Distributed seeing: Algorithms and the reconfiguration of the workplace, a case of 'automated' trading

Highlights

• Management literature studied how algorithms automate or augment work, thereby largely neglecting how work reconfigures.

• The paper shows how algorithms transform work practices, and how knowledge and expertise become spatiotemporally distributed.

• Algorithms introduce new ways of seeing into the workplace which need to be aligned.

• Digital transformation is more than the instrumental restructuring of work but reconfigures it

Abstract

Contemporary organizations increasingly rely on digital technologies structuring how work gets done. Algorithms in particular are fundamental for such technologies. Management literature on digital transformation has studied how algorithms either automate or augment work. In doing so, this literature treats algorithms as largely independent from existing work practices. This paper, on the contrary, theorizes and empirically illustrates how algorithms transform the workplace in a spatiotemporal sense by introducing a new epistemic vantage point through which work is understood. We do so by drawing on previous work on reconfiguration and ‘Ways of Seeing’, and through a qualitative case study on sports trading. Our analysis shows that traders and algorithms each perceive and see the market in specific, though incomplete ways. Since this market is partly virtual and constituted via a range of heterogeneous actors, ‘seeing’ the market entails knowing its distributed nature and pulling spatiotemporal distant elements together. Our paper contributes to the literature on the effects of algorithms on work by putting forward the conceptual lens of ‘distributed seeing’. This highlights that digital transformation is more than an instrumental optimization process by automating or augmenting tasks with technology but that it actively reconfigures the work to be done. We show that digital transformation 1) is reciprocal and thus irreversible; 2) patchworked and thus requires mending work; 3) introduces new organizational vulnerabilities.

Introduction

After several meetings with management discussing the digital transformation of the company, I finally enter the trading room for the first time. It's a secluded and well-secured area and someone with a special pass has to let me in. My first conversations with some traders are contradictory to what I heard about the digital transformation so far. In previous meetings the transformation was discussed primarily in terms of automating the work of traders. Several numbers were given, ranging from a more conservative 70% to a more ambitious 95%, as an estimate of the percentage of trading tasks expected to be automated via digital tech. Algorithms, data analytics and AI were introduced to take over trader's work so that bookmaking could happen “more objectively and precisely”. But here, in the middle of the trading floor, the work is not discussed in terms of how objective it is, or even how objective trading can be, but in terms of how trading knowledge, intrinsically, is and remains a matter of human judgement. The head of the trading department keeps referring to trading knowledge as being some sort of an ‘X-factor’: “traders miraculously ‘feel’ what has to be done”. One of the traders I am following today repeats this sentiment: “It's hard to explain. Decisions usually just seem to come automatically. I tend to feel what to do, and it's not so much a matter of calculation but more one of experience… You know, bookmaking is not a science. It is an art”. (Field notes 4 October 2018).

The puzzle in the observation described above is based on a study conducted at SportOdds365,¹ a state-owned gaming company. The apparent discrepancy between what algorithms are supposed to achieve in the view of management and how this is experienced by traders on the ground forms the inspiration for this paper. Management's expectation of the company's digital transformation strategy resonates with how it is discussed in popular debates. Algorithmic technologies are said to bring about leaps of improvement in the efficiency and accuracy of organizational output and production (Brynjolfsson & McAfee, 2014; Mayer-Schönberger & Cukier, 2013). Concurrently, these changes are predicted to lead to the automation of human labour and even the disappearance of jobs that once required specialist expertise (Ford, 2016; Susskind & Susskind, 2015). Although this is primarily a popular media discourse, played out in the press and pushed by technology vendors with an interest in hyping their innovations, it also has strong resonance in worker-oriented literature, for example in research that is inspired by labour process theory (Braverman, 1998; Kellogg, Valentine, & Christin, 2020; Noble, 1999).

