“Have you learned your lesson?” Communities of practice under algorithmic competition

Abstract

Gig workers are typically thought of as individuals toiling in digitized isolation, not as communities of shared learning. While it’s accurate to say they don’t have the same information-sharing norms as people in traditional employment arrangements, some do gather, in part in digital communities. Online forums, in this space, have become popular sites for gathering, sharing information, and comparing practices. These behaviors provide an opportunity to examine gig workers as emergent communities of practice, and to analyze how work, identity, skills, and workspaces co-constitute each other as sociotechnical environments of work change. In this research, I examine workers’ interactions in an online forum, and focus on how they talk about scams. Analysis reveals that talking about scams is a way for workers to enact belonging in their community of practice. Victims are belittled by other workers, who frame vulnerability, and lack of foresight due to unfamiliarity with the forum itself, as a lack of authenticity. Repudiations are denunciations through which workers assert their belonging. These findings illuminate the practices of what I call “para-organizational” work, with implications for knowledge management in structures of algorithmic competition.

Keywords

Algorithmic management fraud gig economy platform safety scam security work

Introduction

Kari, a young worker in the ride-hailing industry, cruises in her car on the streets of Los Angeles. She’s a native of the city, born and raised there, and her knowledge of its highways and byways has been useful during her 3 years of driving for a company we’ll call RideHail. As she drives on a chilly February evening her phone, logged into the RideHail app and mounted on her dashboard, lights up with a ping for a driving job. The name of the passenger, however, looks odd: “Designated RideHail Ride.” Undeterred, she accepts the ride, and immediately gets a text message from the passenger, explaining the strange name. They actually work in customer service for RideHail, and, mentioning her high driver rating, let her know that she’s qualified for a $200 bonus. The employee asks her a few questions and they exchange several verification text messages, and finally she provides the details of her banking account so the bonus can be deposited. Later that night, settling into her living room with her laptop computer, she checks her online bank statement to see if the bonus is there yet—only to see with despair that the number next to the account is far lower than it should be. The transaction records tell the tale. Within the last few hours, in fact since the call with the customer support representative, dozens of fraudulent purchases have been made. Her account is drained. She’s been scammed.

She immediately emails RideHail Company and makes a complaint. While she’s waiting to hear if they can help her, she hops onto an online forum frequented by other RideHail workers, which she sometimes checks when she has questions or problems with the RideHail app. She creates a new post, and writes up her experience, titling the post “SCAM ALERT!!!” with warnings to the other workers on the forum, “Be careful!” and “Don’t fall for this!” She hits the button to post the message to the forum, so hopefully other workers will see the message. She begins getting replies right away, and opens the discussion to see how people are responding to her post. To her surprise, the thread is full of other workers, just like her, flinging vitriol in her direction: “Why the hell did you think it was a good idea to give a complete stranger your banking information,” “You fell for it, I think you deserved to get scammed,” and “Your warning is pointless. There’s a new victim here every week telling us all about how they gave all their money to a stranger claiming to be RideHail support.” A couple of comments were mixed in admonishing the insults, but by far most of the other workers in the thread seemed eager to cut her down.

As more employment arrangements shift to remote or “alternative” arrangements in the wake of COVID-19, it is important to understand what implications these new sociomaterial and sociotechnical arrangements have for community and learning, and what unanticipated consequences may arise out of the unique arrangement of work, platforms, devices, and associated actors. Workers’ signaling of expertise must contend with new arrangements of competencies and audiences in intertwined ecosystems of multiple platforms (Chan, 2019). To shed light on these questions, this research uses empirical qualitative methods and draws on literature in algorithmic management, online communities, and communities of practice, to investigate how workers communicate with each other in online forums, in particular about scam attacks. Findings include novel algorithmic skills demanded by platform work and new ways of creating and maintaining communities of practice in emergent sociotechnical contexts (Selbst et al., 2019). This paper concludes with an overview of the implications these patterns have for the safety and security of these communities.

