AI curates art shows

A Hong Kong collector just built a machine to do part of a curator’s job: look at a pile of artworks, find patterns, and help decide what should hang together in a show. Artnet reported on April 9 that Alan Lau has developed an artificial intelligence curatorial system for selecting and presenting art. (artnet.com) That is unusual because Lau is not a software vendor pitching museums from the outside. He is one of the people who already helps shape what major institutions collect, serving as vice-chair of M+ in Hong Kong and co-chairing Asia-focused acquisition groups for Tate and the Guggenheim. (artnet.com) In the art world, curation usually means a person makes a sequence of judgment calls that look subjective from the outside. One curator might group works by color, another by politics, and another by a single artist’s biography, even when they start from the same storage room. (amt-lab.org) Artificial intelligence changes that process by turning a collection into data the system can sort, compare, and recombine. Museums and researchers have been testing tools that read catalog records, provenance histories, and object descriptions to surface links a human might miss in a large database. (amt-lab.org) One of the clearest recent tests came at the Nasher Museum of Art at Duke University, which opened an exhibition on September 9, 2023 built largely with OpenAI’s ChatGPT. The museum team first found that the model could not reliably use the collection on its own, so they built a custom system with LangChain, Streamlit, and ChromaDB to connect the model to the museum database. (amt-lab.org) (museumnext.com) Even then, the machine did not replace the museum staff. The Nasher project still depended on curators and faculty to clean the data, frame the prompts, reject bad suggestions, and turn abstract themes like “utopia” and “dreams” into an exhibition of 21 actual objects. (museumnext.com) That is the likely shape of Lau’s system too: less robot tastemaker, more high-speed assistant. Artnet’s report says it is meant to help select and present artworks, which suggests decision support and display planning rather than a fully autonomous machine mounting shows by itself. (artnet.com) The timing fits Hong Kong especially well because Lau sits at the intersection of money, museums, and technology. Artnet described him in 2024 as a fintech executive and major patron, and other recent public bios describe him as a technology executive with long experience in internet and finance alongside his museum roles. (artnet.com) (coindesk.com) There is also a practical reason collectors would try this before many museums do. A private collector can test a new tool on a personal collection, move faster than a public institution, and avoid the board approvals, procurement rules, and public backlash that can slow a museum experiment. (amt-lab.org) (museumnext.com) The hard part is not getting a model to produce a neat wall label. The hard part is that curating also involves context, ownership history, cultural sensitivity, and the politics of why one work is shown beside another, which is why museum standards still treat provenance and verification as core human responsibilities. (amt-lab.org) So the real shift here is not that a machine has become the new museum genius. It is that one of Asia’s best-connected collectors is treating curation as something that can be partly systematized, tested, and optimized in public, and that usually means more experiments will follow. (artnet.com 1) (artnet.com 2)