Abstract
The article explores the normative concerns raised for gig workers by algorithmic management (AM), by embracing an ethnographically sensitive approach to philosophical inquiry. Inspired by Michel de Certeau’s concept of ‘tactics’, the article suggests interpreting workers’ attempts to ‘trick the algorithm’ and escape some of AM’s constraints as ways to reclaim agency, in the absence of suitable organizational conditions for its affirmative exercise. The kind of agency specifically deployed by workers in cooperative settings is referred to as ‘contributive agency’, broadly defined as workers’ control over their contribution in multiple dimensions – epistemic, relational, participatory and protective. Contributive agentic capacities are not mere properties of agents, but organizationally mediated capacities that can be more or less enabled or constrained depending on the contributive context. It is argued that below a certain threshold, AM’s agency-constraining features are objectionable and desirable agency-enabling organizational conditions are identified in the four dimensions.
Keywords
This article explores the normative concerns raised for workers by an increasingly widespread form of organizational practice: algorithmic management (AM). AM is a newly introduced umbrella term (Lee et al. 2015) that describes a series of ‘technological tools and techniques’ aimed at remotely managing disaggregated workers through algorithms, which have the responsibility to make and execute decisions in lieu of human managers (Mateescu and Nguyen 2019, 3). Similar concepts include ‘algocratic modes of organization’ (Aneesh 2009), ‘Digitalized Management Methods (DMMs)’ (Moore and Joyce 2020), and ‘People Analytics (PA)’ (Gal, Jensen, and Stein 2020). AM has several applications. Examples include ‘automated gatekeeping’ practices filtering out candidates to hire; predictive analytics to match jobseekers with companies (e.g. LinkedIn); automated evaluative tools to assess workers’ performance through rating systems. This form of management (semi)automates and de-personalizes the work relation while introducing new forms of control. If the link between technology and managerial control is not new, AI technologies have allowed unprecedented possibilities in this regard. Consider the typical characteristics of AM: ‘data collection and surveillance’ through technology; ‘real-time responsiveness to data’; ‘automated or semi-automated decision-making’; rating systems replacing performance evaluations; ‘nudges’ and penalties to influence workers’ behaviour (Mateescu and Nguyen 2019, 3). Although AM is not exclusive to gig labour, this article specifically focuses on its applications within the gig economy, with particular – although not exclusive – attention paid to the ride-hailing sector. The gig economy refers to a system in which platforms match the supply and demand of jobs (or gigs), implying that workers are not hired and, in most cases, not even recognized as offering work services.
So far, AM has been mostly addressed by disciplines such as sociology, law, and economics. While providing valuable empirical insights, the distinctive normative concerns raised for workers by AM are yet to be fully explored. Recent contributions in philosophy include critical views claiming that ‘algorithmic management limits people’s ability to cultivate their virtue and flourish’ (Gal, Jensen, and Stein 2020, 1) from a virtue ethics perspective, and views on workers’ dignity from a capability approach (Lamers et al. 2022). In this article I focus on workers’ agency and its organizational conditions, at the intersection between the philosophy of work, social philosophy, and AI ethics. More specifically, I explore the idea of an implicit normativity shown by gig workers engaging in practices aimed at ‘tricking the algorithm’ and escaping surveillance. Examples include workers simultaneously logging off the app to cause an increase in fares, making orders via the platform to spread the word about protests, or mocking the app’s GPS location system. I consider these as organizationally creative behaviours with potential normative interest and take them into account in philosophical inquiry. To explore these cases, I rely on media reports and literature in the social sciences (e.g., Ferrari and Graham 2021). Inspired by Michel de Certeau’s (1984) work, I suggest interpreting them as ‘tactics’ reclaiming agency at work: the idea is that, in the organizational shortage or absence of suitable conditions for the affirmative exercise of agency, tactics allow for an ‘oblique’ exercise of agency. I refer to the kind of agency specifically deployed by actors in cooperative settings as ‘contributive agency’, broadly meaning workers’ control over their contribution.
In what follows, I focus specifically on three examples: algorithmic counter-tactics aimed at ‘tricking the algorithm’; alternative bargaining tactics; and ‘collective guesswork’ practices. Before diving into these examples, I outline three broad normative tensions of AM to set up the discussion and highlight some of its stakes (section 2), to then articulate the concept of human agency broadly understood, and contributive agency more specifically (section 3). Section 4 reports examples of workers’ tactics and contestatory practices, to then expand on the contested aspects of AM (section 5). Section 6 focuses on the relationship between agency and tactics in more detail. Then I suggest thinking of contributive agency as involving four main dimensions: epistemic, relational, participatory and protective (section 7). Finally, I suggest some normative criteria in the form of organizational conditions enabling contributive agency (section 8). Section 9 addresses some potential objections. Before diving into the discussion, section 1 clarifies the method and aims of this article.
1. Method and aims
By considering the ‘situated experience of ordinary moral agents’ (Herzog and Zacka 2019, 763) as a source of normative exploration, the article embraces an ‘ethnographic sensibility’ (ibid.) approach to philosophical inquiry. 1 That is, it explores possible normative claims emerging from the practices of situated agents, based on the intuition that experiential sources can be ‘generative of theoretical insights’ (Bell and Zacka 2020, 2) and provide a situated entry point for ‘reflective equilibrium’ (Rawls 1971; Herzog and Zacka 2019). Thus, the article consists of both a ‘descriptive’ and an ‘interpretative’ endeavour (Herzog and Zacka 2019, 764), addressing questions such as: what are some of the agents’ responses to AM, and which normative demands do they seem to implicitly express?
To be sure, this approach does not imply undue generalizations or a mere derivation of norms from pre-emptively selected facts. This would be methodologically problematic, as the examples considered may not be representative of the experience of all actors involved in AM. Surveys on gig workers’ job satisfaction across the world are hardly conclusive, the results being far from showing a black and white picture. 2 Yet, the suggested interpretation is in line with findings about the aspects of AM that gig workers complain about the most (Möhlmann and Zalmanson 2017).
