The first thing visitors see at the new wing of the Kunsthalle is a wall label with no artwork under it. The label describes a painting in precise, loving detail: the brushwork, the light, the particular shade of ochre in the lower left corner. The painting does not exist.
It was generated by a model trained on the museum’s own collection, then deliberately not printed. The curators want you to look at the space where an image would be, and to notice how much of museum-going is an act of reading before it is an act of seeing.
Curation as a model
Across Europe and East Asia, institutions are experimenting with algorithmic curation: shows assembled by recommendation systems, audio guides written by language models, hangings optimized for visitor flow rather than chronology. The results are uneven, occasionally brilliant, and frequently uncanny.
A machine can tell you what is similar. It cannot tell you why the difference matters.
The most interesting work treats the algorithm as a collaborator with a visible hand, not an invisible author. When the system’s choices are annotated — “selected because it shares a palette with room 3” — visitors begin to argue with it. That argument, curators say, is the point.
