EU shipping is temporarily suspended

Google, but for Gauguin: The internet, the future and a search engine for art

Posted on 08 Oct 2020

In a world of targeted ads and personalised recommendations, Matthew Israel reflects on building the ‘first large-scale search recommendation technology for art’ in this extract from ‘A Year in the Art World’.

Image: Olafur Eliasson, Your rainbow panorama, 2006-2011. ARoS Aarhus Kuntsmuseum, Denmark, 2011. Photo: Thilo Frank Studio Eliasson, courtesy ARoS Aarhus Kunstmuseum, Denmark. © Olafur Eliasson

Extracted from Matthew Israel’s A Year in the Art World.


Since the late 1990s, in parallel with so many aspects of the modern world moving online, there has been a consistent and well-funded effort to bring the art world in all its aspects – museums, galleries, auction houses, artists, buying and selling and education – there as well. Today, after a little more than two decades, the internet offers a plethora of ways to engage with art in the virtual world.

For example, you are now able to see images of millions of the world’s most famous – and most niche – artworks on sites like the Google Art Project, which includes digital images from the collections of over 1,600 museums. Or just load up Instagram, where you will find an up-to-the-minute stream of the most-discussed works in the world of contemporary art. You can buy and sell artworks online through international gallery or auction house websites such as those of Gagosian Gallery, Christie’s or eBay. There are endless ways for people to learn online about art, from scholarly or fact-checked resources, such as museum websites or art magazines, to user-generated content on social media apps like Facebook or, again, Instagram. The art world’s increasingly online nature in the last ten years and this growth of adoption has been spurred by a number of factors, principal among them the mass use of smartphones, the birth and rise of social media, the ubiquity of mobile data plans and Wi-Fi, and the increased wealth of the highest echelons of the world economy who are interested in and buy art.

Yet amidst all of this development of the online art world, some aspects of art are still not online – and may never be. For one, not all art can be bought online. Above a certain price point, galleries often do not feel comfortable offering works online or selling them. Their most coveted works frequently have waiting lists, and they want to thoroughly vet collectors to make sure a piece is going to the right collection (ideally, en route to a museum collection). Secondly, for some great contemporary artworks or iconic historical works there are only low-resolution images available online – not useful for sharing or reposting or studying – due to copyright concerns, and also debatable concerns that if a high-resolution image were to be put online then the artwork itself could be copied. Furthermore, scholars cannot undertake all of their research online. Many art books are not available online due to image copyright restrictions, art book formats (as art books are usually not pocket paperback size) and the fact of the content and age of the books; old art books have not been the top priorities for digital scanning projects such as Google Books.

Most importantly, as purists and conservatives love to note – and have noted since the birth of photography – viewing art digitally does not capture the physical experience of seeing a work in person. Some advances have been made in this area; Google took its Street View technology into museums to try to recreate the experience of being in a museum’s galleries and seeing artworks in their actual contexts – rather than as squares and rectangles floating in space – online. The increased resolution of photography and accessibility of video to anyone with a smartphone has made it possible to create highly ‘zoomable’ artworks that show details even the human eye cannot see, or to provide video experiences of three-dimensional works such as sculptures and installations, limited only by the time someone wants to spend filming all the angles. New augmented reality applications generate images of two-dimensional artworks scaled down or up to any environment you find yourself in, to help you envision what artworks will look like installed in real life. (This is something Artsy introduced recently.) At present, virtual reality companies offer some of the most promising projects aimed at bringing more qualities of art to a digital space.

My view of art online has been strongly shaped and influenced by the fact that I spent over eight years working at Artsy, which is considered one of the art world’s most influential online platforms. Artsy is both a website ( and an iPhone app used for collecting and discovering art. Its stated mission is ‘to expand the art market to support more artists and art in the world’.