Not only do we have reason to distinguish between ‘good gig’ and ‘bad gig’ (Wood et al. 2019), but the gig work population is demographically and motivationally heterogeneous. What is a side hustle for some is the main source of income for others: Some drivers sign up because they need extra cash on the side; others do it as their full-time job. Many resort to it as a stopgap solution when businesses fail or unemployment strikes; others take up ridehail work for the fun of it. Some are trying it out to pad their savings; others have little choice, putting in fourteen-hour days just to feed their families. Some tell me that they do it simply to get out of the house and experience a sense of human connection; others are desperate to find a way out of Uber. (Rosenblat 2018, 7–8)
Thus, this article does not ground general normative claims on workers’ tactics as such; the tactics are taken as a source of normative reflection whose coherence with our moral intuitions remains to be investigated, like in the method of the reflective equilibrium (see section 9). Normative conclusions require independent normative grounding – provided in this context by the concept of contributive agency. As Herzog and Zacka (2019, 768) put it, ‘an ethnographic sensibility is in no way a substitute for independent moral reflection. Rather, it provides us with the material on the basis of which we can engage in such reflection’. The reference to workers’ tactics is thus essentially exploratory and interpretative; it aims at enriching our normative reasoning about AM. If the article does not merely infer general norms from particular facts, through the reflective equilibrium, however, it makes claims of more general scope by suggesting some normative criteria for practical orientation.
While some theorists have found inspiration for their normative insights in the demands of existing social movements (e.g. Young 1990), the cases discussed in this article are relatively un-coordinated, impromptu ‘hacks’ against managerial-algorithmic control, lacking the more coherent outlook of organized collective action. Yet, normative demands do not always reach the level of articulation and expression of fully-fledged social movements: there is a lot between the extremes of open collective resistance and strict managerial control. Overlooking them might make us miss out on a more granular understanding of the current forms that frictions between managerial control and labour are taking, where opportunities for unionization are limited. Seeing workers’ everyday practices as ‘mere peripheral phenomena’ may amount to a narrow, ‘one-dimensional’ understanding of politics (Lüdtke 1986, 83). As historian Alf Lüdtke observed in his analysis of daily factory life in the early 20th century, the fact that not all workers would engage in ‘spectacular collective action’ did not make them ‘inactive or apathetic’; they engaged instead in more subtle and everyday forms of resistance, by reappropriating ‘bits of the time formally designed as working time’ (e.g. horseplay, cleaning the machines while still running, napping, daydreaming, coffee breaks and even ‘literal disappearance’, ibid., 79–80). Lüdtke referred to these examples as ‘Eigensinn’ (‘self-will’, ibid., 84): Eigensinn happened in isolated, though in constantly repeated, reappropriated moments, in which the workers created, articulated, expressed and fulfilled needs. In the factory, these needs were blocked or suppressed by the disciplining regularities of work orders, factory regulations and organization, by the constant danger of being under the surveillance of supervisors, masters, and foremen (…). (Lüdtke 1986, 81)
2. ‘Be your own boss’: normative tensions of AM
Three main normative tensions can be identified in AM within the gig economy. They are somehow captured by the slogan used by Uber to attract drivers, ‘Be Your Own Boss’: 3 (i) workers ‘without’ workers; (ii) hyper-management ‘without’ managers; and (iii) autonomy ‘without’ autonomy.
(i) Workers ‘without’ workers. Most gig labour companies describe drivers and riders as ‘partners’ or ‘independent contractors’. 4 Yet, the advantages that partners would normally have are not actually experienced by gig workers. For instance, while they can freely log in and out of the platform, once they have logged in, their activities – for example, working hours and driving habits – on the app are ‘heavily monitored’ (Mateescu and Nguyen 2019, 5) and used to nudge them into working more. Individuals providing services for these companies do not sign a regular contract, they ‘subscribe’ to a platform. Although legislation on this matter is rapidly evolving (with some countries having recognized contractors as ‘employees’), by denying them the status of workers, companies can avoid labour regulations, such as minimum wage laws and other labour rights – for example, healthcare provisions (Rosenblat 2018). In this way, platforms can rely on work without treating it as work, thereby getting its benefits without dealing with its costs.
(ii) Hyper-management ‘without’ managers. With (semi)automated management and a large part of decision-making being delegated to algorithms, forms of control that were previously unthinkable have become possible. For instance, as a result of drivers’ behaviours being constantly tracked, sanctions often follow behaviours that the companies want to discourage (e.g. declining a few orders in a row will result in the algorithm offering drivers fewer rides, or banning them from the platform altogether). Unsurprisingly, the algorithm’s functioning is often perceived as mysterious. These ‘hyper-managerial’ forms of control go with the total absence of interaction between human managers and workers. Questions about the app’s functioning along with problems drivers routinely face can only be addressed through ‘online support’. Hence, AM in the gig economy seems to imply especially pervasive forms of control despite no human managers being directly involved in the process.
(iii) Autonomy ‘without’ autonomy. The prospect of being one’s ‘own boss’ promises a great deal of autonomy to workers. To be sure, most of them tend to express satisfaction with the flexible schedule and discretion in task performance (Möhlmann and Zalmanson 2017; Rosenblat 2018; Wood et al. 2019). Yet, temporal autonomy should be put into context. If algorithmically managed gig labour can in principle allow more flexibility and autonomy in workers’ time management, at the same time, low pay and nudging practices pressure workers into working more, while lacking room for voice and claim-making (Möhlmann and Zalmanson 2017; Wood et al. 2019). In the original sense of autonomy – as self-rule or giving norms to oneself – within this system workers cannot participate in shaping the rules of the game, have little to no access to relevant information about their contribution, let alone make claims, all arguably crucial components of autonomy. 5
In what follows, I address, first, the concept of agency at a more theoretical level, to then outline some ways in which gig workers have expressed these tensions by reclaiming agency through tactics.
3. Agency at work
Several philosophical theories of human agency are available in the literature. Ideas of intentional causation, initiation and ‘sense of control’ (Pacherie 2007, 2) or power are recurring. Hurka (1987, 366), for instance, refers to agency as ‘causal efficacy’, defined as ‘making a causal impact on the world and determining facts about it’. Ci (2011, 261–262) describes agency as a ‘distinctively human form of meaningful causality’, in which power is ‘organized as subjectivity (or selfhood)’. According to Krause (2011), agency involves self-affirmation through action and efficacy. 6 As these definitions point out, agency describes the self being in charge to a certain extent, affirming itself and having a causal impact in the world, as opposed to being merely determined by external circumstances. In this sense, agency marks the difference between ‘happenings’ and ‘doings’ (Ferrero 2022, 1). You cannot be a mere cog in the wheel and exercise agency, as that would imply a lack of intentional causation, self-affirmation, efficacy and control.
