Duncan Alexander, "Brion Gysin Internet Circulation Project", 2010, animated GIFs on HTML page
A few weeks ago, an AFC reader commented on the potential of Art.sy, the still-theoretical “Pandora for art”. They were, it seems, underwhelmed:
Anyone who’s seriously suggesting that software algorithms can replace humans in their interactions with art has overstayed their 15 minutes of fame and should be treated as such. Enough said.
AFC staff were quick to point out that Art.sy's probably not going to put anybody out of work, but then I got to thinking: why not? Art.sy's suggestion engine model is easy enough to swallow, but Google's new Search by Image
feature imagines something more revolutionary: computer-assisted curation.
Now, we're not there yet. The images accompanying this post are the ten images Google Image Search found to be “most relevant”
[for me] to Duncan Alexander's Brion Gysin Internet Circulation Project, and it's apparent that visual similarities are still far more heavily weighted than semantic information. The results are even delivered under the heading of [merely] “Visually similar images”, even though visual similarity is only a small part of the algorithm: search with a black-and-white line drawing of a goat
, for instance, and you'll get lots of black-and-white line drawings, but only of domesticated animals.
So what, exactly, would be necessary to turn this indexing and organizing into curation? Well, what exactly is curation? If we strip out the myriad social and administrative tasks of the real-life curator – the connections, the negotiations, the shipping and hanging and lighting and writing – we can arrive at a pretty simple job description: good curation is the discovery and display of unexpected or heretofore unknown patterns and flows in visual culture. So why can't a computer do that?
The principal requirement of curation is a knowledge of human associations, both visual and cultural, and it seems inevitable that computers will soon understand those associations as well as we do. Each time we search, create, or organize online, we go some way towards ensuring this: the algorithm is always listening. Through our actions and choices, we're continually adding to the body of information available concerning the patterns of human visual culture. Today, simply including an image on AFC creates a network of connections and datapoints recording the possible presence of subjects known to be common to the site: contemporary art, for instance, or GIFs, or even “hipster pussy” (which, incidentally, is a significant traffic driver). Once formed, those connections can be systematically tested for their strength by returning these images as search results, and tracking the responses of users; soon, information emerges about which images humans judge to make sense together. As our actions produce the raw material of machine curation, our presence online funds its production. Even as we produce data, we financially reward the class of information-sellers and ad-placers who have no more pressing or profitable aim than the successful modeling of human culture. There are simply too many interested players: human domination over visual and cultural pattern recognition has an expiration date, and this very post is drawing it nearer.
So what would computer curation look like? In the short term, probably very boring. After all, search algorithms start from calculations of a mean human, and the mean human wouldn't make a very interesting curator. In the long term, though, that problem will be overcome: differentiation is a known goal that Google, among others, is working hard at, because well-targeted ads can pull in much greater profits than mass broadcasts. The same algorithm that ensures my searches give me results tailored to a 20-something male in Brooklyn can, by definition, approximate the choices that would be made by that 20-something male in Brooklyn: the very images accompanying this essay, though taken without omission or alteration from Search by Image's results, reflect Google's (probably very thorough) understanding of who I am, what I do, and what images I might have chosen without the algorithm's help; when I search in a browser without my identifying cookies – the bits of code that identify me as a particular Google user, with known interests and habits – I find a different set of images, and a distinct curator at work. Increasingly, algorithms can be prodded to produce convincingly eclectic “personalities”, whose viewpoints and decisions would be interesting to a public.
That said, perhaps the construction of a convincingly unique curatorial voice is counterproductive. There's a certain tyranny to the curator's role, a certain privileging of one individual's taste, that runs counter to the potential of the medium. Curation's value as an expressive medium, the part of its potential that cannot be replaced by, say, art historical writing, is its Socratic nature – the manner in which an answer is prompted but not completely given. Juxtaposition can imply but not state, so in every exhibition the viewer's participation is in some way essential to completing meaning. There is consciously participative curation – I'm thinking here of Mark Wallinger's “The Russian Linesman”
, which toured Britain in 2009 and was set up more as a game of connections than an authoritative statement – but this is equally true of ostensibly objective historical projects. There's always some work left to the viewer.
This sense remains present when the curator – the plotter of points to be connected by the viewer – is an algorithm. The Google-curated images throughout this essay have this power: the lampooned politician’s weary steed on a cover of Puck echoes in a Dutch drawing of a man riding an elephant, while the latter’s digital watermark of “authentichistory.com” snaps us back to a recognition of our digital gaze as quickly as the flashing rainbow in Duncan Alexander's Brion Gysin piece. All of this is unstated, but available to the viewer; the connections we draw as spectators complete an understanding curation alone can only approach.
The scale and effect of that completion, though, depends on how well we understand the curator: we're more likely to “get” the curation of an organizer with a broadly similar worldview to our own, as we attempt to repeat the logic that generated this series of images. Here, the algorithm lies entirely opposite to the human curator: while a human can always prompt empathy, the complexity of Google's algorithms approaches incomprehensibility. While we can grasp the essential ingredients – keywords, image size, color sets, shapes and color distribution, network proximity – it's clear that the actual reasoning is beyond our grasp. I can determine a similarity in color between two images, or a similarity in content between two more, but I cannot possibly weigh the two relations in the proper proportion. Further, the data these algorithms rely on is generally unavailable to the public, since it constitutes Google's main product.
In this situation, we are faced with the challenge of Surrealist curating: there is a set of rules governing what is collected and displayed, but it lies beyond our understanding, like the laws in a dream-world; it is unknown and unknowable to our daily, waking logic. This is crucial: where a human curator's expressive attempts can be got or missed, i.e. can be understood with various levels of success, we can admit from the outset that all reactions to a machine curator are equal. Further, these reactions are in no way discouraged; we can continue to form our own connections in a machine-generated exhibition just as the religious can establish and enforce worldly laws: the knowledge of an unknowable governing force needs not diminish what it governs.
Indeed, algorithmic curation may well be the most democratic method of curating possible. Internet culture, as cemented in communities like tumblr, invites anyone to curate, but it does nothing to address the top-down nature of curating itself: the best tumblrs, like the best art curators, are known for a taste and judgment which is in some way distinctively superhuman, and itself a subject of spectatorship for the less fortunate (or less dedicated) masses. Democracy is perhaps not always a virtue, but so long as we live within it it must be confronted: is it what we want in our image juxtapositions? Is the viewer-produced meaning in anonymously-curated images less convincing, less moving, or less real than the singular, curator-produced meaning of human-generated exhibitions? Can we yet believe in the meaning we ourselves produce? Can we, should we, must we change that? The technology will be there; the real issue is whether we can accept the radical shift it makes possible.