Travelling Algorithms | Traveling Ontologies

Classification and valuation in today’s society is increasingly done by computer systems and algorithms (Fourcade and Healy 2017). For example, algorithms are used to automatically identify people in surveillance (Neyland 2018), to calculate the risk of disease transmission (Lee 2017), and to assess the risk of recidivism (Kirkpatrick 2016). But algorithms do not create passive depictions of phenomena, they also change how things are classified, valued and handled in practice. For instance how new understandings of the progress of a disease are created when algorithms are used to analyze an infection: The veracity of AIDS patients’ stories can be questioned when their accounts are compared to an algorithmically calculated "normal" disease progression (Lee et al. 2019). Algorithms thus not only depict phenomena in society, but also change how they are understood and handled. Algorithms are performative (Introna 2011).An important aspect of this development is that algorithms are often treated as if they were domain independent—as if they could be translated without friction between different areas of society (Ribes et al. 2019). For example, a US computer system for predictive policing, Predpol, uses an algorithm developed to predict aftershocks to predict future crimes. An algorithm from geology is consequently translated into software that organizes law enforcement (Benbouzid 2019). In sum, algorithms are often translated between different domains and spread different ways of classifying, valuing, and organizing the world.The paper discusses how we can theorize how algorithms travel (cf. Lee and Bj\uf6rklund Larsen 2019). The point of departure is that algorithms fold different things together (Lee et al. 2019): algorithms, ground truth data sets, models, methods, and objects. And that these foldings travel between domains. This approach allows a description of how algorithmic relations reconfigure social and natural phenomena and the social, ethical normative and political consequences of these reconfigurations. Do the algorithms betray and reshape the original ontologies in their new contexts (Law 1997)? Do they then become performative of a particular ontology: That is, as they travel if and how do they reshape the phenomena they are designed to handle (MacKenzie 2003; Introna 2011). The point of departure is to use these two types of algorithms—Agent Based Models (ABM) and Behavioral Algorithms (BA)—as a springboard to theorize the performativity of algorithms in society. The two families of algorithms can be understood as being the inverse of each other: ABM constructs models in a bottom-up fashion, in which the characteristics of particular computational agents are programmed into each algorithm. In contrast, BA are agnostic about the characteristics of an agent, construing action merely as stimuli/response without developing theories about the characteristics of the agent. The two families can consequently be understood to represent two diverging ontological conceptions of complexity: ABM is based on the idea of emergence, and thus a romantic view of complexity, whereas behaviourist algorithms are more aligned with a baroque view of complexity (Kwa 2002)