On the other hand, there is no such thing as ‘absolute’ agency. A core aspect of agency is that it never occurs in a void; it always operates against a web of constraints. As Ci (2011, 267) observes, agency is ‘intrinsically constrained by forms of social external determination, which make up an irremovable background against which human agency is possible’. This ‘irremovable background’ is captured by Reader’s (2007, 589) account of personhood understood as involving both an ‘agent’ side and a ‘patient’ side. As persons, in fact, we act and are ‘acted on by another’ (ibid., 589, emphasis added). If agency and patiency are inherent to personhood, and if it is essential to the nature and conditions of agency to occur against irremovable external determinations, such determinations can in turn have more agency-constraining or more agency-enabling effects. That is, one can exercise more or less of one’s agentic capacities – and hence experience more or less ‘patiency’ – depending on how the external conditions are prone to facilitating or hindering these capacities.
One context in which the inevitable friction between agency and social determinations is particularly salient is work. Whenever we contribute to social cooperation through labour, we experience a tension between our agentic capacities and, among other things, 7 work’s organizational constraints. The latter can be said to include a variety of factors: the division of labour, organizational norms, role expectations, hierarchical structures, the way positions are conceived, productive goals, employers’ demands, managerial practices, the company’s culture, and so on (the list could be longer). I suggest calling the kind of agency specifically deployed when participating in social cooperation through work ‘contributive agency’. Contributive agency refers to the idea that, as participants in cooperation and contributors to value-creation, we demand to exercise our agentic capacities with regard to the content, form and terms of our contribution. At the most basic level, 8 the content of our contribution can be said to refer to the tasks we perform – the ‘what’ of work – while the form refers to the way we perform them – the ‘how’ of work – and the terms refer to the duties, rights, and the broader normative expectations regulating our contribution.
Consider the example of a high-school teacher. She provides a service to a community within a cooperative system called higher education in exchange for money, within a set of rules and normative expectations. For example, she is supposed to teach every day at 8.30am except, say, Fridays. She is supposed to teach about certain subjects and show up on these days unless she has some justified constraint, and so forth. Within this web of constraints, however, the teacher exercises a degree of control on key aspects of her contribution – for instance, over what her classes will specifically be about, how to organize the teaching activity itself, how to deliver lectures, and so forth. Depending on the context, she may even be able to negotiate some terms of her contract (through a union or individually). Thus, we take it as a given that the teacher has no absolute control over her contribution. But we also consider it as a given that she does have some degree of control over it.
Now, if we accept it as unproblematic that the teacher does not, and should not, have absolute control over her contribution, at the same time we would find it objectionable if she had too little control. Imagine if it was school policy that the teacher would be randomly assigned new classes with very short notice and no possibility to negotiate these decisions, abruptly communicated via email, not being able to decline them without consequences. Imagine that the teacher would not have access to relevant information beforehand – for instance, how many teaching hours they will require, for how many weeks they will last or how many students they will involve. Imagine that if she declined, she would be downgraded in the school ranks and lose relevant benefits. We would most likely consider these constraints to the teacher’s contributive agency objectionable. Yet, why exactly object to them?
We may provide reasons focused on consequences, claiming that putting the teacher in that position will undermine the quality of her teaching, for instance. We might say that it would be detrimental to her well-being because of the increased working stress this policy implies. We may argue that significantly hindering the teacher’s agentic capacities at work may affect her broader agentic capacities also outside of work. 9 While these answers make sense, looking exclusively at the possible consequences as a ground for objection would make them susceptible to all sorts of counter-claims about possible alternative consequences (say, management claiming the higher efficiency ensured by this policy). We may thus look for more universal reasons for objection. We may point out that each of us has a fundamental interest in exercising more rather than less agency in the relevant domains of life – work being one of them. We may appeal to the intuition that all workers should exercise at least some degree of agency in their contribution, and that constraining such agency beyond a certain threshold is inherently objectionable because it clashes with our fundamental agentic capacities as persons. We may acknowledge the need for organizational constraints and for leaving room for workers’ agency, pointing to a sort of ideal balance between the two.
As the example shows, organizational forms and managerial practices constitute a relevant external determination to contributive agency. They can be said to have different degrees of impact on workers’ agency, along a spectrum. I suggest thinking of this spectrum as involving a more enabling and a more constraining end, with several nuances along the continuum. One may be in a position to exercise more or less contributive agency, thus leaning toward one end or the other of the spectrum based on the organizational context. Hence, given the variety, complexities, and internal differentiation of work settings, contributive agency is better thought of in degrees rather than in all-or-nothing terms. Contributive agentic capacities are not to be thought of merely as ‘properties’ of agents, but as organizationally mediated capacities, meaning that they can be more or less enabled and more or less constrained depending on the contributive context in which they are deployed. Thus, I take it as consistent with our moral intuitions about agency that it is objectionable if organizational forms and managerial practices significantly hinder workers’ contributive agentic capacities, in the sense of exercising some degree of control (including self-affirmation and efficacy) over the content, way, and terms of their contribution. In the specific context of AM, as the discussion will show, some of AM’s features have agency-constraining effects, which workers try to reappropriate toward an agency-enabling direction through tactics. The latter testify to this unresolved movement between agency and patiency under AM. We are ‘acted on’ also by algorithms. Yet, as the examples of workers’ tactics show, we ‘act on’ algorithms too. Let me now turn to the examples.
4. Tricking the algorithm: counter-tactics, alternative bargaining and collective guesswork
Media reports and empirical findings in the social sciences have widely documented practices aimed at ‘tricking the algorithm’ and escaping some of AM’s constraints by gig workers, turning the expected functioning of the algorithm to their advantage. Scholars have suggested several ways to categorize them. For instance, Brunton and Nissenbaum (2015, 62) have referred to similar phenomena of purposeful interference into algorithms as ‘obfuscation’, described as a ‘weapon for the informationally weak’. Ferrari and Graham (2021, 2) have referred to them as ‘fissures’, defined as ‘moments in which algorithms do not govern as intended’, while Bambauer and Zarsky (2018) have phrased them as an ‘algorithmic game’. Inspired by de Certeau’s (1984) concept of ‘tactic’, I will refer to them as ‘algorithmic counter-tactics’ in order to capture their circumstantial and reactive nature within the context of AM (see section 6).
