Lately I’ve been asking myself: what might artificial intelligence be good for besides answering questions and writing code?

I think the answer is: building a new medium. A medium is any realm that can be used for expression — paint, film, language, music, architecture. I propose that latent space, the high‑dimensional representational space created inside machine learning models, is a new medium.

Latent space is a place where ideas, styles, and forms can be blended, interpolated, and explored. It is operationally different from previous media because it encodes enormous amounts of variation compactly and allows smooth transitions between concepts. This lets creators perform operations that were difficult or impossible before: morphing genres, combining disparate aesthetics, and discovering unforeseen hybrids.

Working in latent space requires new tools and literacies. Instead of mastering a physical technique (brushwork, carving), creators learn to navigate vector fields, compose prompts, and manipulate model weights and embeddings. Collaboration between humans and models becomes a dialog in this space. The artist’s role shifts toward curating, steering, and remixing the possibilities latent models expose.

There are also cultural and ethical questions: proprietary model architectures centralize control over the medium; biases and limits in training data shape what is expressible; authorship and originality are redefined when works derive from learned distributions. Yet la