On Vector Agency

This chapter discusses the challenge generative AI poses to traditional notions of agency and authorship. It argues that agency is not a binary attribute but a distributed and gradual phenomenon emerging dynamically from the interaction of users, algorithms, data, and contingency. Platforms like Midjourney, DALL·E, and Stable Diffusion function as sedimented cultural archives, crystallizing prior creative intentions within their training corpora. Agency is neither exclusively assigned to the machine nor monopolized by the user; rather, it is dynamically computed within the latent space with each image rendering. The concept of vector agency is introduced to analyze how this distributed control operates. This perspective emphasizes that the trained model's latent space embeds vector-like tendencies, directional forces with magnitude derived from cultural data and design choices, which steer outputs. Interaction with generative systems is seen as a continuous reconfiguration of these internal forces across a spectrum of user intervention, from simple prompting and curation to advanced model fine-tuning. Moreover, the chapter maintains a crucial distinction: agency can be attributed to non-human components, but authorship remains a legal concept tied exclusively to human intent and personhood. This distributed nature, amplified by bounded unpredictability, highlights the iterative negotiation within the human-machine ensemble.