The intersection of art and tech: we chat to Artsy’s founder about democratising design

If you are an art lover, you will love this. You know how Spotify and iTunes can tell the kind of music you might like based on previous purchases and listening habits? The same way Foursquare can tell you what restaurants you might like? Ever wish there was a tool like that for art? Turns out there is and it’s called Artsy.

The company’s mission is to make all the world’s art freely accessible to anyone with an internet connection. We got to chat to the company’s founder Carter Cleveland, who believes that by combining art and science, the company — through its art genome project — can foster a new generation of art lovers.

Talking to Cleveland, it’s obvious that he believes art should not be the privilege of only those that can afford to own it or live near the museums housing great pieces. He reckons Artsy can help democratise art by making it more accessible, giving new people the opportunity to discover great works.

Artsy has 195 000+ registered users, serving over 300+ million artwork impressions to an international audience across 180+ countries. Since its launch, Artsy has introduced significant new features including partnerships with leading art fairs, ecommerce payment processing, and the ability for enthusiasts and experts alike to contribute insights and anecdotes, through posts, about the 73 000+ artworks on the site from 1100+ galleries and 170+ institutions.

Memeburn: How did Artsy get started?

Carter Cleveland: It really goes back to my childhood and my parents — my dad’s an art writer, so at an early age he was taking me to galleries and museums and talking to me about the art. I really grew up around it. I’ve always had a deep passion for art.

My mom’s side of the family is all physicists, so what I really excelled at in school was math, physics and ultimately in college I ended up majoring in computer science and engineering. I became really interested in how websites like Pandora and Netflix were applying these artificial intelligence algorithms to music and film to help people learn about and discover other kinds of music and film that they didn’t already know about.

I love that idea of combining algorithms and these forms of self-expression to get out to a larger audience. I was searching around online – searching for art, my main passion – when it occurred to me that there is no website with all the world’s art on it. It just seemed like this giant gaping hole in the internet. Clearly, there should be a website with all the world’s art on it, and of course, such a website (just like Netflix and Pandora) should have a component to it that was about recommending new artworks to people they might not otherwise have known about and helping them learn about and understand those artworks and their historical context.

I always assumed there would be a buying and selling component to it, and that’s how we would make all this money, but that part was a lot less well thought through. In the beginning, it just seemed like a really exciting, fun opportunity to create something that had to exist. I started working on it in my senior year of college, and that was five years ago now. It took a lot longer than I expected.

MB: What is the Art Genome Project?

CC:The Art Genome Project is actually just a team within Artsy. We talk about it because it’s very much its own thing. It is a layer of metadata that maps all the connections between artists and artworks. The way it does this is by categorizing artworks and artists by different criteria. We call these genes, but a gene can be an art historical movement like pop art or abstract expressionism, it can be a stylistic quality like hard edge or chiaroscuro, or more conceptual (like dialogue of the past or stolen moments).

There are over a thousand of them, and every artwork and artist is scored along all these different genes from zero to a hundred. A work might have a hundred pop art, and a thirty in contemporary conceptualism and an eighty in hard edge and a thirty in bright colours. The average artwork has between twenty and thirty genes and these genes are added to artworks and artists by an internal team of art historians who study the artwork and the artist and add all this information to it.

That’s how we figure out what is similar to what, in the same way that we calculate genetic similarity in DNA by doing similarity computations. Which is a great example of an algorithm I was taught about in college — how do you compute similarity between two pieces of genetic code? It’s a very similar idea to computing similarity between two artworks and two artists.

MB: That makes art quite scientific.

CC: Yeah — our core value is that art and science are these two things that typically are like oil and water. You don’t mix them together. But that’s the whole point of Artsy in its most abstract level: bringing together art and science.

MB: What challenges did you face with Artsy? There must be some technical difficulties?

