The first step in knowing algorithms is not to regard the “impossibility of seeing inside the black box” as an epistemological limit that interrupts any “futile atempts at knowledge acquisition” (von Hilgers, 2011: 43). As Ashby recognized:“In our daily lives we are confronted at every turn with systems whose internal mechanisms are not fully open to inspection, and which must be treated by the methods appropriate to the Black Box” (1999: 86). Because opacity, secrecy, and invisibility are not epistemic anomalies but a basic condition of human life, the black box is not something to be feared but something that “corresponds to new insights” (von Hilgers, 2011: 32) When confronted with a black box, the appropriate task for the experimenter, as Ashby saw it, is not necessarily to know exactly what is inside the box but to ask, instead, which properties can actually be discovered and which remain undiscoverable (Ashby, 1999). Because what maters is not the thing but what it does, the cybernetic lead does not ask us to reveal the exact content of the box but to experiment with its inputs and outputs.
What can be discovered and described about algorithms difers from context to context with varying degrees of access and availability of information. However, even in the case of seemingly closed and hidden systems such as Facebook or Google, there are plenty of things that can be known. Speculative experimentation and playing around with algorithms to figure out how they work are not just reserved for hackers, gamers, spammers and search engine optimizers (SEO). In the spirit of reverse engineering—“the process of extracting the knowledge or design blueprints from anything man-made” (Eilam 2005: 3)—we might want to approach algorithms from the question of how they work and their general “operational logics” (Wardrip-Fruin, 2009). There are already some very instructive examples of “reverse engineering” algorithms within the domain of journalism and journalism studies (Angwin et al., 2016; Diakopoulos, 2014) and in related academic calls for “algorithm audits” (Hamilton et al, 2014; Sandvig, et al., 2014).
Mythologizing the workings of machines, however, does not help, nor should we think of algorithmic logics as somehow more hidden and black-boxed than the human mind. While we cannot ask the algorithm in the same way we may ask humans about their beliefs and values, we may indeed atempt to find other ways of making it “speak.” Similar to the way ethnographers map people’s values and beliefs, I think of mapping the operational logics of algorithms in terms of technography.24 As Latour puts it, “specific tricks have to be invented to make them [technology] talk, that is, to ofer descriptions of themselves, to produce scripts of what they are making others—humans or non-humans—do” (2005: 79). Technography, as I use the term, is a way of describing and observing the workings of technology in order to examine the interplay between a diverse set of actors (both human and nonhuman). While the ethnographer seeks to understand culture primarily through the meanings atached to the world by people, the technographic inquiry starts by asking what algorithms are suggestive of. Although he does not use the term himself, I think Bernhard Rieder’s (2017) way of scrutinizing diverse and general algorithmic techniques could be thought of as an atempt to describe the “worldviews of algorithms” I have in mind when I use the term “technography.” Rieder ofers a particularly instructive account of how diferent technical logics—in this case, the Bayes classifier—entail certain values and assumptions that inevitably have consequences for the operational logics of specific systems.
60
I F . . . T H E N