Background

A spectrum of organizational control has emerged in the literature on algorithmic managements: as organizations use digital structures to wield increasing surveillance and control over gig workers (Mateescu and Nguyen, 2019; Rosenblat, 2018; Rosenblat et al., 2014; Ticona et al., 2018), workers also enact resistance and reclaim their own autonomy (Cameron, 2020; Jarrahi and Sutherland, 2019; Kellogg et al., 2020). However, little is known about the interstitial tissue in between the micro-meso levels: the behaviors, workers’ navigations of this spectrum of control: the peer-to-peer communications underpinning workers’ own strategies for understanding and negotiating algorithmic systems and their constituent parts, Workers in gig or alternative arrangements must constantly contend with opaque algorithmic systems (Bishop, 2019; Burrell, 2016; Cameron et al., 2021). Yet, unlike workers in traditional employee arrangements, there are no physically close co-workers of whom to ask questions. Economists and labor scholars point out that the digital distribution of workers creates a “lonely” environment of social isolation (Graham et al., 2017; Webster, 2016). Yet, recent popular press coverage of workers’ organizing efforts detail how they communicate and coordinate, sometimes in physical space (Qadri, 2020), and in part on online forums (Johnston and Land-Kazlauskas, 2018; Maffie, 2020). This work investigates the emergence of online gig-worker discussion, and asks how the structural characteristics of gig work, including the pressures of algorithmic competition, shape knowledge management practices.

In traditional employment models, workers interact with each other and with management in ways which constitute “organizational culture.” The organizations’ values and logics of action are imparted to employees, either explicitly through printed or digital materials such as pamphlets and handbooks, or implicitly, through the normative expectations of peers in the group (Kunda, 2009). Little is known, however, whether and how such community-based learning happens between gig workers, and whether and how spaces of online, collaborative learning and its activities co-constitute each other (Suchman, 1996). Several interrelated research questions at this nexus of workers, platforms, devices, and interactions drive this research. Do the technological and political infrastructures of gig work, including new sociotechnical and sociomaterial arrangements, evoke different patterns of interactions? What new skills are required for successful navigation of these spaces, and how do patterns of legitimation shift to recognize and incorporate new skills? Do these interactions constitute a form of organizational culture or community? These questions will be addressed using three interrelated theoretical frameworks: algorithmic skills, online communities, and communities of practice.

Algorithmic skills

Outside the aforementioned spectrum between organizational control and worker autonomy lies a world of ad hoc work, work that sits outside the core tasks assigned by algorithmic management yet is required to stitch algorithmic outputs into a workflow. Taking a sociotechical view People who work under algorithmic management, typically platform-based and “gig” work, are not simply passive recipients of computationally sorted tasks. While stark information asymmetries are used by platforms to manipulate the work environment (Rosenblat and Stark, 2016), workers themselves are active interpreters and negotiators of algorithms (Watkins, 2020). Algorithms, rather than rigid infrastructures, are unstable objects (Green and Viljoen, 2020). Workers take a hand in constructing what they mean and what function they perform. As the gig economy evolves across the globe, more attention is being paid to the labor that these workers do in interstitial spaces of sorting, seeking, and managing platform participation and legitimation, the “articulation” work (Kellogg et al., 2020; Star and Strauss, 1999) that assures the ongoing sustainability of these systems (Jackson, 2014). These include marginal tasks that are implicitly demanded by platforms, such as their structuralization of a professionalized appearance, and relationships with clients and third-party institutions associated with professional work. Care workers such as nannies and babysitters, for example, have been pushed by care-work platforms to perform visibility to institutions such as the IRS (even though carework in non-platformed formats have been traditionally “off-the-books” or cash-based), and develop strategies to “market” themselves to potential clients in order to gain visibility in the market for carework that the platform cultivates (Ticona and Mateescu, 2018). Mechanical Turk workers develop a range of strategies buffering the actual tasks of “cognitive piece-work” assigned by the platform (Gray et al., 2016), including monitoring MTurk dashboards, scrolling through pages of job postings, and using creativity and ingenuity to juggle vendor-management platforms (Gray and Suri 2019). In her analysis of critical skills used when workers interact with algorithmic systems, Hargittai found that awareness, understanding, and attitudes are key dynamics (Hargittai et al., 2020; Klawitter and Hargittai 2018). Groups of Uber and Lyft drivers have coordinated their platform log-in times to “manipulate” and drive up available fares (Sweeney, 2019). Amazon delivery drivers in American suburbs have been documented maximizing their chances of getting jobs from the platform’s location-based dispatch system, by hanging their phones in trees (Soper, 2020). AI researcher Meredith Whittaker described this process of workers “observing and assessing opaque algorithmic management systems” as a form of “folk tradecraft” (Whittaker, 2020). This research contributes to the discourse by examining how new skills emerge for navigating algorithmic work, and the organizational structures by which skills, identity, and work practice are co-constructed. This contribution is made through an analysis of how workers encounter, negotiate, and interact in online forums, in response to threats from third-party scammers.

