qualities of ferality and frailty bear convincing images, fraught with intensified and defamiliarized signs of life. Patterns of interiority emerge in these raw images which allow us to realize an architecture exuding patina and growth, and is suggestive of inhabitation. This flavor of interiority emerges within intimate, human-centric, sensuous, and private demonstrations of fantastical, architectural space. This representational technique creates fictional spaces that allow viewers to inhabit the images, becoming fully immersed within these imaginary spaces, much like absorbing into an imaginary tale.
Auto-Prompt
Surrealists employ the automatic process in their work to tap into the subconscious mind, which they consider the source of all truly creative thinking. Prompt engineering with the machine learning model is reminiscent of surrealist procedures: automatic iteration through the subconscious mind and critical refinement using the conscious mind. This prompt engineering is the same method of language enhancement that designers use in order to refine their own special language, the architectural drawing. In this way, prompt engineering and architectural drawing share in common the process of iteration and critical refinement. The other important consideration here is that ideas not fully worked out have the opportunity to be nurtured through a collaboration with the AI bots. One can coax out the desired qualities in the final image by making changes, introducing new language, and refining the prompt. Tapping into the unconscious allows the process of becoming an image to become interiorized, a collaboration that allows one to nurture their ideas until they grow and develop into fully fledged concepts.
The Value of the Surreal
Our innermost fantasies are revealed in the surrealist exercise of describing a building into existence, allowing our subconscious thoughts and feelings to guide the creative process one word at a time. As Jennifer Shields writes in Collage and Architecture, “…surrealists sought to unify their inner world of imagination with the outer world of reality, a synthesis termed ‘sureality’ by Surrealism’s founder, Andre Breton.”1 Surrealists believed the rational, conscious mind repressed the power of the imagination. According to Thomas Mical, who cited Jameson, surrealism has the power to pierce our stifling and mechanical habits of behavior and reintroduce us to the world’s existential vitality and terror.3 He posits the misalignments of form and content otherwise known as defamiliarization, the process through which the familiar becomes unfamiliar, restores a sense of life and curiosity. Like surrealism, artificially generated images can juxtapose content that appear in tension to provoke new understanding and experiences. Furry buildings, chairs made of candy corn, interiors sculpted from paper- never before have we had the agency to explore so freely discrepancies of content and form, breaking free of the conventions and concerns that define contemporary theory and practice, and bringing dreams to life that we might not have had the courage to externalize.
1. Shields, Jennifer. Collage and Architecture. New York: Routledge, 2014, ix
3. Mical, Thomas, ed. Surrealism and Architecture. London: Taylor & Francis Group, 2004, 4
Fantasy Architecture
The relationship between surrealism and dreams have long been discussed but a fascinating addition has been thrown into the conversation with the emergence of artificial neural networks. Neuroscientists such as Erik Hoel propose dream theories can benefit from research related to artificial neural networks and specifically the concept of overfitting, the idea that a neural network stops being generalizable when it learns too explicitly.4 In other words, learning from fixed datasets inevitably will lead to overfitting. Ways to counter this phenomenon include dropping out information, warping and corrupting data, or feeding the network its own outputs. Hoel proposes that dreams serve as the biological equivalent of a combination of these methods, a type of abstraction or generalization that takes place after the day’s learning. The goal, in dreaming, is not to enforce the day’s memories but rather to counteract the negative effects of their memorization. Hoel goes on to describe that allure of fiction and art is that they artificially externalize dreams which prevent the overfitting of our perceptions, representations, and conceptualizations of the world. It is through this dream-like abstraction, that artificial intelligence can produce fictional novelty, nurture our deepest architectural fantasies, and shake up our comforting preconceptions surrounding architecture. We share in common with AI the foundations of deep learning: AI learns better when information is abstracted or generalized through various counter-overfitting methods, in other words when the model is dreaming; Humans learn better when we do not rely solely on the blueprint of our daily life, but also the abstraction of ideas, the warping of information, and misalignments that emerge while we are also dreaming.
4. Hoel E. The overfitted brain: Dreams evolved to assist generalization. Patterns (N Y). 2021, 3-11
Concluding Thoughts
Thus, learning and creation are intrinsically tied, both naturally and artificially. Reciprocity between learning and creation is enhanced by artificial intelligence, a space to learn while creating and to create while learning. Questions of authorship arise within the discussion of artificial intelligence and examining the roles and relationships of humans and AI can help answer some of these questions. Collaborating with artificial intelligence allows users to develop and nurture their ideas before externalizing concepts to the public. The private nature of this collaboration is what is fascinating, the relationship between human and AI, where the AI becomes a sounding board and mirror to augment the human’s imagination, becoming their greatest champion. What makes an image valuable, interesting, and critical is defined by the human and our ability to stand back, scrutinize and evaluate the collage from our own subjectivity. That is the power of the individual, the power that subjectivity carries within our discipline, and the power that theories of interiority rely on. Artificially generated images complement and advance ideas of subjectivity and interiority, privileging and empowering the individual to not only iterate and create within a supportive collaboration, but ultimately take part in the experience of these artificial images as an active participant.