CC: I prefer the word ‘fun’ instead of technical difficulties. Nerds just love this kind of stuff. There are several challenges involved. On one hand, you need to have a really good system for entering this data in an efficient way — that’s more of a user experience and a work flow challenge in order to make the jobs of our Genome team really efficient.

We’re constantly getting better at that. For instance, if you start off by genoming the artist, then you copy the artist’s genome to all the artworks, then for each individual artwork you might look if there are different series. So you might genome one work from each series and then copy those genomes over to the other works in the series and then you make individual adjustments.

Probably the hardest challenge is calculating similarity between artworks at scale. If you give me just two artworks it’s really easy for me to tell you the similarities between them. The simplest way to calculate that is called the Euclidian distance metric, or computing the dot product. Those are just fancy ways of saying ‘what’s the distance between two points in space.

If someone’s on Third Avenue and Fifth Street, and someone’s on Tenth Avenue and Sixth Street, the way you calculate the difference between them (the way you learn at school is Pythagorean) is you take the width and height of that triangle, square them, add them together, square root that. We’re doing it in almost a thousand dimensions. Every time we calculate the similarity we’re calculating those distances and squaring them and adding them up in over a thousand different dimensions.

That’s actually not even the hard part. The hard part is doing that at scale, because if I have two artworks, then how many similarities do I have to calculate? Just one. But if I have ten artworks, I now have to calculate the similarities between all ten. That’s going to be closer to a hundred different similarities I have to calculate. That goes up as a square of the artworks in my database.

So once you start getting thousands, and tens of thousands, and hundreds of thousands of artworks, this starts to be millions of computations, and every single time I add a new artwork, I now have to recalculate similarity between that artwork and every other artwork in our database. You as a user, you enter Artsy and we want to know what to recommend to you, so now we have to find out what’s similar to you.

So that is a completely unscalable problem. That’s where really clever algorithms come in called heuristics. They don’t give you the exact answer that is 100% correct 100% of the time, but there are really clever heuristics that if you are willing to accept a small amount of uncertainty, let’s say 95% of the time, turns out you can get massive increases in speed and scalability. The art of choosing these algorithms is figuring out the correct heuristic.

What is actually pretty cool is that the research that our algorithms are based on here is pretty cutting edge.  The beauty of scaling systems and coming up with these heuristics is that there really is an art as well as a science to finding just the right balance.

Those are the main technical challenges. We’re also working on algorithms that help us automate the genoming process. So computer vision algorithms. It’s very hard to look at a work and know that it is a pop art work because there are such a diverse array of pop art works, but you can look at a work and say ‘how geometric is it?’. You can eye-ball it. Is this hard edge work versus soft edge, is it blurry, is it bright colours? So there are a lot of visual characteristics that we’re looking to automate with computer vision algorithms. That’s another very challenging area.

MB: How do you decide what artworks to recommend to users?

CC: As you’re using Artsy, you’re actually defining your own genome. The great thing about the genome is that the recommendation system doesn’t pigeonhole you. The more traditional recommendation systems you see on the internet are called collaborative filtering, and the premise behind that is that birds of a feather fly together. So the idea is that the people who like the same kind of things you like, you will like the things that they like that you haven’t yet discovered. The problem then is you can become very pigeonholed, and it can be hard to break out of that, and you all end up converging towards the same items in your group.

What’s cool about a taxonomical approach (which is what Artsy and Pandora do) is you might spend a whole day checking out Kandinsky and the certain period and genre of artists, and it knows great, Michelle is clearly very interested in abstract expressionism, she’s influenced by music, and you’re building on your genome there. Another day you might come back and be obsessed with black and white landscape photography and that’s going to count as well.

Some people who have had no exposure to art might not have any particular strong passion. Some people may have a few favourite artists or one genre that they really love. Some people who have spent a long time in the art world may love a ton of different genres. So the genome can very accurately capture that simply based on your interactions in the site – what you’re clicking on, what you’re searching for, what you’re following, what you’re sharing with your friends. That all helps tell us what you like.