Online communities

No ethnographer is immune from the influence that their own personhood and implicit biases have on the research they do or how their research subjects present themselves to the researcher (Meadow, 2013), and inarguably, the presence of biases remains in digital spaces, only becoming more invisible and perhaps insidious, as algorithmic recommendation direct attention toward content in ways that are the most beneficial for the platform owners. However, in some online forums, in particular Internet messageboards, online communities are a rich site for observation of social interactions without the vagaries of algorithmic sorting and metrics (Christin, 2020). An additional advantage of non-participant observation on digital forums is its unobtrusive nature (Holtz et al., 2012; Webb and Weick, 1979), with no researcher influencing the responses of participants.

Online forums, a popular site of sensemaking for gig workers (Gray and Suri, 2019), predate gig work itself. Online forums are a modern descendent of digital bulletin boards, which themselves are descendants of electronic listservs. The earliest online forums, like Usenet networks, date to the 1970s. Key for our purposes here, is that online forums are a common place for workers to turn when they’re trying to solve a problem or address a breakdown in the work system. Breakdowns are an ideal place to study the interactions between social and technical structures, as they make visible the articulation work which always underpins such interactions but which often go unnoticed (Bowker and Star, 1999; Jackson, 2014; Sachs, 2019). Workers across a variety of domains have turned to online forums to address breakdowns in their work, or to address breakdowns in knowledge practices, such as sharing knowledge which their organizations, either by accident or by design, overlooked or does not share. Da Cunha and Orlikowski (2008) examined how workers in the petroleum industry used online forums to engage in discursive opposition to recent organizational changes. UPS workers have long had their own online forum, according to its homepage established in 1999. An influential voice examining online forums for workers is Lilly Irani, whose work on Turkopticon, a platform for crowdworkers on Amazon’s microtask platform Mechanical Turk, helped to overturn the invisibility with which such workers are often treated (Irani and Silberman, 2013). In addition to messageboard-style forums like these, workers of all stripes also gather in groups on social media sites like Facebook and Reddit.

Methods for studying online communities have been popular in public health communications research, which seek insight into how people make sense of their condition, treatment options, and healthcare system in online spaces (Johnson et al., 2020). Nancy Baym has studied the role of online forums in the emergence of community since the mid-1990s, through the lens of computer-mediated communications (Baym, 1999). Online forums, she and others have found, primarily host three respective communication purposes: information sharing, coordination, and emotional support (Da Cunha and Orlikowski, 2008), often all three at once (Baym, 1999). Early studies of computer-mediated communications’ effects on group patterns of communication argued that these channels (in particular, early email lists) were determinant, and had a normative influence on group communication, that is, discrete norms for interaction emerged within the online groups as they spent time communicating via the computer-mediated channel (Postmes et al., 2000). In contrast, Pablo Boczkowski’s (1999) study of an Argentine Mailing List found that technological and social elements engaged in “mutual shaping,” where the meaning and function of the mailing list shaped, and was in turn shaped by, social processes.

