Course Structure
A hands-on workshop to deploy AI capability in modelling a series of houses that offer increasing complexity in their geometric constraint. We aim to discern new speed, accuracy and design agency for both architectural design and its construction, respectful of the history of disciplinary norms but aiming to exceed current design>build aptitudes. The 2 professors have developed design>build capability through 3-4 years of AI-oriented software development, supported by technically skilled TA(s). Powerful computational servers + API offline usage will be provided by MIT’s central Office of Research Computing and Data (so no issues with token limitation).
Prior Experience / Computational Skill / Equipment
Beyond everyday laptop habitude, none is required, our increasing belief being that architecture students no longer need be concerned by a lack of software expertise: through generative AI, processes such as parametric or relational modelling (typified by Grasshopper) are quite simply engaged by anyone with clarity of thought (that’s key!) Programming, coding, even drawing! can be by-passed by direct use of natural language, textually or even vocally(!), even gesture becoming instrumental in precisely defining sophisticated architectural CAD-CAM description.
* Where David Sedaris’ hilarious, Me Talk Pretty One Day describes his cleverly evasive strategies for (not really) mastering French, we will attempt to write (or speak!) “a house” (actually, houses). But our mischief will from the outset aim to deliver hyper-precise architectural drawings and construction documents. If this seems surprising, or even demoralizing, we’ll reflect back to the origins of architectural description long before the widespread adoption of drawing in the Renaissance, itself ushered in under influence of the new technologies of surveying, cartography and perspective (plus the availability of paper, as the printing press — another technical apparatus — took hold of intellection and imagination). This implies that architects will not so much design a house, as define house-ness, or the possibility of house.
We've worked with decently powerful laptops, so with MIT CPUs, we assume they'll be adequate.
Course Description
We aim for you each to generate 3 houses:
- We’ll start with Henry Thoreau’s simple hut at Walden Pond, a modest retreat-for-his-soul, born of his wariness of industrial technology in stripping us of contemplative well-being. As we aim to cerebrally replicate his implicitly industrial(!) dwelling, seen either as skilling-up or losing-our-minds, the Walden Paramorph offers itself as historical counterpoint per uptake of this latest technical apparatus. The goal will be to set up a symbiotic generative AI logic that can automatically optimize the design + fabrication of Walden AI through multiple iterations.
- A next target will aim to parametrically AI-model DOMhouse, which is offered as a formally simple rectilinear mono-pitch 2-bedroom summer house. But implicit is a high degree of internal constraint, starting to raise questions as to how to balance between many competing design influences. The goal here will be to bypass Grasshopper and build a sophisticated parametric model inside Rhino without manually modeling a single line! This will aim at a sophisticated design interface operable by anyone, to be able to hone a nuanced house, albeit imbued with your implicit architectural qualities. Both dumb and sophisticated, it will announce capability for any architect to effectively create design software (Rhino only for visualization!)
We will attempt to attain full design-build (fabrication-ready) output as an Dessent ially automated AI-assisted logic, but this will require that base paramorphs are essentially perfect to permit their serving as progenitors for quite advanced AI algorithmic capability. In other words, deploying electrical and plumbing optimization routines will require that your base geometric building envelopes are highly disciplined: the generative system has to be conceptually pure.
We've attained step-by-step fabrication files that offerremarkable speed/dexterity IF and WHEN our base geometry is fastidious: this will be one essential measure of success. - We will then offer a choice to each student:
- We can provide a yet-more-complex house as a learning-curve in non-orthogonal geometric constraint, offering the benefit of our comprehensive understanding of it’s implicit rule-sets (which are essential to defining it as logically computable).
- Students can choose an historic exemplar to try to independently conceive of the seemingly essential “cascading schemas” and bi-directional constraints, interpreting the design prerogatives of le Corbusier, say, to witness a new level of relational understanding (NB formally-simple houses would seem to us prudent in this early effort at probing not only AI systems, but our own capabilities for system-thinking buildings)
- Students would elect to apply their AI-learning to one of their own projects, looking to attain disciplined generative capability, ideally able to be parametrically nuanced.
Schedule
Lecture 1: M 2-4
Lecture 2: R 9-11
Location
Lecture 1: W41-1101
Lecture 2: W41-2319
Can Be Repeated for Credit
Yes