That’s the basis for recommendations – it’s just purely to do with similarity. It has nothing to do with other users on the site are doing, although in the future I’m sure we’ll look into that data.

MB: Was the social element of sharing art key in starting Artsy?

CC: It’s funny. In the early days I was actually very interested in social – this was when Facebook was first opening up – and a lot of my original idea behind Artsy was all about social. It’s interesting because when we had to start really building it, we just really realized we needed to focus on just getting the basics right. It’s hard enough just to get the imagery. We have to clear all the rights for every work you see on Artsy, so a huge part of our business is simply getting the copyright to display all the art you see. That’s why you can’t find it anywhere else.

What’s interesting is that now that we’re starting to get the fundamental building blocks in place, now I’d say that a lot of what we’re excited about and what we’re going to be building today is more and more social elements. It’s hard to know exactly when these things will happen, but without a doubt in my mind I know the ability that pretty soon on Artsy you will be able to have friends, follow friends, see what they’re sharing.

If you’re a collector, you might want to keep your activity private, but I think the majority of users on Artsy are not actually collectors — the majority of collecting is done by a small subset of our users. The vast majority just love looking at the art and learning about the art. I think for those users, I’m really excited about the opportunities to create a really vibrant social community.

MB: Would you say you’re bringing art to the masses?

CC: Absolutely. That is the whole point. The buying is important because that’s how we become a successful business — that’s how we make money for our gallery partners and our artists – so it’s extremely important that we are very successful as a business. But the long-term goal is that we want everyone to be passionate about and learn about art.

Of course, the beauty of that means that there will be more people who buy art, because every single person who buys art today, it all started at some point with them seeing an image, and then learning more about that art. No one just hands over their credit card and spends a lot on art. It all starts with education.

I think there’s this beautiful harmony between the two; the more people are educated about art, the more people will buy art. The more people who buy art because of Artsy, the more we can make all the world’s art free to anyone with an internet connection. Even if they’ll never be able to afford it, it’s okay.

MB: How do you sell an idea about marrying art and technology to investors?

CC: We have a great list of investors, but I was probably rejected literally over a hundred times. I had at least a hundred meetings with investors (an hour at least per meeting if not many hours) and was told no. That’s fine. The beauty of investment is that you only need one yes.

You have to find investors that share your vision for the company, and our vision is very specific. It’s a very long-term focus. Yes, if we’re successful, it will be extremely lucrative. It’s all about finding those investors who are passionate, on an emotional level (not just a business level) about changing the world, making art more accessible, increasing the size of the art market.

They also have to be willing to hand over their cash and be comfortable never seeing any return on it at all, or, if they do, being comfortable waiting ten years for that to happen. We’ve been very fortuitous, after many attempts, at meeting those kind of investors.

MB: What’s next for Artsy?

CC: This is just the very beginning. At our very core, we are an innovation company, which means we’re driven by not just world-class people on the team, but by technology, by design, by product – we’re fundamentally about innovation based on top of the value of art and science. Today, that means making all of the world’s art accessible to anyone with an internet connection. Who knows what that might mean in five or ten years? It’s a very timeless value. We’ll always be evolving, always changing and innovating.

Dialing it back to the short-term, I think we’re very excited about opportunities in mobile, in tablets. Today we’re doing a good job of serving a lot of the existing market, and we hope to be doing a much better job in the next nine months. But I think over the next year, it’s really about that next wave of people who (on the business side) can afford to buy art but are currently sitting on the sidelines because the market can be so intimidating, so opaque. I think as we feel better about the value we’re driving for the existing market, we’re going to start thinking more about how we can expand that market.

For every household that currently buys art in the US, there are 37 other households with the same average income. So if you can even get one of those households buying at the same rate, you’ve doubled the market. That’s where we think a huge opportunity is. That goes back to your investor question – that’s why we have people giving us money. They believe we can expand the market.