One example is data manipulation – for example, GPS localization – by using fake accounts and modified software (Ferrari and Graham 2021; Möhlmann and Zalmanson 2017). This allows gig workers to make informed decisions on which rides to accept, because platforms do not make information about distances and earnings available before accepting the ride, a feature often criticized by drivers (Amorim and Moda 2020, 112; Rosenblat 2018). ‘Information asymmetries’ (Mateescu and Nguyen 2019, 6) like these are created by platforms such as Lyft and Uber through passenger acceptance policies that hide information about how much drivers will earn before accepting a trip. This way, workers may end up taking on trips that are economically disadvantageous to them (Mateescu and Nguyen 2019, 6). Therefore, manipulating the data flow enables workers to prevent sanctions – which include bans or temporary suspensions from the platform – in case of order cancellations. Note that this tactic seems far from sporadic: almost 40 percent of drivers in China reported having downloaded bot applications and registered their cars on several devices (Chen 2018, 2705); 81 percent of drivers declared they had manipulated orders; and 61 percent of drivers said they were aware that their colleagues did the same (ibid.; Ferrari and Graham 2021). Managerial ‘counter-counter-tactics’ often follow, for example, through sensors capturing ‘spoof sensors’ creating mock GPS locations.
This example is relevant since it affects the epistemic conditions to exercise the relevant contributive agentic capacities – that is, information and resources for understanding and making decisions. Beyond the epistemic dimension, privacy concerns are raised as well. Pervasive data tracking is pursued through sensor intelligence and location intelligence, which allows the apps to track workers’ ‘GPS location, speed, and acceptance rate of customer requests’ (Möhlmann and Henfridsson 2019). Uber has a remarkable grasp on the geolocation of any active user, thanks to its pioneering yet ‘algopticon’-like ‘global grid system (H3)’ (Jamil 2020, 7). Control goes beyond real-time ID checks requiring drivers to submit a selfie for verification (Jamil 2020). Micro-sensors capture drivers’ overall mobility patterns, including working hours and driving habits, and motion sensors such as accelerometers detect driving behaviours through GPS receivers. This recalls what has been referred to as ‘datafication’ (Mayer-Schönberger and Cukier 2013). Without drivers’ (and passengers’) real-time data, these platforms ‘would not work’ (Sadowski 2019, 6). This means that workers’ (and customers’) data itself constitutes a key source of value extraction. 10
Another example is the manipulation of fares through the organization of collective log offs. It has been reported (Mamiit 2019; Ferrari and Graham 2021) that Uber and Lyft drivers at several airports have simultaneously logged off the apps for a few minutes to determine a temporary price surge. Surge pricing is generally a way through which platforms push workers to operate for longer hours in certain areas and times, through – often disattended – promises of more money (Rosenblat 2018). In the absence of a labour contract setting clear rules and expectations, companies use these subtle strategies to pursue the desired productive outcomes. For example, a driver who is about to log off the app will receive a notification like: ‘You’re $10 away from making $330 in net earnings. Are you sure you want to go offline?’ (Scheiber 2017). Data monitoring is thus not only used for tracking purposes, but also to influence workers’ behaviour (Mateescu and Nguyen 2019). Given their nature of psychological incentives influencing behaviour, they are often referred to as ‘nudges’. Some scholars have pointed to the manipulative nature of this practice. Susser, Roessler, and Nissenbaum (2019, 6) see it as ‘hidden influence’, in that it operates ‘by targeting and exploiting decision-making vulnerabilities’ – for example, workers’ desire for earnings (Susser, Roessler, and Nissenbaum 2019, 8). The idea itself of manipulation clashes with that of agency, as by definition it interferes with agentic control. In this sense, price surging counter-tactics may be seen as ways to regain some control.
Tactics are also more directly aimed at escaping surveillance. Consider the case of platforms setting up a system of automated surveillance by taking screenshots of workers’ computer screens at random intervals to verify whether they are actually working. Wood et al. (2019, 64) have documented a worker, Simon, saying: ‘It comes once every 10 minutes. Once it has shown up in a 10-minutes block you leave…you have 9 minutes to do everything that is totally non-related to work’, and another one, Nicole, saying: ‘Since I’m technical I connect my laptop to my TV…so I have two screens, I’m watching YouTube while I’m working on the platform…because the screenshot is only for the main [monitor]’. These examples show workers’ response to surveillance through tricks that allow them to regain some control over the information about themselves they are willing to share. What may be called the protective dimension of contributive agency seems to be at stake here, even though it is also about reclaiming temporal agency – control over how to manage one’s own working time.
A challenge for workers in the gig economy is to organize strikes despite having few chances to meet their colleagues, given gig labour’s lack of structural conditions for consistent peer sociability. Unlike more traditional workplaces where workers regularly share working time and space – which has historically facilitated the organization of collective action – algorithmically managed gig labour is based on a disaggregated workforce. While not making unionization impossible, this still functions as a structural barrier. Workers have to solve the puzzle of how to mobilize as a disaggregated workforce. These organizational challenges to unionization have prompted gig workers to find creative solutions, by repurposing certain features of the platforms. For instance, UberEats delivery workers in London have organized a protest by spreading the word through chains of food orders. When the food was delivered to them by fellow riders, they informed them about the protest, inviting them to join and do the same (O’Connor 2016). This provides another example of attempts to loosen the grip of AM’s constraints to contributive agency, in what may be called the relational (sociability) and participatory (collective action) dimensions.
