Working for an Algorithm: Power Asymmetries and Agency in Online Work Settings

Abstract

Drawing on interviews with 77 high-performing eBay business sellers in France and Belgium, this article investigates the power asymmetries generated by customers’ evaluations in online work settings. Sellers revealed a high degree of sensitivity to negative reviews, which, while infrequent, triggered feelings of anxiety and vulnerability. Their accounts exposed power asymmetries at two levels: the transactional level between sellers and customers and the governance level between sellers and eBay. Our findings highlight three main mechanisms underlying power asymmetries in this context. First, online customer evaluations have created a new form of employee monitoring in which power is exercised through the construction of visibility gaps between buyers and sellers and through an implicit coalition between buyers and the platform owner, who join together in the evaluation procedures. Second, by mediating and objectifying relations, algorithms reproduce power asymmetries among the different categories of actors, thereby constraining human agency. Third, online customer evaluations prompt sellers to exploit their practical knowledge of the algorithm to increase their agency. Through the lived experience of working for an algorithm, our findings contribute new understandings of power and agency in online work settings.

Revising the ‘science of the organisation’: theorising AI agency and actorhood

“…In practice, AI agents change the nature of organisational design, decision-making, strategy, knowledge production and learning, power and governance. Beyond our observations on how this may challenge several core and relevant theories, this has further theoretical implications for leadership and contingency theories that view organisations (Curchod et al, 2019;Donaldson, 2008;Fiedler, 1964;Morgan & London, 1998) in needing to accommodate AI leaders with different objectives, rules and norms than human leaders. Additionally, conceptualising AI as an independent agent within organisations has implications for governance theories especially agency theory, as traditional governance practices involved with human actors no longer apply to AI agents (Bostrom, 2014;Someh et al, 2016;Van Rijmenam & Schweitzer, 2018).…”

“…What we observe today is AI agents having the capacity to change their behaviour (Curchod et al, 2019), and collaborate, strategize and make decisions independently and autonomously, thereby altering context without being subject to further human action.…”

“…Consider the Hong Kong venture capital firm Deep Knowledge Ventures, who appointed an algorithm to its board of directors to improve decision-making (Hanson, 2017). With AI agents becoming more advanced, this could result in fully automated organisations, where human agents are managed by artificial agents (Curchod et al, 2019;Ma, Yuan, Ghafurian, & Hanrahan, 2018;Rosenblat & Stark, 2016;Van Rijmenam, 2019). Already, we can see the first consequences of what this means within Uber, where AI agents control human drivers (Rosenblat & Stark, 2016).…”

From panopticon to heautopticon: A new form of surveillance introduced by quantified‐self practices

“…Other dimensions could also be tested. For instance, the feelings of reassurance and guilt could be complemented by other emotions, such as frustration and anxiety that both play a role in power asymmetries on online platforms (Curchod, Patriotta, Cohen, & Neysen, ). Since power can shift over time (Jasperson et al, ), a longitudinal study of QS practices could offer a richer description of the empowerment/disempowerment tensions.…”

On the way to Ithaka[1]: Commemorating the 50th Anniversary of the Publication of Karl E. Weick’sThe Social Psychology of Organizing

“…Moreover, how are distributed forms of organizing, as manifested in open-source software development, crowdsourcing, algorithmic-mediated forms of work, spontaneously organized social movements, and agents distributed in space and time, accomplished? (Curchod et al, 2019;Ribes et al, 2013;Puranam et al, 2014).…”