In the tradition of other American start-up origin stories like that of Facebook, where an enterprising and ambitious young person identifies a need for something and then builds it in their dorm room, Artsy started out as an idea in the head (and then the dorm room) of Princeton University student Carter Cleveland. Born and raised in New York City, Washington, DC and London, Cleveland was a computer science major with an interest in art history (his father was an art writer and collector). He wanted to put original art on his dorm room walls rather than posters, but he couldn’t find a place on the web to connect with younger artists to buy their work in order to make this a reality. So Cleveland built a website called Exhibytes, which he intended as a social platform for young artists. He thought artists would post their work there and use the site to socially network (as the creative community uses Facebook and Instagram today), and in doing so provide an opportunity for people to buy their work online. It would be a great replacement for the posters he didn’t want to have, and would serve as an accessible way to connect with young artists.

Unfortunately, Exhibytes didn’t work out so well. The upside was that the idea was good enough for Cleveland to win some business grant award money from Princeton to develop it, but the downside was that artists didn’t use the site. Most importantly, it seemed artists didn’t want to socially network there as they had on Facebook and other platforms.

Like any talented entrepreneur, Cleveland digested this feedback and data and then pivoted. He became interested in the idea of developing a ‘genome’ for art, based on what the music website (and now app) Pandora had done with its Music Genome Project. His thinking was that this could be an exciting and new way for people to engage with art: users would be able to, as on Pandora’s site, key in the name of an artist they knew, and get tailored recommendations. In so doing they would expand their knowledge of art, and eventually they might even buy something. It had the potential to be an ideal combination of art education and collecting, and it promised something that really didn’t exist in the art world – because at the time, searching online for art and getting intelligent recommendations was quite difficult. Google searches could only bring up artists’ names and movements; quality control was poor, and the images were not consistently good. Museum websites offered only limited search capabilities. And there was no one-stop shop for the world’s art – art existed all over the internet, on gallery and museum websites, but nothing knitted them together. The existing resources for aggregating art worked only on historical art or contemporary art, never bringing the two into one space.

With this idea of an art genome in mind, Cleveland reached out to Pandora’s CEO, Joe Kennedy (who was also a Princeton computer science alum), and asked if there was any issue with him pursuing it. Kennedy gave his blessing, and so Cleveland began developing what would eventually become Artsy. He tried to buy the domain, but the price was too high, so instead he picked At that time, site names divided by dots were in vogue; more importantly, it gave the site the shortest possible domain name with ‘art’ in it. Artsy, like Pandora, was ‘powered by The Art Genome Project’. I came to Artsy serendipitously at the beginning of 2011, about a year before the site had launched.


For the next four years, as director of The Art Genome Project, I led a team of people to create a genome for art. The project (we internally referred to it as TAGP) was initially inspired by Pandora’s processes. These were to create a list of possible attributes for pieces of music – such as genre, beats per minute, vocals; basically anything you could think of to describe music – and then go through musical pieces and rate them for these attributes, so when you were done you would be creating a list of attributes (‘genes’) for each piece of music that would comprise something like its ‘genome’. There would also be genomes created for each musician, apart from their individual songs, since a general search for The Beatles should bring up the diversity of what they produced, an aggregate sense of who they are; but each song could be given quite different genomes to account for their differences.

Also, for Pandora, genes were importantly not tags, or things that are binary (you are either tagged in a photo or not; you are either tagging a location or not) but could be given strengths from 0 to 100, to rate the strength of the connection of the gene to the musical piece. This range allowed for a significant amount of nuance and a much more detailed way to connect musical pieces with each other, which was important over time, especially as their database got bigger. What this data set worked out to for the user was that through an algorithm on the back end of the site that located similar genomes, they were able to (on the front end) be recommended an artist or a piece of music based on whatever they input into Pandora’s website, and learn about new music even if what they knew themselves was quite limited.

