Link to article

There is no such thing as a new idea. It is impossible. We simply take a lot of old ideas and put them into a sort of mental kaleidoscope. We give them a turn, and they make new and curious combinations. We keep on turning and making new combinations indefinitely. but they are the same old pieces of coloured glass that have been in use through all the ages.

Mark Twain

This chapter delves into a design workflow that embraces the concept of “Harmony in Co-Creation.” Exploring the profound implications of and relationships between quasi-autonomous design and non-autonomous design provides insight into the future of architectural practice. By employing a multidisciplinary approach, the chapter introduces two workflows. The first workflow integrates artificial intelligence (AI) algorithms in the form of search engines and diffusion models to elucidate and enhance students’ understanding of site dynamics and design potential. Meanwhile, the second workflow employs a digital collage methodology, using manually discovered and collected data and a hand-crafted approach to ideation, representation, development, and iteration. The chapter explores projects undertaken in densely populated metropolitan contexts to test these two approaches. Concluding with reflections on the innovative methodologies embedded in the workflow and their potential impact on architectural education, this chapter emphasises the fusion of digital technologies and the representation of physical experiences in shaping innovative architectural solutions. “Harmony in Co-Creation” encourages readers to contemplate the profound implications of quasi-autonomous and non-autonomous design strategies as a natural continuation of foundational methodologies and past technological introductions rather than as an existential threat to the profession.

1 Introduction

The concepts, strategies, and terms of computer science systems originate in architecture. Now, following technological advancements, architecture is finding value in importing these systems back, and, due to the similarities between the fields, it is proving easy and natural for architects to do so.

It is reasonable to expect that architecture’s real-world explorations and advancements will continue to influence thought and strategies in digital systems in a continued symbiotic relationship between the professions. The collective imagination around futuristic spaces depicted in film and advertisements more often than not utilizes architecture that already exists. Architecture contributes its own cutting edge to the collective cultural momentum largely because of its design process.

Architecture’s slower evolution over millennia disciplines the rapid advances of digitalization, though experts in computer engineering remind us that critical advancements are slower than the public perceives. One would imagine that this history would protect architecture from the hype cycle currently attributed to seemingly easily adopted digital advancements, like generative AI, that some speculate point to the demise of the architecture profession. A more measured response gives the opportunity to reexamine which elements fall within the architect’s scope. The prioritization of these built and systemic responses is conditional and context-dependent and is largely determined by the architect. A short list of these issues includes climatic response, material behavior, spatial programming and planning, and the ability to inventory and prioritize such elements experientially in ways that combine to yield varying degrees of human comfort, resource and energy efficiency, and forms and experiences with varying degrees of success.

AI’s strengths aid architecture’s understanding of context, long-term planning, and human experience. Rather than replacing architects, AI can be viewed as an “instrument” or “collaborative partner” that expands the space of possibilities. It offers the ability to analyze vast data inputs and generate design options that architects can evaluate and adapt based on more subjective human knowledge. The future of architecture lies not in replacing its core practices but in a dynamic partnership with AI technologies, where the measured pace of architectural evolution meets the speed and scale of digital advancement to redefine the creative process.

Since Alan Turing set the stage 70 years ago, AI has developed in two directions: the first focuses on reducing human intelligence to symbolic manipulation, while the second, more widely used approach develops mathematical models that detect data patterns and enable automatic learning.

Current advancements in AI are centered around generative algorithms, particularly diffusion models. Figure 2 illustrates a series of automatically generated images. These images were produced entirely by AI algorithms trained on vast datasets. We are generating an overwhelming amount of visual content, and as image and text generation become faster and more automated, it becomes cheaper and more accessible.

Diffusion models create complex content by diffusing information across neural network layers. Each step takes a latent array, or noise, and produces another latent array that looks more like the text input and all the visual information on which the model was trained. The arrays are real numbers that correspond abstractly to the objects in the image, essentially digital data forms of the image and the text descriptions. Combining all the images encoded as arrays creates a latent space ripe with potential for new imagery; we will call this space the space of possibilities. Since we have trained the model with images and their corresponding text descriptions, this space of possibilities can generate similar images that contain objects previously identified in our world, e.g., an aeroplane or a cat (Fig. 3). Let’s say we want to generate an image of a tree; this new image is generated by altering the numerical array in the space of possibilities labelled as an aeroplane. Understanding how the space of possibilities works motivated several platforms to allow users to use predefined settings that control or modify the array of numbers representing a particular object or combination of objects.

These settings control the influence that a text or image prompt has on the final output. This sense of control is addressed later in this chapter because this model has been used only to generate images aligned with the human perception of the world.