To date, such findings from studies of online communities have been only shallowly applied to gig-worker communications. Gig-worker communications have taken a largely positivist track, with a dominant narrative around worker organization (Gray et al., 2016; Irani and Silberman, 2013; Lehdonvirta, 2016; Yin et al., 2016). For example, while Wood et al. (2018) recognize that workers’ fragmented experience of work may contribute to diverse subjectivities, their focus on online communities activity is restricted productive behaviors, including how workers go to forums to ‘‘obtain labor market opportunities and information, reduce social isolation, maintain professional norms, provide support and advice, and create feelings of community” (p. 9). Some research has recently emerged examining how algorithmic competition may influence whether and how workers share strategic information in online communities (Yao et al., 2021). For the most part, however, discourse lacks a deeper engagement with workers’ knowledge practices under algorithmic management, and how their communication patterns in online forums may be shaped by the materiality of the forum itself, or how these communication patterns may be informed by workers’ identities as platform workers.

Little is known about peer-to-peer interactions in these communities as a form of organizational or work-based knowledge management, and how the technical infrastructures of such spaces shape, and are shaped by, social structures, values, and interactions. The mediation of work by platforms is expanding, touching more aspects of professional and occupational life in processes like hiring (Ajunwa and Greene, 2019). Implicit, cultural conceptions of values are no exception, and are enacted and exercised through platform communications of workers. The governance of privacy in apps sold on the Apple Store and Google Play store, for example, takes place not just through explicit rules but also through community-based negotiation in developer forums: In their study of app developers on GitHub, a popular site for software engineers, Greene and Shilton (2018) found that developers building products for the Apple App Store and Google Play Store discuss and collectively negotiate the rules and norms set by these large organizations with which they have little direct contact. The labor politics of such spaces, such as the role of moderators and the labor exerted by users to manage these spaces, has also been recently examined (Matias, 2016; Seering et al., 2019). Recently, however, examining online communities as spaces of learning has taken a back seat to questions of platform governance and content moderation on social-media platforms (Gillespie, 2018).

In this article, I am concerned with the intersection of online forums and gig work platforms. I’m particularly interested in how algorithmic management creates groups of workers who labor, ostensibly, for an organization, but whose “membership” in that organization is limited or liminal, and the knowledge-management practices of these workers. Gig workers’ primary contact with these organizations are so digitized, dictated by their algorithmic bosses, that they may never actually meet or interact with a human employee of the platform company. In one online forum I consulted for my dataset, one worker commented that they’d “been driving for [RideHail Company] over two years . . . the whole time I’ve met one person. the girl who handed me my [branded delivery bag, for food deliveries through the related app].” Another delivery driver had that beaten, replying, “I have never spoken to or met anyone at [RideHail]. . . they mailed me my insulated bag.” The implications of this intersection for organizational membership and related knowledge-management practices have gone largely unexamined in critical spaces. For insights into knowledge-sharing practices attending group-based informal learning for people who share an occupation, I turn to the literature on communities of practice.

Communities of practice

One way that organizational culture is analyzed is through organizational learning that takes place through “communities of practice.” “Communities of practice” (CoP) is a concept drawn from anthropology, used primarily in management and communications fields to examine how information-sharing and learning happen in group settings, outside the classroom in dynamic and multi-actor interactional processes (Lave and Wenger, 1991). These can include interactions with peers and with artifacts such as documentation, pamphlets, and manuals. The CoP framing has been applied to journalists (Meltzer and Martik, 2017; Schmitz Weiss and Domingo, 2010), creative workers (Comunian, 2017), and healthcare workers (Antonacci et al., 2017). The CoP literature emphasizes that learning is a social phenomenon situated in specific sociotechnical contexts. Applying the theoretical framework in a new setting and examining the interactions between gig workers in online forums can illuminate the relationship between information, group practices, and organizational membership when the boundaries of organizations are blurred (Scott, 2004), and answer questions as to whether gig workers truly constitute a practice-based community. In this work, I contribute to the CoP literature by examining more closely the linkages between work, interactions, and identity (Bechky, 2006) in gig work, and in particular, workers who choose to interact with each other in groups via online forums.