People managed by algorithms do not know how the system works: it is, to them, a ‘black box’. In the lack of transparency and of consistent opportunities for sociability, workers engage in practices of collective guesswork to figure out how the system works via online forums (Ferrari and Graham 2021; Möhlmann and Zalmanson 2017). A great deal of time and energy is devoted to trying to make sense of platforms’ obscure decisions and mysterious functions through a sort of extra epistemic labour. Consider that the online forum Uberpeople.net currently has 7.5 million posts, many of which are attempts to figure out how decisions and fares are determined and to catch up with abrupt app updates. Complaints about sudden changes in fares are common, just like the perceived unfairness of the rating system and the lack of company support. Möhlmann and Zalmanson (2017, 10) have found that ‘drivers dealing with uncertainty and information asymmetry developed theories, stories and urban legends regarding the system and its reasoning’. In this way, they seem to implicitly reclaim contributive agency in the epistemic dimension (access to information about aspects of their contribution).
5. Contested aspects of AM
I suggest that these responses to aspects of AM share an ‘oblique’ attempt to regain contributive agency against some of AM’s constraints. This section spells out more clearly the nature of these contested aspects. This is helpful to prepare the ground for identifying some positive normative criteria of desirable forms of agency-enabling AM (section 8).
Epistemic helplessness: informational asymmetry and algorithmic opacity
The unilateral, abrupt, and inscrutable nature of AM’s decisions limits workers’ ability to make their own choices about their contribution. This includes things like how shares are determined, unpredictable surge pricing, and details about the rides. Sudden platform updates replace timely, explanatory communication of decisions. For instance, UberEats opened in London with an initial salary offer of £20/hour. In a few weeks, the formula abruptly changed to £3.30 per delivery plus £1/mile minus a 25 percent ‘Uber service fee’, which was later eliminated (O’Connor 2016). Following Creel (2020), this may be referred to as AM’s ‘functional opacity’. For instance, drivers are not informed about the destination of the customer before accepting rides. Once they log in, they only have a few seconds to accept or decline an order in the absence of further information, for example, a drive might take 2 hours or 10 minutes, with related differences in earnings. Algorithmic opacity undermines workers’ understanding of their work and thereby hinders the epistemic components of their contributive agency. In fact, workers are deprived of the necessary information to make sense of their contribution. As Brunton and Nissenbaum (2015, 2–3) note, ‘obfuscation may serve to counteract information asymmetry, which occurs when data about us are collected in circumstances we may not understand, for purposes we may not understand, and are used in ways that we may not understand’.
Behavioural control
As observed earlier, data collection not only monitors and surveilles workers, but also actively influences their behaviour (Eyert et al. 2018). Not relying on contractual expectations, companies follow alternative routes to obtain the desired productive outcomes, like psychological tools influencing workers’ behaviour, such as nudges and sanctions. An example is surge pricing, that is, messages promising higher pay in certain moments and areas to incentivize drivers and riders to work for longer hours. When they log off the app, a notification invites them to work more in the expectation of higher earnings. Similarly, workers complain about sanctions following the automated performance assessment or ‘rating system’ (Mateescu and Nguyen 2019). Reasons for low rating and hence possible account deactivation can be as trivial as the type of music in the car, an unpleasant level of conversation, the lack of particular comforts (Jamil 2020, 11). Note that workers can spend months trying to improve their rating by 0.1 (Jamil 2020). While traditional taxi drivers are ‘free to sit behind the wheel with a grumpy face’, for Uber drivers it could mean a significant loss of income (Royakkers et al. 2018, 138). Not to mention the discriminatory downsides of ratings, which outsource to customers the biases once held by employers – especially worrisome in the absence of anti-discrimination protections (Rosenblat 2018, 155; Rosenblat et al. 2017) and considering that most gig workers are non-Whites (Pew Research Center 2021), more likely to feel unsafe or harassed on the job (ibid.).
Corporate surveillance
AM has allowed increasingly pervasive forms of surveillance. Companies engage in data collection and constant monitoring of workers’ habits and behaviours through sensor and location intelligence, but also through more explicitly invasive monitoring. Another example is Deliveroo’s algorithm sending workers personalized monthly ‘service level assessments’ on their average ‘time to accept orders’, ‘travel time to restaurant’, ‘travel time to customer’, ‘time at customer’, ‘late orders’ and ‘unassigned orders’ (O’Connor 2016). The algorithm compares each courier’s performance to its own estimate of how fast they should have been. The Financial Times reports a worker’s assessment: ‘Your average time to customer was less than our estimate. Your average difference was −3.1 minutes’ (ibid.). Despite the obvious historical and contextual differences, strict time monitoring for profit maximization recalls Taylorist forms of labour management. 11
Lack of sociability and of the organizational conditions for unionization
The structure of labour in the gig economy has encouraged the erosion of the organizational conditions for unionization, promoting workers’ isolation. This is concerning not only for its impact on workers’ well-being (Wood et al. 2019) and alienation (Vredenburgh 2022), but also for their contributive agency. Workers’ attempts to organize collective action despite this barrier, as well as the practices of collective guesswork in non-working hours, can be seen as attempts to resist this isolation and powerlessness.
6. Tactics and contributive agency
In this context, it is helpful to read the notion of contributive agency through the lens of Michel de Certeau’s concept of ‘tactic’, which he defines in opposition to ‘strategy’. On de Certeau’s view, strategies are planned actions resulting from mastery of time and space, ‘sustained by the power to provide oneself with one’s own place’ (de Certeau 1984, 36). By contrast, tactics are acts that ‘produce without capitalizing’ in everyday contexts (ibid., xx), lacking a ‘proper locus’. The space of a tactic is thus ‘the space of the other’, in that it operates by playing ‘on and with a terrain imposed on it and organized by the law of a foreign power’ (ibid., 37). Operating by definition against a ‘foreign power’ – certain AM features, in this case – tactics can be thought of as always being ‘counter-tactics’, in a sense. Lacking its own place, lacking a view of the whole, (…) limited by the possibilities of the moment, a tactic is determined by the absence of power just as a strategy is organized by the postulation of power. (ibid., 38) [A tactic] does not have the options of planning general strategy and viewing the adversary as a whole (…). It operates in isolated actions, blow by blow. It takes advantage of ‘opportunities’ and depends on them, being without any base where it could stockpile its winnings, build up its own position, and plan raids. What it wins it cannot keep. This nowhere gives a tactic mobility, to be sure, but a mobility that must accept the chance offerings of the moment, and seize on the wing the possibilities that offer themselves at any given moment. It must vigilantly make use of the cracks that particular conjunctions open in the surveillance of the proprietary powers. It poaches in them. It creates surprises in them. It can be where it is least expected. It is a guileful ruse. In short, a tactic is an art of the weak. (…) [T]rickery is possible for the weak, and often it is his only possibility, as a ‘last resort’ (…). (de Certeau 1984, 37–38)
Overall, tactics’ rationale is that they seem to allow agents to operate within the contested system because of the advantages it provides (e.g. economic gains), without having to bear the costs of quitting. Since the terms of cooperation are deemed questionable, agents’ participation in it looks conditional to the enactment of these tactics. They allow workers to keep a distance from the system while remaining in it, when engaging in conflict is not an option in the absence of claim-making opportunities. In this sense, gig workers’ attempts to trick the algorithm recall Lüdtke’s account of workers distancing themselves from the ‘demands of the supervisors and from the constraints of the work process without directly fighting them’ (1986, 83). In a situation of shrinking agency, tactics constitute an attempt to regain it – hence, it is an ‘art of the weak’.