At Artsy, we took an approach similar to Pandora’s. We created a list of all the attributes that you could apply to art (for all art, but in practice it was much more tilted to contemporary art, reflecting the focus of the site); and then we spent a great deal of time applying these terms to the site’s artists and artworks, whose numbers started to rapidly increase as we established relationships with galleries, museums and image rights societies. (Importantly, all of the images we posted to the site were used with permission, not ‘scraped’ from the web.) As with Pandora’s technology, we were able to provide nuance to the genes and not just tag artworks. This allowed for simultaneous similarities to happen – something could have many connections at once. When you searched for an artist, you were given a list of multiple artists – not just one; the same thing happened with artworks, and these lists presented a range of ways in which art and artists might be similar, from formal characteristics to more conceptual qualities.

Because we personally were creating recommendations for potentially millions of people around the world, we consistently acknowledged the subjectivity of the project. We knew that another group working in the same way might have created an altogether different genome; and we knew that the project might have been undertaken in a different way a few years earlier or later. But we had confidence in our art-historical and art world expertise. The overriding concept was that this would be a jumping-off point for learning about art, aimed more at the wider public than at experts (who might have taken issue with our user-friendly presentation of quite sophisticated art-historical connections).

When I started at Artsy, the number of genes was in the twenties and there were people with somewhat limited art historical knowledge running the project. After I started, the numbers grew quickly into hundreds of terms encompassing things like art-historical movements, formal qualities, concepts, places an artist had lived or worked – and what we decided to call ‘contemporary tendencies’, or soft groupings in contemporary art that we were hesitant to define as more formal movements. We also brought more and more people onto the team who had the breadth of knowledge and the skill set to do the hard work of genoming day in and day out, making the endless edits that needed to be made to the genomes as we invented new genes, decided others were not working so well, and critiqued our own genoming through a review process.

At first, I asked Cwilich (to whom I reported) not to tell anyone in the art world I was working at Artsy. Because I was set on an academic career, I was worried about being perceived as a sell-out, or being labelled ‘too commercial’ for an academic role. At the same time, I was enjoying growing our terminology list and creating what would be the first large-scale search recommendation technology for art. I felt it was more exciting than any academic art project I was aware of – and much more collaborative than anything I had experienced in a highly individualistic academic environment.

Before Artsy, the search capabilities for art online were quite limited. On most art websites (primarily those of museums), you could only search for things like artist, medium, technique, date, subject matter and tags; Artsy would change all of this drastically. (At the time of writing, it still remains the only dynamic search system for art in existence: the basic search fields mentioned above are still the only ways to navigate most museum websites with tens of thousands of works.) I sensed that the other people I was able to work with at the company – who had studied art history and worked in galleries, or were from art graduate schools, and who saw Artsy as a rare opportunity to break free of academia and the museum and build something that potentially millions of people would use – were just as excited as I was.

The Art Genome Project is now just one part of the expansive online platform that Artsy (which changed its name from Artsy in 2012) has become; but I, not very objectively, still see it as one of the most fascinating art and technology projects created in the past decade. In addition to my work leading the development of the project, I functioned as an ambassador for Artsy as an educational platform, delivering talks around the country at museum conferences focused on art and technology, as well as arts education conferences. I took part in countless discussions about art online – its past, present and future. After four years, as company projects grew beyond the genome, I was given the opportunity to take on a new role at Artsy, exploring how the company might exist offline. I became curator at large and subsequently head curator, leading and expanding Artsy’s real-life programming. This began with conversations with artists – in galleries and museums – before growing to include art exhibitions and installations around the world.

When I started at Artsy, we were about eleven people, crammed into a few tables in a start-up incubator in the Flatiron district. Between 2011 and 2019, the company grew to employ over 200 people, with headquarters in New York and offices in Los Angeles, London, Berlin and Hong Kong. Artsy eventually partnered with galleries to promote and sell work online, with art fairs to list their galleries and artworks, with museums to promote exhibitions and with auction houses to promote and run auctions. It also has its own highly popular magazine and millions of social media followers. One wonders, over the next decade, whether any other online art venture will match Artsy’s growth, influence and ambition – and how and in what form Artsy will sustain its place in the art world ecosystem.

Discover the book

A Year in the Art World

Matthew Israel