 |
The Analytics of Art: Some observations on Art.sy
The Internet has great taste.
iTunes seems to know all about my passion for blues music.
Amazon keeps suggesting books that would help me brush up on my neuroscience.
And, Epicurious even picked up on my preference for spaghetti alla carbonara.
For the past decade or so, the Internet has been making
great recommendations to its consumers. Now with the creation of Art.sy, the
self-proclaimed online, artistic “genome”, the Internet now seems equipped to help
Internet trolls and normal people alike hone their taste in fine art.
 |
| Carter Cleveland, Founder, Art.sy |
While “analytics” may be a buzzword these days, the method
Art.sy uses is nothing special or cutting-edge. Currently, Art.sy employs a
dozen art history majors to organize artworks by particular “genes”. These
genes are categories or labels that denote different kinds of information. Some
genes describe an artwork’s historical context or geographical origin. Others
attempt the feat that countless pretentious artist statements have failed to
do; to distill the greater meaning of a work into a handful of characters, like
“Identity Politics” or “Documentation of Social Life”. Additionally, each
descriptor’s relevancy to a particular artwork is scored on a scale between 1
and 100. Once enough artwork has gone through the initial labeling process, this
information is combined with user-generated preferences. In turn, this data is used
to power algorithms that helps suggest new artworks to Art.sy users.
 |
| An overview of Art.sy's "gene" organization |
Art.sy qualifies as an element of Net Art because it relies
on the Internet to help art lovers make connections with new types of art. The
initiative’s innovation lies in the fact that it is the first platform to
extend this idea of organization and recommendation of fine art – a class of
artistic media that has practically no intrinsic ties to the Internet or
digital technology. In turn, however, this goal has a series of inherent
challenges, the solutions to which will have a large impact on the way art is
understood by Art.sy’s viewers.
First, there is no obvious way to analyze the huge breadth
of art’s visual qualities in the same way algorithms can compare music’s aural
qualities. Unlike translating sound waves into bytes, fine art does not posses many
qualities that are easily translated into digital data. For example, how does
one quantitatively differentiate cubism from impressionism? Or, tell the
difference between oil paints and acrylic paints? Furthermore, the challenge of
analyzing an artwork’s visual qualities is compounded by the fact that Art.sy
uses a two-dimensional image to represent a three-dimensional art work. This
begs the question, how could a computer program quickly distinguish a three-dimensional
ceramic work from a photographic work, if both are represented in the same
two-dimensional digital format?
Second, fine art’s significance relies to a large degree on
its context and meaning. Based my own experience, Art.sy’s current system is quite
consistent in its ability to suggest art works that share thematic
similarities; however, how will this system fare when Art.sy has increased its
database from 2,700 artists to a number around 5,000 – the current scale of the
Google Art Project? Will the paradox of choice ultimately drown any system of
organization?
Finally, can Art.sy really afford to cater to “everyone”?
Many people in the tech world hold the opinion that social media platforms have
an inherently flawed business model. To some, Art.sy seems to be heading down
the same, flawed path. Currently, its business platform is based off taking a
commission for online art sales that are negotiated through its website.
Although more and more art is being bought and sold online, will this
e-commerce traffic be sufficient for Art.sy’s continued growth? If not, Art.sy
may have to revert to its original business model: catering to a private group
of wealthy individuals. Indeed, Art.sy only recently opened up its membership
to the public; previously it was an “invite-only” club.
All of these challenges have an impact on the way users
“discover” new art works through Art.sy. In providing a means to navigate such
a wealth of artistic information, Art.sy has adopted a “browser” oriented
interface. The ease of use and accessibility of recommendations in this browser
encourages “artistic speed dating”, where users briefly flirt with one artwork
before moving on to the next recommendation. Consequently, Art.sy could
potentially create a very reductive art-viewing experience in which users feel
compelled to cursorily scan over a huge breadth of art instead of delving more
deeply into a handful of selected art works. Also, Art.sy’s business model
could have an effect on the actual recommendations it makes to users. For
example, since, Art.sy’s main method of monetization is based of sales commissions,
it has an incentive to direct users to pieces of artwork that are currently
“for sale”. If this were to happen, Art.sy’s recommendations would likely be
skewed towards contemporary artworks. This situation might come about because contemporary
artists will be drawn to use Art.sy as an online marketplace more so than
owners of older, more famous artworks who have pre-established art trading
connections.
