The Black Box of the Present: Time in the Age of Algorithms

as i click on the link to the news site, a cascading series of actions unfurls behind the static pixels of my screen, obscured from human view not just by design but also by the boundaries of my temporal perception. In a matter of milliseconds, my request for the page has reached one of the company's servers, triggering a flurry of activity as the site identifies me through cookies, IP address, and other means. The server contacts other machines in the company network to pull together not just "the news" but my news, curated in various ways to capture and maintain my attention (e.g., through recommendations for what to read next) and establish a familiar context by referencing previous interactions with the site (e.g., items saved in my "history"). But before the page loads, my query also triggers a number of other requests to companies, including Google, Facebook, and Twitter. Advertising algorithms engage in an automated auction to decide which ads will be shown to me on the news site, taking into account prevailing bids as well as my own history, social media presence, and membership or nonmembership in various predetermined target populations.

Over the space of perhaps a half-second, tens if not hundreds of machines have participated in a complex set of automated negotiations and interactions to shape, format, and present to me an experience of the "current news." The experience of clicking on a link might seem instantaneous, but the gap between request and answer is vast by computational standards, and filled with activities in which I am not the observer but the observed.

Eliminating latency has long been a fetish of technologists, and the history of the modern Internet can be measured by the diminishing gap between query and response. A rule of thumb in interface design states that "to create the illusion of direct manipulation, a user interface must be faster than 0.1 second," which roughly correlates with the speed of our own nervous system's sensory responses (Nielsen 2009). Designing systems that react so quickly leads to a sense not only of immediacy, but also of control. It creates the illusion that the Internet is an extension of ourselves: a structure integrated into the central nervous system. But this immediacy obscures all the work we are not in control of, the framing and targeting that increasingly tailors the foreground and the horizons of our unique, individual informational worlds. What seems immeasurably fast or instantaneous to us is quite slow for the machines, giving them plenty of time to do the real work.

From the static-laden gurgles of early dial-up modems to Google's decision to include a measurement of response time in all its search results, we have been constructing a computational model of contemporaneity for decades. Today the logic of immediacy, of the update feed, of the news cycle sublimated into an endless live stream, has transformed our understanding of the present tense in ways that reach beyond user interfaces and server architecture. Manuel Castells has called this "timeless time," both ephemeral and eternal (Castells 2000, 494).

When we speak of the contemporary in the twenty-first century, we are almost always talking about something that requires computation to access. After all, there are billions of dollars changing hands over the question of who gets to construct the present for you. When you access a website, perhaps to find out what is happening in the world "right now," hundreds of servers are involved in the micro-auctions that determine which advertisements will appear on the page, and maybe in organizing its content according to models that predict your interest in different topics. Services like Google Now seek to understand and anticipate our desires and interests—ideally, to predict the entire topography of the contemporary for each of us.

The stakes of the algorithm are fundamentally about the social production of time and a fundamental shift in our experience of the present. I will pursue this in three stages: framing the algorithm as culture machine, defining the...