The use of generative AI models increasingly involves the reliance on a few black box pretrained and centralized models where design intent is conveyed through language prompts. However, language is an imperfect medium for conveying design and artistic intent especially in the visual arts and architecture.

Underpinning modern AI is a multitude of techniques borrowed from data science, digital signal processing, information theory, computer graphics …. The machinery of Machine Learning and information theory provides a unified framework for representing, analysing and synthesizing data, regardless of whether these data map to images, sound, 3d models or something else.

In this course we will explore the creative potential of Machine Learning and Signal Analysis techniques when we treat the environment and the things we create as signals. This year we will focus on the relationship between form and structure in the context of design and architecture using a mix of techniques borrowed from machine learning, computer vision and structural analysis
The architecture historian Eduard Sekler coined the term “atectonic” to describe structures whose form is not driven or expressive of their materiality and especially their structural behaviour (the interplay of loads and supports). By this definition the terms tectonic/atectonic are aesthetics categories that describe the degree of alignment between the underlying structure and its expression.
This tension between the material support and its envelope carries echoes of a divorce that happened in the 19th century when architecture and engineering formally separated. This lingering tension contains a multitude of dipoles, structure|form, skeleton|skin, engineering|architecture, convergent|divergent, deterministic|ambiguous, reductionist|wholistic, physical|percetual. It manifests itself in theory, pedagogy and technology.

In one particular sense with the introduction of computational techniques in design a unified space opened in which engineering processes through the various simulation and structural optimization tools can be integrated into design workflows. Many of these methods depend on a very systematic approach starting from first principles of the mechanics of materials and building the complex finite element analysis algorithms that are commonplace nowadays. This has been going on for decades now. More recently the development in AI and Machine Learning introduced a new set of tools that can handle less deterministic aspects of design. Perceptual and aesthetic, questions of expression can be to an extend also modelled within these digital design processes. In addition, these new techniques enable the creation of unified models that can integrate both sets of constraints simultaneously through approaches like Physics informed neural networks.

In this class we will visit and combine tools opportunistically from both domains to explore methods for giving expression to material systems. Through a series of workshops students will be introduced to tools that combine Machine Perception, Structural Analysis and Computer Vision to design a physical object.

Students will work on a bigger semester long project that consists in designing a structural system that derives its aesthetic expression from its mechanical / material properties infused with some external aesthetic constraints. This final design could be a single building component (beam, column, branching column, truss) a larger space (hall of columns, shell structure, bridge) or a functional object (chair, table).

No prior scripting knowledge is necessary though a basic knowledge of rhino + grasshopper would be useful.

The semester will unfold as a series of workshops cantered around custom tools , each one exploring one relationship between form and structure or a specific technique.