We normally associate the jobs at potential threat from AI over coming years as those that involve repetitive, standardised tasks that work to clearly defined rules. But recent advances in machine learning algorithms mean even jobs that require broad knowledge and understanding of a particular area and wouldn’t immediately strike as fitting the above description, could come under threat.
One interesting recent example is an AI developed by researchers that is able to do at least part of the work that is the responsibility of museum and gallery curators. The algorithm is able to detect subtle links between works of art that could offer clues as to who the artist behind pieces of unknown or unconfirmed origin is. The process could also be used to select combinations of works of art that would fit well together in an exhibition.
The algorithm can quickly sift through visual databases of millions of paintings and sculptures and find connecting themes and motifs that group them together. The research project’s lead Mark Hamilton had the idea for the potential of applying AI to artwork curation while visiting the Rijksmuseum in Amsterdam.
While there, similarities between the poses, colours and textures used in two paintings, The Martyrdom of Saint Serapion, by the 17th-century Spanish painter Francisco de Zurbarán, and The Threatened Swan by the Dutch artist Jan Asselijn, struck him. While the two artists had never met nor had any known communication, Mr Hamilton noticed “a rich, latent structure underlies both of their works”.
A PhD student at the Massachusetts Institute of Technology (MIT), Mr Hamilton was inspired to embark upon a project that involved his team at MIT and collaboration with Microsoft. The created an algorithm designed to spot connections between an image and a defined culture or media. The tool is called MosAIc, and is already being experimentally used to look for connecting features between artworks in the databases of New York’s Metropolitan Museum of Art and the Rijksmuseum where the idea was born.
Some interesting findings have already been uncovered by MosAIc. The tool connected an 18th-century sketch of a blue crane made by Dutch artist Robert Jacob Gordon with a 16th century Persian vase shaped like a bird and painting of the bird-headed Egyptian god Seth, created in blue and dating from 521BC.
The AI is adapted from a common algorithm called “K nearest neighbours”, which is used to group together objects or images based on similarity. The MIT team tweaked the algo to group images in a ‘tree-like’ structure of connecting similarities.
From the trunk, which is general similarity, the AI applies layers of new filters such as colour, context and texture, following a chosen branch until it finds the most ‘linked’ images across a database. That could be all images with a bird theme, or more subtly, that use similar textures.
Mr Hamilton explains:
“These fields are rich with information that has never been processed with these techniques and can be a source for great inspiration. This can be expanded in new ways to understand the connections between works.”
Despite the potential of MosAIc, it seems traditional human curators are not yet overly concerned about being replaced by software. Rosalind McKever, a curator at the Victoria & Albert Museum, thinks the tool could help make a curator’s job more efficient. And it could be used to throw up interesting links between works that may not otherwise have been spotted. But ultimately falls well short of what a curator contributes to a collection. She comments:
“As interesting as it is for algorithms to find connections, if you want to know why the connection exists, or what it means, you need an art historian or curator. One of our roles is researching, questioning and communicating the stories behind the connections. And we are always looking at the bigger picture, beyond the collections of even the largest museums.”