Let’s talk about agency in mediated creative practices. Scholars have historically been hesitant to ascribe agency to nonhuman actors, a resistance that stems from conceptions of agency as rooted in human subjectivity and intention (Reyman, 2017, p. 116). Yet, rhetorical scholars have already shown interest in the ways that technological structures exert agency on the experience produced during human-computer interaction. For example, scholar Jessica Reyman (2017) redefines the concept of rhetorical agency to account for the ongoing and dynamic collaborations between humans and algorithms within the interactive networks of digital culture. To Reyman, algorithms enact agency when their products influence the outcome of a human-computer interaction, since such collaboration gives algorithms the power to “speak and be heard” (2017, p. 115).
Reyman is particularly interested in algorithmic glitches, when algorithms steer users down the wrong path or produce interactions that are dangerous or harmful. Reyman argues that the very possibility that an algorithm can get it wrong necessarily implies agency. By asking where the responsibility lies for unexpected negative outcomes, Reyman points out the nuanced nature of hybridized ways of making, doing, and knowing.
As an example of such negative outcomes, Reyman directs us to a 2015 project in which Google and Flickr developed an algorithm to recognize objects in users' photos and to automatically tag those photos with concepts like ‘car’ or ‘tree’ or ‘beach’. However, the algorithm “got it wrong” in many ways. Their system began to autotag images of African Americans with “ape” and “gorilla”.
Other examples included images of concentration camps, such as photographs of train tracks leading to Auschwitz or of bars on the gates at Dachau, tagged with “sport” or “jungle gym”” (Reyman, 2017, p. 119).
And this is not the only time that Google’s algorithms have resulted in accusations of racism: in 2016, one user noted the start contrast between Google image search results for the terms ‘three white teenagers’ and ‘three black teenagers’. While the former returned stock images of smiling kids posing with backpacks and basketballs, the latter returned police mugshots.
As Reyman points out, we have to ask, who is at fault for such outcomes? The programmer, the algorithm, the user’s input? Google says its algorithms merely reflect the biases that exist in society and that show up in what and how people search online. Such examples of algorithms gone wrong illustrate the moments where corporations are able to hide behind the supposed neutrality of their algorithms.
Hodgson might suggest that these problems arise “through a distributed set of relations of human and nonhuman capacities of engagement. These capacities include the heuristic impulses and procedural proclivities of people (individual and collective), computationality, and the mechanical—all setting upon one another in the very moments of making” (2019, p. 115). In agreement with Hodgson we might also consider the concept of rhetorical actancy, which posits that “rhetoric is always produced from the dance of various actants engaged in intra-actions within various assemblages. The capacity to persuade, then, and to effect change is a distributed process created in the relationship between multiple and various actants” (Gries, 2012, p. 81).
Since the photo-tagging debacle in 2015, the powers of bots and AI have only become more pronounced, with algorithms producing complete visual and textual works of art based on simple and limited input from human users. As Hodgson points out when he quotes Vittanza, users often approach a machine with an idea, only to find that the machine produces something completely different than was expected (2019, p. 114). The location of agency in questions of AI generated artworks is still up for debate, but the concept of rhetorical actancy suggests to us that these works of art might be considered the work of co-authors: a humans concept formed by algorithmic execution.
