Viewing Algorithms as Institutions

• In Event: AI, Governance, and Power

Abstract

This paper summarizes the arguments developed in a book titled Algorithmic Institutionalism: the Changing Rules of Social and Political Life, which was recently published by Oxford University Press. It conceives of algorithms as institutions in contemporary societies, focusing on different dimensions of how algorithms structure decision-making and enact power relations. It addresses the need for new analytical lenses to make sense of algorithms’ rising ubiquity in decision-making and to foster democratically infused interactions between humans and algorithms.

The paper starts with an introduction to the underlying concepts of algorithms as institutions. It uses the analytical lens of 4 different approaches to institutionalism (sociological institutionalism, historic institutionalism, rational choice institutionalism and discursive institutionalism) to propose a framework for studying algorithms to comprehend their social implications properly. Such a framework focuses on six dimensions: institution building and design; historical processes; rules and norms; gaming; power relations; and discursive dimensions. These dimensions can, on one hand, drive a comprehensive and nuanced reading of algorithmic systems in contemporary societies. On the other, they also help us reflect on the novelties that algorithmic societies bring to make sense of contemporary institutions. Examining institutions provides insight into the influence of algorithms, while exploring the workings of algorithms illuminates the necessity for innovation in institutional theories. In the paper we develop this framework of Algorithmic Institutionalism, explaining its theoretical roots and illustrating its explanatory capacity through a few examples.

Most of the examples employed are extracted from a specific subfield of algorithmic usage: security. The cases of predictive policing, extensive use of facial recognition technologies, and development of lethal autonomous weapons illustrate a multitude of social dilemmas related to the growing algorithmization of security. Algorithmic technologies employed in security are building a new type of Leviathan in which private organizations perform public functions. Algorithms are becoming institutions that implement security policies, thereby changing routines of the state’s coercive forces. In many situations, algorithms nurture authoritarian, unfair, and exclusionary practices, negatively impacting society. Security assemblages imply new dynamics of action and resistance.

On the grounds of these examples, we argue that Algorithmic Institutionalism seems particularly productive, as it helps us navigate some tempting dichotomies: (1) Animism vs. instrumentalism; (2) Structure vs. Agency; (3) Collective vs. Individual; and (4) Determinism vs. Novelty.

After applying the framework of algorithmic institutionalism, the article argues that algorithmic systems are institutions deeply connected to the growing temptation of epistocratic regimes. Algorithms play an essential role in shaping technical decisions and power relations. People use these systems that influence how they work, communicate, consume, participate in the public sphere, build public goods, appreciate culture, and build social relationships. Likewise, algorithms nurture new forms of injustice, dilemmas, and problems. We argue that democracies’ survival requires algorithms’ democratization as central institutions. This should be thought of as a normative horizon that guides practices and allows continuous criticism of existing institutions. Algorithms must be inserted in political dynamics oriented by the values of participation, equality, pluralism, accountability, public debate, and liberty. Democratization should not be conceptualized as an end state but an ongoing process that revolves around democracy adapting to ever-changing contexts.

Authors

• Ricardo Fabrino Mendonca, Universidade Federal de Minas Gerais

• Fernando Filgueiras, Federal University of Goias and ENAP

• Virgilio Almeida, Federal University of Minas Gerais and University of Harvard