This course aims to introduce students to concepts and techniques from Machine Learning and Computer Vision as a way to revisit questions of perception and aesthetics within the context of an AI-mediated world and its implications for creative work.

As machines are increasingly used not only to generate images but also to censor, curate, filter, organize and contextualize what can be seen (the distribution of the sensible, as Jacques Rancière would say), these indefatigable systems that are able to make aesthetic judgments for millions of images per second increasingly mediate our relation to culture and cultural artifacts — not to mention their political and social implications.

In this introductory course we take the position that in order to understand, engage with and critique the generative capabilities of AI models, we need first to understand the underlying processes and especially the perceptual models that guide any type of generative process. In addition, we need some fundamental concepts and ways of thinking: what an embedding space is, and how the mathematical structure of a vector space maps into semantic relations in a symbolic order, or into perceptual affinities.

In the larger context of computational design there are also some new ways of thinking, like the transition from procedural algorithmic programming to training a model as a sort of programming by example. Questions of culture and aesthetics can also now be incorporated into computational design workflows, where machines can act as surrogates for human perception and aesthetic judgment.

Rather than focusing on large language-based generative AI models, we are going to explore the design potential and implications of some of their constituent components and the software frameworks that enable them. Through a series of workshops and small projects, students should develop an intuitive understanding of how model architecture, dataset curation, training and inference work, and what the opportunities are for injecting creative intent beyond the use of language and prompt manipulation. The emphasis will be placed more on the perceptual capabilities and idiosyncrasies of ML models, with some forays into proto-generative processes.

We will start with simple language embedding models to discuss the structure and operations on vector spaces that underlie most ML applications. We will later introduce the classifier and autoencoder models as two archetypes of artificial perception and building blocks of other models. We will also visit RankNet as a simple model that can be trained to act as a surrogate for individual taste. Through a series of targeted projects, students will train and deploy these models to use them as surrogate perceptual systems that can curate, filter, organize and ultimately modify visual content.

Note regarding the Fall 2026 GSD academic calendar: The first day of classes, Wednesday, September 2nd, is held as a MONDAY schedule at the GSD. This course will meet for the first time on Wednesday, September 2nd.