Why are museums almost absent from our debates about AI?
In a new editorial for AI & Society, I argue that this silence is a mistake. Museums are not catalogues of objects in storage—they are active epistemological agents where knowledge is created, contested, and circulated. That makes them one of the few institutions positioned to challenge, rather than absorb, the dominant ideology of contemporary AI.
The warning sits in their own history. The Pitt Rivers and the Petrie were built through colonial extraction and organised typologically—every object flattened and pinned to a label. Kate Crawford and Trevor Paglen describe the same impulse in ImageNet: taxidermy butterflies in a display case. When Pereira and Moreschi fed the Van Abbemuseum collection to commercial vision systems, Duchamp’s Fountain came back as «plumbing fixture» and 90% of the artworks collapsed into IKEA-catalogue categories. Contemporary AI inherits the nineteenth-century drive to classify, rank, and commodify—an epistemology that leaves no room for ambiguity.
What the industry calls «generativity» is, strictly, statistical resampling—extraction, not creation—within a fixed distribution. The museum works the other way. Ten visitors generate ten irreducibly different paths through a space.
As the conversation shifts from generative to agentic AI, the question of whose knowledge is being produced—and on what authority—only becomes sharper. Museums have spent centuries curating exactly that: careful reasoning, disputed authority, the coordination of competing knowledge claims.
It’s time they stopped being passive actors in this transformation.