This newly trained diffusion model can be considered human-aligned since it is trained on all the objects we know. We narrow the possibilities to a defined set of constraints based on labels. It becomes intriguing to explore the “in-between” space—the space between our labels and concepts—leveraging the full spectrum of possibilities the model offers (Fig. 4).

In Fig. 4, the central images represent the most accurate and accepted interpretations of the text query, while a spectrum of intermediate representations radiates outward. For example, when we prompt the diffusion model with “A city with timeless connectivity”—a somewhat abstract concept—the generated outputs attempt to balance elements like dense urban fabric, watch-like, time-related patterns, pavement materials, and alleys. We are given the opportunity to explore how a more atmospheric understanding would look by examining the abstract blended images surrounding the initial concept. This exploration offers an opportunity to push a design concept forward by examining and iterating upon these intermediary images with more isolated interpretations.

This approach to representation can be contrasted with the most stable one, perspective drawing, which integrates geometry and spatial logic to communicate experiential architectural designs and played a pivotal role in the Renaissance’s intellectual and artistic revolution.

What is understood as perspective was defined by Vitruvius and is considered a particular projection of architectural work through the eye of the architect. We can see a parallel here with the way the diffusion model can be masterfully articulated; perspective drawing is a representational instrument that combines floor plans and facades into one. It too necessarily includes latent space and is similar to the way generative algorithms are used today. Perspective drawing was also used as an instrument in design, as it brought architects into closer collaboration with artists, mathematicians, and engineers, enriching architectural theory and practice.

1 Worldmaking

Now that we have introduced the space of possibilities, let’s explore the aim of the generative algorithms, which is to map the data points we generate within a normally distributed curve (Fig. 5). The center of the curve is our past data. At the boundaries, we encounter outputs that cannot typically be categorized and that represent exceptional instances of creativity. To produce innovative work, we must focus on representations of in-between concepts and on how they relate to architectural space and the projection of creativity. A collaborative effort between machines and human intelligence allows this to be achieved efficiently. By combining AI’s ability to process and explore vast datasets of known information with human creativity, training, and intuition, architects and designers can make better and more widely informed decisions, generate novel ideas, and realize designs and spaces that were once unimaginable.

2 Tool and Instrument

AI can be a practical tool, like a hammer or lever, to address problems or tasks such as playing chess, image recognition, or text generation. However, it is also essential to consider AI as an instrument comparable to a pen, paper, or violin. The primary distinction between a tool and an instrument lies in their interaction with the user's creativity. That is, tools such as a lever or a motor, and instruments such as a typewriter or a computer, can be easily distinguished by their use. The instruments themselves have no particular use and are so generic that they require human interaction to articulate a result. In contrast, a tool is created for a particular and specific use and is defined for a certain action.

In architecture and generative design, using AI as a “tool” is typically more deterministic, providing immediate solutions based on defined parameters. On the other hand, AI as an “instrument” fosters collaboration, integrating human inputs and interpretations to enhance and expand the design process. This adaptability allows AI to shift roles depending on the stage of the project, specific needs, or creative objectives, making it a versatile asset in design workflows.

The reaction to AI as the death of the architecture profession echoes the emergence of BIM, CAD, digital modelling, digital image manipulation, and printing. AI’s apparent ability to develop and represent all conceivable design solutions for evaluation by a project team, and its ability to recombine existing and known images, styles, documents, systems, relationships, and conditions for a new context, is better understood as the latest iteration of the design process, deriving from strategies that have been in practice for more than 100 years.

Ludwig Mies van der Rohe pioneered manual photo collage as a means of exploring and representing design proposals. Historically, firms with more resources could create numerous design iterations, giving them a significant advantage. Famously, Eero Saarinen’s office is reported to have manually drawn through all imaginable combinations of programmatic elements for some proposals in order to find success. More recently, Richard Meier and Partners, now Meier Partners, hand-collaged recombinations of elements from the firm’s own expansive catalogue of projects and honed the final relationships between the parts to create highly detailed and renowned designs.

Instead of eliminating the architecture profession, AI development is an opportunity for more high-quality design solutions to be available to communities with a broader spectrum of budgets. These AI tools can lower the resource and time requirements that come with generating many versions of designs and have the potential to democratize great design. The introduction of more options and a reduced cost of entry is one step toward levelling the playing field and expanding the profession’s opportunity set.

AI becomes a powerful instrument in the hands of the knowledgeable. Skilled teams can translate the spaces illustrated in AI’s visual outputs into robust technical solutions that deliver on the suggested experiences. As calculable and predictable conditions are programmed into more project options, architects can offer greater accountability and prioritize the variables they choose. This expands the opportunities to identify and implement the best solutions for a design problem. Indeed, AI also enables inelegant and unknowledgeable responses, but that is neither new nor surprising to the profession or the construction industry.