Overall, Art.sy is an interesting concept that is worth
watching. At the moment, however, it is still much too early to view it as a
major force of change in the world of art. The real impact of Art.sy hinges on
whether it can break in to popular culture and then dramatically increase its
user base. Alternatively, if Art.sy stays within the niche of the art world
elite, its greater influence will be reduced to an over-glorified mail-order catalog.
Additional Reading:
If you're interested in the 2d v 3d thing, you should check out some of what Google Street View has done with museums and art work.
ReplyDeleteAlso, I suspect that reading up on how Pandora's recommendations work (the Music Genome Project) would be interesting to you -- a lot of it isn't quite as automatic as you make it out to be.
It would be considering to take this discussion into the newfound realm where crowdsourcing and machine learning is put together. As Sam points out, there is actually a great deal of constant human input and curation in the whole process, and for the sake of scale and economics, companies are starting to train whole droves of people in India (to name one of few countries) to specifically 'give recommendations' and then combine and integrate these recommendations into the algorithms.
ReplyDeleteIt would be worth exploring the intersection of 'crowdsourcing and machine learning'.
There is a lot of interesting discussion here, but first I have one technical quibble.
ReplyDeleteI’m not sure I agree with “Unlike translating sound waves into bytes, fine art does not posses many qualities that are easily translated into digital data” A digital image translates vision into bytes in exactly the equivalent way to sound. Just as it doesn’t qualitatively differentiate cubism from impressionism, the sound wave representation doesn’t differentiate soft rock from pop. That is in the form of human-generated annotations. It is possible to create algorithms that correlate digital qualities with these human labels, and it is just as likely to do so in visual as in sound. And with 2 and 3d one might equally ask, how a computer might distinguish a real musical group from a set of tracks done by a synthesizer.
Having said all this, you are raising some fundamental questions about interpretation, context, and meaning. Whether they are simpler in the audio vs. visual space isn’t really the issues.
I’m not sure that the “paradox of choice” applies to computer systems. It means that humans when asked to choose between a set of alternatives get overwhelmed, but in general computer algorithms for this kind of task get better with more data, not worse.
The economic question is a very interesting one where the digital/real distinction becomes key. How much money could you get for an original Picasso, compared to a copy, or even more, a digital scan. Clearly the “real thing” makes a huge difference in art. So that means that no matter how good the representation on the site, people will have additional motivation to get the real thing, and this could be enough to make the business model entirely different from the social media and publishing platforms. Whether it will be sufficient depends on the quantity and quality of art they can recruit.
The art viewing experience is another interesting issue. When I go to a museum, the pace naturally is slow, standing for a while at one artwork before moving on to the next. With a browser, it’s click, click, click. Of course I could sit with one thing on my screen for as long as I want, but the medium prompts quick motion. This is interesting for things like education as well as art.
As to skewing the results for commercial reasons, it seems likely, and isn’t bad as long as it is done a transparent honest way. This depends to some sense on what one sees as the goals of the site. If its purpose is to provide a better commercial marketplace for art, that’s fine. If has a loftier goal of expanding art appreciation in general, then it may be at conflict. What is the avowed purpose?
RE: audio vs. visual digital translations
DeleteMy reasoning behind this point was that there are already computer programs that can accurately determine some organizationally, relevant qualities of audio recordings, such as tempo, Key signatures, modal patterns and musical themes. However, to the extent of my research, I had not found many similar programs that were applicable to the visual art realm.
RE: "paradox of choice"
In my original post, I intended to communicate my belief that the number of different labels that Art.sy has created to organize its database could be very overwhelming to individual users navigating through the website. I meant to associate that phenomenon with the individual's sorting/browsing experience rather than the algorithmic sorting process.
RE: Skewing the results for commercial purposes
Art.sy currently markets itself as a platform to “help you discover the art you’ll love”. However, it originally started as an invite-only online club for the art world habitue. In fact, you still need to "request" an account, instead of just signing up. As a result, Art.sy's avowed purpose is somewhat unclear, and the incentive to promote marketable art is probably responsible for switch from private art dealer's catalog to semi-public online art viewing experience. They are probably still looking for the best way to monetize the service Art.sy is "supposed" to provide.