These tactics and practices entail benefits and costs for workers, and it would be problematic to think of them as ‘empowering’ (Ferrari and Graham 2021, 827). Consider, again, the organization of collective log offs to influence surge pricing. It requires a certain amount of vigilance, group concertation efforts, not to mention the opportunity costs. The costs include the psychological investment required to engage in these behaviours as well as the exposure to risks of managerial ‘counter-counter-tactics’. The effectiveness of this form of tactical extra labour is largely limited, as it does not substantively alter the overall functioning of the system. Much like de Certeau’s tactics that ‘produce without capitalizing’, it does not lead to lasting change. Similar to factory workers’ resistance, in these practices ‘victory perpetually coexist[s] with defeat’ (Lüdtke 1986, 86). In the long run these tactics more likely inspire ‘counter-gaming’ responses from management, which ultimately perpetuate if not exacerbate the contested features of the system. They prompt a ‘moves and countermoves’ dance where companies respond by further refining their tools of algorithmic control (Bambauer and Zarsky 2018, 1), within a ‘cat-and-mouse game without a clear winner’ (Ferrari and Graham 2021, 9). Far from idealizing them, tactics carry a risk of further perpetuating the problems they try to challenge.
While normatively insufficient and politically limited, these practices can nonetheless provide us with some direction to think about the objectionable aspects of the system and guidance toward the assertion of positive ideals. If my reading of tactics as exhibiting the agency-constraining aspects of AM is accurate (see for instance Möhlmann and Zalmanson 2017) and in line with our considered moral judgements (see section 9), we can turn in a positive direction. Let me first spell out the dimensions of contributive agency.
7. Dimensions of contributive agency
Keeping these examples in mind, I now articulate the relevant dimensions of contributive agency. Recall the discussion of section 3, about contributive agency and its external determinants. We may consider several types of external determinants within a cooperative setting – some have to do with how organizations distribute access to information among their members (think about the teacher’s lack of information about the newly assigned classes), while others have to do with how much room they allow for voice and claim-making (think about the teacher’s lack of room to contest or negotiate the new responsibilities). Consider the following dimensions of contributive agency: epistemic, participatory, relational and protective. When trying to exercise our agentic capacities at work, we are demanding access to relevant information about our contribution and its terms (epistemic dimension), to have voice and room for claim-making (participatory dimension), to be treated with dignity, have our contribution recognized, as well as relate to and connect with peers (relational dimension), and to exercise control over information about ourselves (protective dimension). In what follows, I further articulate these dimensions.
Epistemic
The epistemic dimension of contributive agency affected by AM concerns access to information about workers’ contribution, specifically with regard to the algorithmic functions affecting them. For instance, the lack of information about rides prevents them from making informed decisions, impacting their ability to keep working for the platform itself (e.g. they are at risk of being banned or sanctioned if they refuse rides). Lack of ‘functional transparency’ (Creel 2020) prevents them from understanding the conditions in which they work, and epistemic helplessness undermines their ability to control significant aspects of their contribution. The concepts of information asymmetry and algorithmic opacity capture the epistemic nature of the constraint on workers’ agency.
Relational
The relational dimension of contributive agency can be said to involve three main aspects: dignity, recognition, and sociability. Dignity is involved insofar as workers express a sense of being treated as ‘mere means’ by AM, exemplified by the slogan ‘We Are People, Not Uber’s Tools!’ (O’ Connor 2016). While every employment relation can be said to be inherently instrumental to an extent, there is a limit to workers’ instrumentality, that is, dignity and respect. Relatedly, claims of recognition stem from the perception of not being recognized as fully-fledged contributors to value-creation. A clear example is companies’ representation of drivers and riders as partners or independent contractors or even as customers instead of workers to avoid labour regulations, thereby refusing to recognize the nature of their contribution. The slogan ‘Be Your Own Boss’ feeds into a narrative of autonomy and horizontality which does not match the reality for gig workers, who provide services in exchange for earnings and are expected to follow rules others have determined.
Finally, the relational dimension involves concerns for workplace sociability. The erosion of workplace conditions for sociability impacts workers’ ability to organize collective action to have their voice heard. By structural isolation I mean the organizationally induced lack of access to the ‘social goods’ relevant to workers’ agency – for example, connection to peers. Sociability can be linked to a sense of being part of a broader community of peers. Several benefits go with it: ‘[R]elatedness, social support, organizational commitment, organizational identification, psychological ownership, and socio-moral climate’ (Schnell, Höge, and Weber 2019, 165). Estlund (2003) highlights that even though workplace relationships are mostly unchosen, they can be very successful in providing a sense of connectedness that other, more freely chosen associations do not provide: ‘more people say they get “a real sense of belonging” among their co-workers than among any group other than family and friends’ (Estlund 2003, 22–23). This, she argues, impacts civic engagement and democracy. Hence, the structural isolation suffered by gig workers can be said to undermine their contributive agency because it prevents the organizational conditions for collective action, but also because of the possible effects on civic engagement (Estlund 2003) as well as well-being (Wood et al. 2019) and their ‘practical orientation’ (Vredenburgh 2022).