4 Exploration

4.1 Physical Collage as an Instrument in Collaborative Design

Ludwig Mies van der Rohe’s collaboration with Peter Behrens in a magazine called G: Material zur Elementaren Gestaltung (G: Materials for Elemental Form Creation) laid the foundations of the modernist collage representation process.

The architecture profession has constantly experienced technological advances that lower the amount of time it takes to represent the intent of spaces. For example, intaglio copperplate printing helped distribute and increase the reach of previous hand-drawing techniques.

Photocollage emerged as another technological advancement in architecture, making the representation of design intent an order of operations and much faster. These new media, like collage and eventually printed drawings, were still bound by their own skill requirements, accessibility, and production timelines but sped up output time. Print media’s limitations reflected the modular building techniques of the Industrial Revolution and led to increasingly realistic depictions of space or space-making elements.

Many of these now custom-built techniques continue to be used by designers who want their drawings to be literally assembled like the construction of a building. Some schools continue teaching this process to students so that they can coordinate their decisions, the impacts of those decisions, and their order of operations. This process of learning to think is valuable training for imagining work and the processes that will be made real by the building construction process (Fig. 6a, b).

Van der Rohe’s collage techniques introduced the idea that representation could be layered, iterative, and nonlinear—a process that encouraged combining diverse materials and ideas into cohesive spatial concepts.* Similarly, today’s AI diffusion models and clustering algorithms have redefined this process, allowing designers to explore the in-between space of concepts, generating collages that are not limited to predefined outcomes but instead emerge from exploring potential relationships between form, materiality, and context (Fig. 7).

AI’s ability to explore many possibilities mirrors the intent of early modernist collages but with incredible speed, flexibility, and depth. It is a new form of collage-making in which the user directs rapid combinations and interpretations of images in a collaborative partnership. Looking at these various levels of image possibilities provides opportunities for architects to engage with unexpected design outcomes, atmospheric representations, and moods. The collages in Fig. 8 showcase some of the results of an exploration that tried to blend spatial, material, and contextual elements in ways that traditional collage techniques could only hint at.

4.2 AI as an Instrument in Collaborative Design

At the beginning of this chapter, we talked about the tendency for generative AI to lean toward the center of the curve; however, knowing that we want to distance ourselves from the center of the curve, which means that we want to approximate the distribution’s tails, it is necessary to think of AI as being able to point in a particular direction. Still, the architect ultimately decides what direction to take the project, but we should not consider AI to be under command and control; rather, it should be viewed as an active collaborator helping humans explore. Let’s use AI as a tool or instrument so that it becomes an active collaborator that aids creative work, utilizing its capabilities for discovery and our capabilities for imagination and creation.

To demonstrate our approach, we collected a diverse dataset of over 5,000 images from social media related to our research topic, the housing system in New York City, using an AI-powered image crawler. Keywords such as “affordable,” “inclusive,” and “re-build” were selected to define the scope of our investigation. AI can be used as a tool to extract relevant images from platforms like Instagram by inputting keywords (Fig. 9, Top).

The collected images were then processed further using an AI feature extractor to identify key visuals that would inform and enrich the design exploration (Fig. 8, Centre Row). The images were categorised through SOM into clusters of similar colour schemes and characteristics. These categorised images were then assembled into “atmosphere collages” (Fig. 9, Bottom), designed to explore various moods and environments that could shape the foundation of a design concept. This approach heightens engagement with architectural concepts by synthesising vast amounts of imagery into coherent and navigable colour schemes. AI supports the creative process by drawing attention to subtle transitions and nuanced relationships within the design, inspiring new ideas and perspectives (Fig. 10).

AI’s adaptability highlights its versatility in architecture, from task-oriented functions (as a tool) to actively contributing to the creative ideation and conceptual development of design projects (Fig. 11).

4.3 Navigating the Dual Roles of AI in Architectural Design

When utilized as a tool, AI operates under the designer’s direction, optimizing processes, solving problems, and calculating parameters in a repetitive and goal-oriented manner. In contrast, when AI acts as an instrument, it becomes a collaborative partner, enhancing the design process by synthesizing ideas and generating innovative outcomes. As AI evolves, its role and relationship with humans will deepen. The question is not about choosing between AI as a tool or an instrument but about understanding how its dual nature can inspire, enhance, and prototype designs. By embracing this potential, architects and designers alike can unlock new creative dimensions, pushing the limits of what is possible and redefining the future of design.