Methods

For my observations of peer-to-peer worker communication patterns, I gathered data on one of the most active forums, which I’ll call GigWorker.net. This study and all methods were approved by my university Institutional Review Board. I collected data in the form of digital comments in discussions of scams targeted at RideHail workers. This forum is open to the public and requires no account approval or log-in in order to view any of the content posted. Using the native search function and searching for the term “scam” yielded several hundred discussions, or “threads.” Many of these threads were irrelevant to my research questions. These included when the term “scam” was used for issues outside of scope, including discussions of whether the gig economy itself is a scam, where RideHail’s pricing is a scam, whether new pricing models are scams. Some of these were found to be simply links to other threads, or to press stories, which were also removed, because these posts would not yield evidence about how members of the ridehail industry community interacted with each other to engage in knowledge-sharing practices. In setting aside threads about RideHail’s business model and pricing schemes being discussed as scams, I narrowed my search to focus on threads motivated by personal stories of attacks from third-party scammers that is, where scams targeted at RideHail workers were the primary topic of the first “post,” or “comment.” These threads were directly relevant to the research question, because these posts constituted evidence of how community members communicated laterally, sharing knowledge with each other, including direct feedback on each other’s work practices. Data collection was conducted in two rounds, first in 2019 and again in 2021. The resulting focused dataset was composed of 110 discussion threads with 1855 comments, with an average thread length of 16.68 comments. All threads in my final dataset were made between 18 October 2014 and 2 January 2021. To protect participants’ privacy I have ensured that no usernames or “handles” are listed in this paper. The wording of all quotes has been altered slightly, while ensuring consistency of meaning, to minimize the possibility of identification.

I chose not to analyze discussions on social-media sites like Twitter and Facebook because of their algorithmic sorting structures. Generally, posts on Internet forums are comparatively static: once they’re posted, they remain accessible by following the topic branches or using the native search function to search by keyword. Posts on forums can be accessible at least as long as the web host pays for hosting costs, and even defunct web forums can be found on web archiving services like the Wayback Machine. This is in contrast to social media platforms and their sorting and targeting algorithms. These privilege new posts or posts that are perceived, by a variety of metrics, to be of interest to the user. Older, or “less relevant” posts, or those with less engagement, are pushed down. This means that only a select number of comments can be easily accessed. For these reasons, I focused my data gathering on forum-style messageboards.

An abductive approach to coding was used to analyze the posts (Tavory and Timmermans, 2014). All threads were read and analyzed for thematic patterns, while I concurrently consulted with the literature on online communities, communities of practice, and algorithmic skills. A limitation of this method is my single-coding approach. As a result, these codes have not been verified by a second coder, and so may be less robust than a double-coded dataset. Drawing on my initial reading of the data and iterative consultation with discourse, I created a set of five codes focusing on members’ attitudes and forum behaviors toward other members in response to scam-related discussions. Armed with this codebook, I then re-read all of the relevant threads for the most dominant and relevant themes.

Research setting

Putting the “place” in this “workplace,” I’ll first provide some thick description of the site itself, to give a better idea of the environment where workers are choosing to spend their time.

GigWorker.net’s aesthetics are inherited from the earlier days of messageboards. The homepage is simple and geometric, with a neutral color palette of pale blues and dark grays. All fonts are sans-serif. Discussions are presented in a tree-based, “branching” structure. The top-line topic of Community branches into subtopics of Advice, People, and Complaints, with subtopics of Pay, Technology, Ratings, and Vehicles, among others. Columns bear figures showing participation in numbers of comments, with the Advice sub-topic bearing over 300,000 messages. Discussions are structured in threads, arranged chronologically with the first “post,” or user-generated comment, beginning the discussion and followed by replies from other users.

The site has a wide geographic reach. Geographic subsections include 106 city-based regions in the United States containing 3.3 million messages. Four regions in the United Kingdom and three in Canada have collective content of over 600,000 messages. Six cities in Australia have collectively half a million messages. Subsections for nonwesterm and Global South regions include Guadalarjara, Indonesia, Malaysia, Manila, and Singapore, containing across them hundreds of thousands of messages.

