John Hejduk Soundings Lecture: Mario Carpo, “Generative AI, Imitation, Style, and the Eternal Return of Precedent”

Generative AI does not create new images out of thin air; it generates images that have a “certain something” in common with a selection of images we have fed into it.  This selection, often called a “dataset,” can be generic or custom-made; either way, Generative AI automates the imitation and replication of some of its common visual features, often known in the past as styles.  Imitation was for centuries the backbone of the classical tradition in European art, and it was de facto banned by 20th-century modernism for many good reasons.  As the rise of Generative AI is bringing the practice of imitation back to our design schools and to the design professions, we urgently need to learn again what imitation is, how it works, what it does, and how we can deal with it today, in critical and creative terms.  Every dataset is a canon, but every reference to precedent is based on preference, and we know all too well that preference is often a proxy for prejudice.