Participatory
The participatory dimension is involved insofar as workers demand room for claim-making and for their voice to be heard. This is undermined not only by the lack of suitable conditions for unionization just discussed (relational dimension), and by the systematic exclusion of access to information (epistemic dimension), but also by the unilateral, often abrupt and opaque way in which companies make decisions affecting workers’ activities. Participation is a salient strand of contributive agency in that it gives a sense of co-shaping the terms of workers’ contribution, in contrast to just passively being subject to them.
Protective
Contributive agency also involves a protective dimension, meaning control over the data and information workers are willing to share with companies. The tactics manipulating the data flow limit companies’ intrusive data collection and related sanctions. Likewise, surge pricing manipulation allows workers to regain some control over nudges. Resistance to surveillance, such as monitor screenshots, also signals a demand to protect a space of agency from intrusive monitoring. Notice, again, the informational asymmetry: companies have access to virtually unlimited information about workers, but workers know very little about how companies operate.
This is not meant to be an exhaustive list. 12 This multi-layered notion of contributive agency aims to encourage a more granular understanding of workers’ demands and to reflect on some normative aspects of the examples discussed, as well as the broader implications of AM for workers.
8. Enabling contributive agency
Which criteria should algorithmic management meet to enable workers’ contributive agency?
In the epistemic dimension, we have seen how epistemic helplessness with regard to AM’s functioning and the obstacles to making informed decisions significantly affect workers’ contributive agency. Enabling contributive agency in the epistemic domain thus would require increasing transparency and providing explanations about the functioning of the algorithm and decision-making processes. An agency-enabling form of AM in the epistemic dimension provides explanations, increases transparency, and reduces informational asymmetry.
To recall, the relational dimension of contributive agency refers to the quality of treatment of workers (dignity), workers’ recognition, as well as strategies to reduce or eliminate workers’ isolation and improve peer connection (sociability). Structural isolation makes it significantly harder for workers to exercise their contributive agency, which contributes to powerlessness. Given that their contribution is often not recognized as ‘work’, this entails structural impediments to contributive agency in that workers cannot rely upon labour guarantees, besides having the very status of their contribution misrecognized. Likewise, resistance to corporate surveillance may be seen as a demand not to be ‘infantilized’, as AM’s hyper-management features do not seem to treat workers as competent, responsible adults (consider the example of the random screenshots monitoring their activity).
The participatory dimension would require the elimination of obstacles to participation, as demands for voice remain unsatisfied in current forms of AM. Such obstacles include the recognition of the status of workers, the overcoming of their structural isolation (both intertwined with the relational dimension), and the introduction of structural ways to actively participate in shaping the rules of the game. Unionization is one such way, while a more demanding reading of the participatory dimension would require forms of workplace democracy. A substantial revision of AM decision-making structures may even lead to gig workers’ cooperatives replacing investor-owned companies, ensuring greater agentic capacities in the participatory dimension. 13 Finally, the protective dimension points to the elimination of corporate surveillance as well as manipulative influence.
Hence, contributive agency-enabling AM requirements include the following criteria: the reduction of informational symmetry and algorithmic opacity through increased transparency and explanations; recognition and the organizational enablement of dignity and sociability; greater structural opportunities for voice and claim-making; the elimination of surveillance and behavioural influence. Note that each requirement is intertwined with the others and cannot really be thought of in isolation. In order to enhance the participatory dimension, for instance, it is key to enhance the relational dimension too, in the form of increased organizational conditions for sociability. The same goes with the epistemic and the participatory dimensions, as one cannot truly participate without access to relevant information. Underpinning contributive agency, then, is a complementarity of its dimensions.
It might be argued that this is too demanding. Companies have efficiency priorities, and trade-offs with them are inevitable, especially in an aggressively competitive market. Whether making organizations more enabling of contributive agency is compatible with current or desired productivity levels remains an open, largely empirical question. Nonetheless, at least beyond a certain threshold, reasons of economic efficiency should not undermine contributive agency. Consider the elimination of surveillance. Data extraction is key to these gig economy platforms’ business model (Couldry and Mejias 2020). Eliminating it might lead to challenging the rationale behind their economic success at its core. Yet the fact that something is rentable does not mean that it should be. 14 Corporate surveillance significantly weakens workers’ contributive agency beyond acceptable levels in that it prevents them from exercising control over the information they are willing to share with the platforms – where mere consent is not unproblematically resolutive (see Nissenbaum 2011) – and makes them susceptible to manipulative influence (Susser, Roessler, and Nissenbaum 2019) thereby considerably undermining contributive agency. Contributive agency, at least below a certain threshold, may be seen as setting a moral limit to economic efficiency – the pursuit of the latter should not come at the significant expense of the former.
But where exactly is this threshold? How to draw the line between acceptable and unacceptable constraints to contributive agency? To help answer this question, we may identify two levels of realization of contributive agency-enabling AM in a non-ideal world: affirmative and transformative. 15 The affirmative level operates by refraining from hindering contributive agency in the relevant dimensions or by mitigating the constraints to agency, rather than structurally eradicating them. The transformative level operates instead by changing the underlying structures and positively enabling contributive agency. One may consider the affirmative level as the minimum threshold below which agency-constraining features are not acceptable, outweighing efficiency trade-offs, and the transformative level as the most desirable scenario.
For a practical example, consider the participatory dimension. The affirmative level of realization would require mitigating obstacles to unionization by recognizing gig workers as employees. Unionization opens the room for claim-making and voice, although without directly changing the decision-making structure of the organization. That marks the threshold below which contributive agency is unacceptably constrained, given the stark lack of opportunities for voice in AM. The transformative level of realization would demand instead to introduce stronger forms of democratization, changing AM decision-making structures.
These criteria can facilitate actors’ practical orientation by providing a standard with which to measure the distance from reality and adjust the latter accordingly in degrees. The contributive agency approach holds AM organizations accountable for their managerial practices, by requiring certain agency-enabling organizational means to be exercised. As for the implementation, philosophical inquiry can help identify and clarify the normative concerns raised by organizational practices, bringing them to scholarly attention and critique. However, it is up to the actors involved in these practices and policymakers to ultimately specify the content and practical details of each requirement in the various contexts.