In the management literature on algorithms in the workplace we identify two dominant perspectives. First, algorithms are discussed from a perspective of automation. Algorithms are seen as more objective and fairer than human judgement (Mayer-Schönberger & Cukier, 2013), giving rise to the potential to replace human work. Second, from an augmentation perspective, algorithms do not replace the role of humans within work processes, but they are complementary to, though different and distinct from, human capabilities (Schwab, 2018). In both perspectives we trace two assumptions we want to discuss here. First, it is assumed that the work itself and what is required of it is something that can be ‘known’ such that algorithms can then automate or augment specific tasks. Second, implementing algorithms in the workplace is a matter of ‘adding on’ a technology to facilitate specific tasks or make them more efficient. Implementation is seen as an instrumental process with reversible consequences: it does not interfere with the knowledge practices of professionals nor fundamentally transform the nature of work.

These two assumptions do not align with the observations of algorithm scholars who take a more critical, performative point of view. Previous work has shown, for instance, that algorithms are not neutral nor objective, but that existing inequalities and political goals are reinforced in the process of designing and implementing algorithms (e.g. Eubanks, 2018; Gillespie, 2014; Glaser, Pollock, & D'Adderio, 2021; Introna, 2016; Irani, 2015). Studies in this tradition, while exploring a wide range of societal and organizational contexts, converge on the insight that algorithms do interfere with the knowledge practices of professionals, to such an extent that they “have the potential to profoundly transform how work is done” (Orlikowski & Scott, 2016, p. 91). Algorithms, in other words, transform what workers can know and what they can ‘see’ in their work practice. Recent research has shown that algorithms challenge and change existing professional expertise, occupational boundaries and roles, and how coordination and control are organized (Faraj, Pachidi, & Sayegh, 2018; Pachidi, Berends, Faraj, & Huysman, 2021). These changes in the workplace do not occur because algorithms ‘have’ objective power, but because they, as Neyland and Möllers (2017) frame it, produce a particular distribution of knowledge between algorithmic technologies and professionals. In this paper we are concerned with understanding how algorithms transform work practices, with a specific focus on how knowledge and expertise become spatiotemporally distributed among a diverse set of people and technologies.

We do so by drawing on a qualitative study on digital transformation at SportOdds365, a state-owned gaming company operating in sports trading. The digital transformation of trading at SportOdds365 was primarily concerned with automating the work practices of traders via the introduction of an algorithm in their workplace, as the excerpt of our fieldnotes with which we opened this introduction illustrates. We observed how traders worked with this algorithm in their day-to-day practices and interviewed them about their experiences with working with this new technology. For traders at SportsOdds365, the most important change brought about by the use of algorithms was the introduction of the ‘auto-feed’, an automated trading platform that is populated by historical data on sporting events generated by an external vendor. The odds, previously created by traders' own understanding of their evolving market, were now also calculated by an algorithm.

While trading is a difficult field to access for research, it provides a particularly well-suited context to explore the issues we are interested in. This is specifically so because trading floors are historically concerned with the spatiotemporal distribution of work. Knorr-Cetina and Bruegger (Knorr-Cetina & Bruegger, 2002a; Knorr-Cetina & Bruegger, 2002b) have investigated the transformation of global trading terminals, arguing these social systems can be considered as ‘communities of time’ dealing with the spatial and temporal distribution of a market. The transformation of traders' work is most poignantly illustrated by the gradual move from trading pits, with traders co-located in a shared physical space, to a form of trading entirely constituted on computer screens, with traders in different geographical locations watching the same market as it unfolds on an interface. Gariban, Kingma, and Zborowska (2013) have, in a similar vein, looked at how virtual and material dimensions become ‘interwoven’ through such interfaces; while online betting may seem to happen in virtual space and through global networks, it remains firmly grounded in its material organization. Other studies on trading have confirmed these findings and, importantly for our paper, highlighted that there is a perceptual quality to how traders and technologies operate together. Seyfert (2018) has illustrated the affective dimension of working with algorithms, while Zaloom (2006) describes trading floors as ‘sensoryscapes’, attuning us to the idea that even virtual space is experienced in a sensuous and corporeal way. Digital, algorithmically mediated work in trading thus has a specific rhythm that is embodied and perceptual, something which Borch, Hansen, and Lange (2015) identify as a matter of traders calibrating their bodies to markets and algorithms. A prime concern for studies on new ways of working, then, should be studying how digital and embodied or material aspects of practice reconfigure how work gets done (e.g. Barrett, Oborn, Orlikowski, & Yates, 2012; Kingma, 2019).