Findings

The vast majority of scam attack accounts were told in the first person, from people narrating their own experiences handling scams attacks, though a small number were from people telling stories of people they personally knew who had been attacked. Multiple different kinds of attack were described on the forums, operating in different ways toward different goals. Some scams were perpetrated by passengers attempting to extort a free or lower-cost ride out of the driver by manipulating the spatial or temporal logic of the app. Sometimes riders placed their location “pin” close to their position yet outside of the higher-priced “surge” bubble, then when drivers approached the “wrong” pin and found no rider, the rider would feign ignorance, claiming that the app had placed the pin in the wrong location, and then provide their “actual” address either via text or voice—channels that the app cannot detect and monetize resulting in a lower fare for the passenger. Other times passengers would request a ride during a high-priced surge bubble, and then when the driver got close, the passenger would cancel the ride and then immediately re-request it, gaming the system to see if the pricing bubble had popped and gone back to lower rates. Other passenger tried to surreptitiously “cancel” the ride after they got into the car, so that the unknowing driver would deliver them to their destination without knowing that the ride—and the fare—had been canceled. Other “passenger” tried to game the app’s tipping function, by telling the driver they’d “accidentally” overpaid the tip, and would need to cancel the ride in order to recoup the cost. Other “passenger” attempted to cancel the ride following arrival at their destination, by claiming in the app that the driver had picked up the wrong person, to get their fare refunded.

Other scams targeted not the driver’s labor but their money. In the majority of these attacks, scammers impersonated the RideHail Company, reaching drivers either by calling them while the ride request was open (and the app created a channel where the driver and the “passenger” could call each other through the app), or by getting the driver’s phone number and calling at a later date and claiming to be RideHail Company. These stories as posted to the forum are usually highly detailed personal accounts of the worker’s interaction with the platform ecosystem and technical stack, including complex interactions between the platform app, banking or cash-transfer apps, and communication channels including voice calling and texting. Most of these stories follow a similar script: a driver gets pinged for a driving job on the app, then before they can pick up the passenger they receive a phone call, through the app itself, from the passenger, claiming to be an employee of the ride-hailing company. On this call, the driver is told that their high rating had made them eligible for a cash bonus, and that the employee needed the driver’s banking app information in order to deposit the cash. In other accounts, drivers were told, again by someone impersonating RideHail either via voicecall or text message, that their account needed to be reviewed and they would need to supply their account-login information to avoid being deactivated.

These scams exploit information asymmetries, where workers had not been provided with, or built up, enough accessible knowledge of the interaction between platform, devices, and various communication channels used by the company to detect discrepancies between actual, and fraudulent, company activities. Many stories conclude with some kind of call-to-action for other forum-goers in the form of a warning, that is, “Be careful everyone” or “watch out,” conveying that the purpose of sharing the narrative is to act as a warning for others, lest they fall prey to the same fate.

In the subsequent discussions motivated by stories of scams, participants’ behaviors display a markedly different pattern than literature might suggest. Literature on online communities, particularly examining public health, details many forms of support that participants both provide and seek out, including information sharing, coordination, and emotional support (Baym, 1999; Da Cunha and Orlikowski, 2008). Respondents to stories of scams oftentimes reply, however, not with support, but with derision. Responses can be grouped into a typology of five types: derision, straightforward information (where drivers simply shared stories of what had happened to them, with no overt implication), credibility checks (where commenters rebuked victims’ legitimacy as drivers), and, rarely, empathy (where commenters express solidarity or sympathy with the victim), or respect (in threads where drivers defeated scam attacks, some responses carried a tone of respect or admiration). Table 1 shows the frequency of these comments, as a percentage of total comments in scam-related threads.

Table 1.

Typology of responses as a percent of all scam-related comments, 2014-2021.

