Came across this article in the NYT about "art genome mapping." The idea is to create a massive data base of art through the ages which supposedly can analyze your artistic inclinations and direct you to works of art you may enjoy. Amazon has done this for years with books and Netflix does it with films and Pandora with music. Most internet sites selling things do the same to an extent. For some reason, I feel that the algorithm for predicting taste in visual art may be harder to crack than music and literature. We rarely can say what exactly catches our eye in a painting - the color, the composition, the subject matter, the light, the shadow, the reputation of the artist and sometimes even whether one comes across it in a tasteful museum setting or a roadside flea market. We can to a fair degree of accuracy say for ourselves or about others the *style* of artwork that could have an aesthetic appeal for a particular viewer but not a specific painting. I suppose the Art.sy project will be useful in putting a person in a particular artsy 'box' of sorts but I will wait to see how often its You may also like... recommendation hits the mark with art browsers.
I have written a fair number of times here what I think about formulaic thinking about visual aesthetics. Here is an old post which may be relevant to Art.sy's analytic scope.
Putting viewers into an "artsy box" is exactly right. The enterprise says more about viewers than whatever they may be seeking. It certainly has little or nothing to do with art.
I'm reminded of a story from years ago about a woman from a prominent Southern family, patrons of the arts and pillars of a business empire, who was hosting a party at her home. The house was a study in architectural excellence, including a lighted shelf where little artistic treasures were displayed.
To appreciate the story you need to know about the inertia nutcracker. (I haven't thought about this for years and sure enough they are still available new or used from Amazon and ebay.) It consists of a two-part anvil into which a pecan is secured, and a sliding guided hammer driven by a big rubber band strikes one end, shattering the shell (usually) without breaking the nut. I don't know about other nuts but it works quite well with pecans.
Anyway, somewhere in her trips around the country, indeed around the world, this wonderful lady had come across a piece of one of the anvils. Liking the shape and design she included it her collection of small art pieces. And there it was, along with archaeological relics and some jewelry originals.
One of her guests who recognized what it was asked about it.
"Well I don't know much about it, actually. I saw it and liked it. I don't know what it's supposed to be, but isn't it lovely?"
I love the story and think no less of her as a collector. I recall when the dove soap bar was first introduced someone mentioned that it was really a piece of art even though it was mass produced. I'm sure there are enough people with imagination deficits that art.sy will do well.
(Incidentally, when I saw the .sy domain name I instantly recognized it as "Syria." Most people have no idea, I'm sure. But it makes me curious if any of the charges for that name are going to Bashar Assad and company. The only reason I caught it was that before Ghaddafi was overthrown someone pointed out that "bit.ly" (the third-party site that assigns shorter links to urls too long for Twitter use) was of Libyan origin. Many of us switched from bit.ly to shr.tn when we found out to support Tunisia, the first country in the Arab Spring sequence.)
Posted by: John Ballard | October 15, 2012 at 02:58 PM
John, you are right. Art is indeed in the eye of the beholder which is why I think Art.sy has its work cut out. No matter what the mavens tell us, we will continue to see art in the most unlikely places. When Christo's much publicized installation Gates went up in Central Park, I remember saying that I liked it because it reminded me of the unintended 'art installations' I saw in my childhood like the fluttering colorful banners in Buddhist monasteries and the billowing laundry that the Indian washermen hang on river banks.
Posted by: Ruchira | October 15, 2012 at 09:53 PM
Every thing in the world is art, if you look at it with the right eyes. It's all a matter of perspective.
Art.sy is maybe trying to blend the algorithmic approach to determining and nudging peoples' taste in high-end art materials, similar to what etsy.com has done for the world of crafting, sort of a giant 'better than flea market' online. Ultimately, whether it catches on depends on how well it is marketed to the public, and obviously NYT has its work cut out in trying to popularize art.sy as a site for the discerning and well-heeled.
For the rest of us, there's always etsy.
Posted by: Sujatha | October 16, 2012 at 05:41 AM
And we have the specter of "Rock star Eric Clapton sells an abstract painting by German artist Gerhard Richter for £21.3m – a new record amount for a living artist."
http://www.businessinsider.com/gerhard-richter-work-sells-for-34-million-2012-10#ixzz29T5AyHxA
...as well as the viral popularity of the botched restoration.
https://www.google.com/search?q=botched+art+restoration&hl=en&rlz=1C1SKPL_enUS414US446&prmd=imvns&tbm=isch&tbo=u&source=univ&sa=X&ei=9l99UPbTEYq88ASm8YAw&sqi=2&ved=0CCwQsAQ&biw=1024&bih=649
Posted by: John Ballard | October 16, 2012 at 08:25 AM
The Times article starts off on the wrong foot by assuming that software that classifies art (or anything else) by similarity needs to understand art deeply. However, classifying items of X isn't about understanding X; it's about understanding differences among items of X. And this task can be vastly simpler than understanding X itself. I have little or no idea how cars work, but child can tell a bus from a tractor. I use neural nets etc to classify and count physics events routinely, and quite often the machine is actually going to be better at this than I would be. I'm pretty sure my laptop doesn't know more about the underlying physics than I do. Figuring out that someone's singing a bandish (and not a thumri) in raga malkauns (and not rageshri) is a very different task from being able to produce high quality alap yourself.
Having made this mistake (or so I see it) the reporter, predictably worried that his soul is being stolen by a machine, gibbers about 'reductionism.' But this is confused for a few reasons. First, there's nothing new or wicked about the idea that genres of music or styles of art have distinctive features. Much of what you do in a typical art appreciation course is to acquire those features - baroque music sounds like this, swing like that, prog rock like something else altogether. Styles of music and art have their rules, both explicit and tacit, and every student learns them before he creatively bends or ignores them. It's strange to forget that fact just because you're worried about your status as a sufficiently magical entity.
Also, pattern recognition of this sort in a very real sense is anti-reductive. The "reductive" way of identifying some genre or style is to say "rock music needs to have more than X beats per minute, use at least five of the following instruments, and use/omit various chords." But that's not how recognition (both for people and for machines - BFPAFM) actually works. You indeed start with features like tempo or types of instrument used (if you hear a harpsichord it's probably old music!) or scales. (Pandora probably uses tens of features subclassified to thousands of subtler gradations. People, with better hardware and software, probably use even more.) But you never impose hard requirements on these things - the things you're classifying are fuzzy, multivariate and continuous valued. So you perform some kind of global/gestalt assessment and assign some overall weight. And that assessment involves both design and learning, using very complex ideas connecting different sorts of features. bfpafm there are going to be marginal cases where two or more labels both seem plausible. there are going to be hybrids of classes that ping part of the classification for multiple categories. bfpafm the labeler won't be able to give a crisp, succinct answer as to why any given instance was classified in a particular way - it's done with lots and lots of properties weighed in a complex way, not with a "reductive" simple series of logical decisions.
Finally, I don't know what all art.sy does under the hood, but recommendation systems like Amazon or Pandora or Netflix actually get a fair bit of their classification power not from number-crunching the product but from number-crunching the users. If people who like movies X1 and X2 I've liked also like X3 and hate X4, then you can reassess the odds that I'll like X3 and X4 myself. This approach, far from understanding "reductively" or "non-reductively" doesn't bother with understanding to begin with - it's the equivalent of asking my mom what I'd like for a birthday gift.
Posted by: prasad | October 26, 2012 at 06:01 AM