Any music fan knows that there are myriad ways to find new songs online: a scroll through digital playlists and streaming radio services like Pandora, which serve as musical recommendation engines. Likewise, Netflix subscribers are regularly showered with suggestions for, say, romantic comedies and horror films, based on previously viewed movies.
But until now, there was no automated guidance for art lovers seeking discoveries online - no “If you like Jackson Pollock's ‘No. 1,' you may also enjoy Mark Rothko's ‘No. 18.' â€
Enter Art.sy, a start-up whose public version went live on Monday. An extensive free repository of fine-art images and an online art appreciation guide, it is predicated on the idea that audiences comfortable with image-driven Web sites like Tumblr and Pinterest are now primed to spend hours browsing through canvases and sculpture on their monitors and tablets, especially with one-click help.
After two years of private testing and with millions of dollars from investors, including some celebrities in the art and technology worlds, the site aims to do for visual art what Pandora did for music and Netflix for film: become a source of discovery, pleasure and education.
With 275 galleries and 50 museums and institutions as partners, Art.sy has already digitized 20,000 images into its reference system, which it calls the Art Genome Project. But as it extends the platform's reach, Art.sy also raises questions about how (or if) digital analytics should be applied to visual art. Can algorithms help explain art?
Robert Storr, dean of the Yale University School of Art, has his doubts. “It depends so much on the information, who's doing the selection, what the criteria are, and what the cultural assumptions behind those criteria are,†Mr. Storr, a former curator of painting and sculpture at the Museum of Modern Art, said. In terms of art comprehension, he added, “I'm sure it will be reductive.â€
The technology, at least, is expansive. To make suggestions successfully, computers must be taught expert human judgment, a process that starts with labeling: give a machine codes to tell the difference between a Renaissance portrait and a Modernist drip painting, say, and then it can sort through endless works, making comparisons and drawing connections.
For the Art Genome Project, Matthew Israel, 34, who holds a Ph.D. in art and archaeology from the Institute of Fine Arts at New York University, leads a team of a dozen art historians who decide what those codes are and how they should be applied. Some labels (Art.sy calls them “genes†and recognizes about 800 of them, with more added daily) denote fairly objective qualities, like the historical period and region the work comes from and whether it is figurative or abstract, or belongs in an established category like Cubism, Flemish portraiture or photography.
Other labels are highly subjective, even quirky; for contemporary art, for example, Art.sy's curators might attach terms like “globalization†and “culture critique†to give ideological context. “Contemporary traces of memory†is an elastic theme assigned to pieces by the Chinese Conceptual artist Cai Guo-Qiang and the photographer and filmmaker Matt Saunders.
A Picasso might be tagged with “Cubism,†“abstract painting,†“Spain,†“France†and “love,†all terms that are visible and searchable on the site. Jackson Pollock's works typically get “abstract art,†“New York School,†“splattered/dripped,†“repetition†and “process-oriented.†Predictably, some of those criteria show up on paintings by Pollock's contemporaries Robert Motherwell and Willem de Kooning, but also on artists from different eras and styles, like Tara Donovan, whose contemporary abstract sculptures using stacked and layered plastic foam and paper plates have also been marked with “repetition.â€
As the categories are applied, each is assigned a value between 1 and 100: an Andy Warhol might rate high on the Pop Art scale, while a post-Warholian could rank differently, depending on influences. Software can help filter images for basic visual qualities like color, but the soul of the judgment is human.
“Literally, a person goes in by hand, and they enter a number for all the relevant fields,†Mr. Israel said.
The technical complexity is outweighed by the curatorial challenges. “We learned that the data matters much more than the math,†said Daniel Doubrovkine, 35, who is in charge of engineering at Art.sy. “How are you going to pick something that shows ‘warmth' with a machine? We're not.â€
Similarly, Pandora has a roomful of musicologists deconstructing each tune; their analysis is then fed into an algorithm, called the Music Genome Project, that recommends songs in its player based on users' taste and the ratings they give each track. (Joe Kennedy, the chief executive of Pandora, served as a consultant to Art.sy.)
But Art.sy aims to make connections among artworks that are seemingly from different worlds, with a catalog that encompasses pieces from the British Museum, the National Gallery in Washington, the Los Angeles Museum of Contemporary Art and others. A recent partner, the Cooper-Hewitt, National Design Museum in Manhattan, a branch of the Smithsonian, has added objects to the mix, which will be a test of the site's technology and the parallels it draws, said Seb Chan, the Cooper-Hewitt's director of digital and emerging media.
Culturally, “what does it mean to recommend a painting from seeing a seventh-century spoon, for example?†he said. Anticipating such questions, the Art.sy staff has a blog explaining how its process works.