Derision	Information	Credibility check	Empathy	Respect
41.68	40.21	8.52	8.10	1.47

Derision

Derision is the most popular thematic response to stories of drivers “falling” for scams, appearing more than any other theme in my typology of community responses. This was most apparent in response to the type of scams where drivers interacted with the attacker, and faced a choice of whether to “comply” with, or “fall for,” the ploy. Typically, these were the attacks that exploited information asymmetries, where attackers impersonated RideHail Company either by phone call or text message, and asked drivers to turn over information about their RideHail accounts. (This is opposed to those attacks beyond the control of the driver, such as riders trying to claim a refund by falsely reporting in the app, after the ride, that the driver had picked up the wrong person.) In one thread, the original poster¹ describes what happened: they’d gotten text messages from someone claiming to be an employee at RideHail Company. In an ironic twist, the person messaging accused the driver of being the fraudulent one, saying there’d been suspicious activity on their account and telling them they needed to submit their account information so their identity could be verified. The driver dutifully turned over their login information, and their driving funds account (where all of the wages for the rides they’d recently completed were held) was promptly emptied.

Replies followed a punishing theme. Other workers wrote that the victim had been “really dumb,” with exhortations to “Pay attention,” and thematic comments including “I can’t believe how gullible people are,” and “I can’t believe that anyone can be this stupid.” One response contained a YouTube clip from the popular American game show “The Price is Right,” of a sound effect called the “Loser Horn,” blasted on the show whenever a contestant loses. In another thread, in response to a victim who’d lost a week’s worth of earnings, one reply read, “Where is P.T. Barnum when you need him?” in reference to the American circus showman’s apocryphal remark, “a sucker is born every minute.” In another thread, a commenter wrote, “You have to be one dumb box of rocks to fall for that.” In yet another, a commenter took the time to explain, in patronizing language, how straightforward and easy-to-spot the scam operation had been, to refute the victim’s claim that the operation had seemed complex. They added, “I’m not trying to be rude with this, just trying to save you the mental pretzels of figuring out the situation.” In one thread, someone facetiously told a victim of a scam, “Gee maybe it was the Russians,” and when the victim replied, “I’m looking for help, please stop,” yet another commenter replied, “You must be new, snarky smartass is the default around here.” Derision was a theme in all calendar years for which I gathered data. As far back as 2014, someone responded to a scam story with, “People actually fall for this?”

Included in this scorn is the positioning of accountability, not with the scammer, but with the victim: “I think people deserve to get scammed if they fall for it,” “Any driver not aware of this by now deserves to get their money taken,” and “People like you are why scammers exist.” Another victim, after sharing his story, maintained that RideHail Company should do more to educate drivers on the prevalence of this scam, which elicited the reply, “How is that [RideHail Company’s] fault: it’s like you gave your bank card and pin to a stranger and then blamed the bank.” In yet another thread, where the original poster used the word “hacked” to describe the attacker’s breach of his bank account, someone retorted, “nothing was hacked. You just let them in.” In another thread where the victim said they’d contacted RideHail Company for reimbursement, a commenter replied, “your money is all gone! Time to accept it and learn from your awfully stupid mistake.”

Information

The second-most popular theme in community replies to scam stories is straightforward information sharing. In these comments, drivers typically share experiences that had happened to them, or share links to stories of similar events in the news. Many times these stories included how they had managed the situation, either what they’d done to avoid falling victim to the attack or what had happened when they’d failed. In some cases, these stories were told with a condescending tone, in which case they were coded as “derision” rather than straightforward information. In other cases, they were told with phrases expressing solidarity, such as “this isn’t so unusual, [RideHail Company] does call drivers sometimes, they called me,” in which case they were coded as “empathy,” Often “information” comments included tips and advice shared in a straightforward manner, displaying neither empathy nor derision, such as “buy a dashcam,” or “make sure to rate the rider after the ride.”

Credibility check

Revealing are replies that equate the work of driving itself, or capacities associated with driving, with the ability to avoid scams, including “Anyone dumb enough to give away your phone number or social [security number] just shouldn’t be driving,” and “It’s shocking to me that people who can manage to hop through the hoops of becoming a driver would disclose this information to strangers.” This expectation extended to familiarity with the forums themselves, with an indication that anyone who has been spending time on the forums would already be inoculated from attacks: “There have been multiple threads on this scam. . . . Sorry you got taken, but really you should have known better,” “there is a post about these scams a few times a week and everyone here should know better …. Sorry this happened to you, but wise up,” and “this is the oldest trick in the [Ridehail] book of knowledge.” In another thread, a reply subjected the original commenter to what they called a “credibility check” and wrote out a list of ostensible black marks against the poster’s “credibility,” where items included “New user. Just joined today. Did not check the board on the possibility that the scam may have already occurred.” The original commenter is very literally told they aren’t credible because they hadn’t consulted the forum. When yet another victim writes that the purpose of sharing their story on the forum was “to help protect others,” one comment made in reply denied any need for information, writing, “Appreciate your largess but everyone around these parts are hip to this shenanigan.”