To better grasp the contribution of this approach to the debate, it is worth outlining some differences from the virtue ethics view (Gal, Jensen, and Stein 2020). Whereas the virtue ethics approach focuses on workers’ pursuit of ethical excellence and the building of personal character, the contributive agency approach is centred on the organizational conditions for agency. It thus shifts from concerns for ethical virtue to concerns for workers’ capacities to act. While being distinct and not necessarily incompatible approaches, in a sense, organizational conditions for agency can be said to bear some sort of priority over the pursuit of ethical ends of excellence. The latter presupposes the exercise of agency, while the contrary is not true. In fact, we can exercise control over our contribution without excelling in personal character building. 16 In fact, while we might still be living a decent life without pursuing excellence and ethically flourishing, it is doubtful that this is attainable without exercising our very capacity to act in the relevant domains of life (work being one) in the first place. Second, the cases discussed earlier suggest that more than asking to be virtuous and build character, workers seem interested in increasing their agentic control over their contribution. Finally, this approach is more compatible with pluralistic conceptions of the good life. In a culturally complex and value-pluralistic society, beliefs about ethical excellence may largely differ. By contrast, regardless of people’s different ethical goals and concerns about the good life, everyone can be said to have a fundamental interest in increasing their agentic capacities. These differences do not make the virtue ethics approach dismissible, the two approaches address different problems.
9. Objections
One worry about an ethnographically sensitive approach to philosophical inquiry is that it may lead to an undue generalization of contingent behaviours, by overly charging them with normative value. For example, one might object that just like workers’ counter-tactics, students’ plagiarism in exams might signal a normatively significant claim to regain agency against the constraints of academic control, or that wealthy people engaging in tax evasion seek to regain agency against the constraints of the local fiscal system. 17
The objection misses a relevant point: as pointed out in section 1, agents’ behaviours are only a starting point of the reflective equilibrium (see Herzog and Zacka 2019). They cannot, by themselves, ground normative generalizations. The reflective equilibrium requires us to assess the coherence of our considered moral judgements. In this case, I have considered the experiences and behaviours of agents involved in normatively sensitive contexts only as an entry point – discussing their coherence with our considered moral judgements about the exercise of agency in organizational contexts. The intuition with which these practices cohere is, I have argued, the idea of contributive agency. Given their power over individuals, organizations are prominent external determinants capable of considerably constraining or enabling members’ exercise of that agency. 18 Being organizationally prevented from the exercise of the epistemic, relational, participatory and protective dimensions of contributive agency hinders relevant capacities to act in cooperative settings, which is especially concerning given a lack of significant alternatives.
Hence, not all ‘tactics’ are of normative interest. Plagiarism significantly conflicts with our considered moral judgements about intellectual property, as does tax evasion with respect to basic duties of fiscal justice. Given that they do not fit our considered moral judgements, they are not a sound inspiration for norms. Workers’ tactics, as discussed earlier, on the contrary, in my interpretation (in line with Möhlmann and Zalmanson 2017) signal an attempt to regain agency which fits considered moral judgements about our agentic capacities in social cooperation, captured by the concept of contributive agency. 19 That said, more stringent conclusions about contributive agency would require further systematic development. As stated in section 1, the ambition of this article is more limited. It essentially accounts for a normative exploration as a first step toward a systematic theory of contributive agency, yet to be fully developed.
Another objection notes that workers can always opt-out: after all, they are not coerced to stay in the system. This objection overlooks two problems. First, AM tools are increasingly pervading all sorts of workplaces, even the most ‘traditional’ ones (especially since the pandemic), in a way that we do not have reason to think will stop in the future (Aloisi and De Stefano 2022). Thus, ensuring agency-enabling organizational structures does not seem merely a problem of a particular company or sector that one may choose to leave at any time. Far from being resolutive, opting out of an organization might mean having to enter another with similar characteristics.
An analogy with data privacy can further illustrate this point. If popular social media platforms heavily rely on data extraction and behavioural influence, opting out of a particular platform (say, Facebook) does not really make the issue go away, as data extraction now characterizes the internet in pervasive ways (Véliz 2021). Second, in 2021 it was estimated that 16 percent of Americans have worked for a gig labour platform, most of whom of Hispanic (30 percent), Black (20 percent) and Asian (19 percent) ethnicity (Pew Research Center 2021). While data on the number of people working in the gig economy on a global scale is difficult to obtain (Heeks 2017), the growth of this sector in the last few years is ‘exponential’ (Abraham et al. 2019; see also Heeks 2017) and expected to increase even more (Kuek et al. 2015). The gig economy tends to attract the most socially vulnerable populations, as for more than half of workers (58 percent) the earnings received through gig labour were ‘essential’ or ‘important’ in them meeting their ‘most basic needs’ (Pew Research Center 2021, 31). Finally, one limitation of the opt-out argument is the cost of exiting. With reference to data privacy, Brunton and Nissenbaum (2015) have observed that ‘any real opt-out policy’ would have to allow for ‘choices that lay between the extremes of refusal and compliance’. By analogy, the contributive agency approach may be seen as providing changes between the extremes of mere refusal and mere compliance to AM’s features.
10. Conclusion
The contributive agency approach provides a way to overcome the normative tensions outlined above. If consistently realized, workers would be recognized as workers and not misrepresented as partners, customers or independent contractors. They would gain more control over their contribution, limiting the ‘hyper-managing’ tendencies of current forms of AM. Finally, increased organizational means for workers’ agentic capacities would provide more room for autonomy, in the sense of participating in norm-building processes about their contribution. Overall, AM need not be inherently incompatible with contributive agency. If AM were reorganized to enable rather than significantly hinder it, AM’s potential benefits (Lamers et al. 2022; Wood et al. 2019) would be made more available to workers as well.
Footnotes
Acknowledgements
I am thankful to audiences and organizers of the Annual Meeting at the Centre de Recherche en Éthique, Chaire Hoover – Université de Louvain, and the UNC Parr Centre; the workshop ‘Transformations of Work’ at the University of Barcelona; the annual congress of the Société de Philosophie du Québec, ‘Philosophie de l’économie’ panel; the Radboud Centre for Philosophy and Society Workshop for helpful questions and comments. Thanks to François Claveau, Thomas Ferretti, and Tamar Sharon for helpful comments on an earlier draft.