These prior studies align with our own finding: that the autofeed algorithm specifically affected the spatiotemporal arrangement of the workplace of traders, and that this required new forms of human-machine interactions that are perceptual in nature. While traders traditionally have a specific way of seeing their market (in our case, bookmaking and setting the odds for sporting events), algorithms have introduced new means by which this market can be ‘seen’. Heterogeneous, spatiotemporally distributed actors not only ‘see’ organizational reality in a specific way, but each individual way of seeing only provides a partial vantage point. Our traders, for instance, had an in-depth understanding of the market by their experiences ‘on the ground’ which became interwoven with an algorithmic way of seeing that same market when the data-driven auto-feed was introduced. This phenomenon of algorithms and human workers joining up in mutual efforts to ‘see’ and act in their domain does not align with existing dominant perspectives in the management literature with algorithms leading to automation or augmentation in the workplace. The goal of this paper, then, is to develop an alternative understanding of working with algorithms, which we capture using the term ‘distributed seeing’. We develop this understanding by drawing on the literature on ‘reconfiguration’ (e.g. Barrett et al., 2012; Suchman, 2007) and on insights from what we coin the “Ways of Seeing” literature (Berger, 1972; Bush, 2019; Scott, 1999). Both have examined in detail the interactions between people and technology, with a specific focus on how these relationships shape how we get to know and see our world, as well as the performative implications of changes in these relationships for objects of concern.

A key contribution of the paper is “distributed seeing” as a new conceptual apparatus that permits describing the spatiotemporal-material entanglements within new work practices (Aroles et al., 2018; Aroles, Mitev, & de Vaujany, 2019) that underpin digital transformation efforts involving algorithms. We argue that digital transformation is an ongoing and partially indeterminate process that is characterized by developing a new way of seeing. Because this seeing is distributed across actors and across spatiotemporal dimensions, it involves reciprocity, mending work, and new vulnerabilities. These characteristics highlight that digital transformation, rather than representing an instrumental optimisation process, can result in not only a change in how work is organised and the kinds of work that need to be performed, but also a change in the phenomenon that is ‘brought into view’ through distributed seeing (MacKenzie, Muniesa, & Siu, 2007). Digital transformation and by corollary new ways of working in this sense are not easily contained - their performative effects can spill over from the organizational context to the wider world. As algorithms and other digital technologies become involved in what it means to see and know and compete in the world of work, the very phenomenon of trading, how markets are made (MacKenzie et al., 2007), and how this market is regulated, (Gariban et al., 2013) is transformed in irreversible ways.

It is with the above conceptual context of distributed seeing in mind that we view trading, while unique, as a paradigmatic site to study the coordination of work that is distributed across time and space thereby enacting a specific reconfiguration of work practices (Gherardi, 2012, p. 43). The contributions of this paper are then also of relevance to discussions on digital transformation through algorithms in general. Especially with the Covid-19 pandemic and growing mainstream interest in harnessing large data sets for business intelligence in the background, contemporary workplaces will increasingly have to deal with coordinating work from a wide range of different spatial and temporal sites (cf. Orlikowski & Scott, 2021), which generates new opportunities for organizational ‘vision’ while further complexifying the task of distributed seeing. Understanding how algorithms partake in digital transformations is key to providing more fine-grained analyses of how work is dynamically being reconfigured across time and space.