Empathy

Once in a great while, in threads featuring scam victims subjected to group derision, someone chimes in to scold the group and question their motivations. In the midst of one such thread, one commenter asked of the group, “Why must you kick him when he’s down?” In another thread, where the original poster was, again, derided, a different commenter wrote, “Thank you for putting yourself out there and trying to help us!!! The replies you’re getting—saying people deserve this crap and we should know better—are ridiculous.” Another asked, as an original poster was bombarded with scornful comments, “Aren’t we here to support each other?” To these replies, commenters retort that these interactions are meant to be educational. One commenter asked, after an original poster described how they’d lost hundreds of dollars to a scam, “Have you learned your lesson?” Another commenter wrote that insults were meant to “make “him” (maybe her?) stronger. If not, the streets will eat them alive.” Another wrote, “We are trying to toughen the [original poster] to survive on their own.” When that last comment received pushback, from the same person who’d questioned the educational value of abuse, yet another commenter retorted, “Quit coddling [them].” While workers may perceive that the purpose of their retorts and abuse are educational, however, the message content is distinct from the purely educational information-sharing which typifies other online communities. The degree of derogatory fervor contained in these discussions suggests an additional reading.

Respect and admiration

A different tone characterizes conversations sparked by stories of forum-goers who had circumvented or defeated attempted scammers. These are far less popular, appearing only a few times. These stories are deftly told in triumphant play-by-play form, with drama built around the moment the worker gains control over the scammer. In a common script for scams, drivers are persuaded to call a phone number they’re told leads to an employee of RideHail Company, who promises them a bonus and that they need simply to produce their account log-in information. Those drivers who have fended off these attacks describe in detail their subsequent manipulations of the scammer, where they often play along, using wheedling messages to coax the scammer into spending precious extra time on what’s already a failed attempt. These posts are met with resounding respect and admiration, with replies of “Well played!” and “Beautiful!” Respondents to these stories shared similar victories from other workers, who’d attempted to trick scammers into losing money or time on their end, for example tricking the scammer into canceling the fake ride, producing a small cancelation fee for the driver. Sometimes the reported purpose of the counter-attack provides no end other than idle fun: “Actually enjoyed screwing with him. Cheap fun in the early morning.” Another wrote, “I would have taken the scam call just for entertainment value.”

Discussion

A multi-layered analysis, drawing on lenses from communities of practice, algorithmic skills, and algorithmic articulation, reveals relationships between organizational structure and knowledge practices. Two broad themes emerge.

Scam susceptibility as illegitimate community membership

Gig-work platforms, and their inhabitants, are no strangers to scams. Workers looking for jobs on carework platforms, for example, already must contend with scams waged on the platform, where “scammers pose as prospective employers with the goal of defrauding careworkers” (Ticona et al., 2018). Vulnerability to bad actors is created by participation in these gig economy platforms. In this research, a slightly different interaction effect emerges between bad actors, platforms, and community-based information. Vulnerability is transitioning out of a sideline effect and into the realm of core capacity. The discussion forum, rather than a simple support system, plays a critical role in the formation of the community and in the selection of signs which can be used to signal community belonging. The scorn heaped on workers who display vulnerability reveals that protecting one’s self from scams is seen as a core component of legitimate membership to the community. Navigating scams becomes not just handy street smarts but a new skill requirement for gig workers. Possessing prowess around scams, and successfully demonstrating this skill, have become ways to signal one’s legitimate membership in this community of practice. This is revealed through observation of the inverse: displaying a lack of such acumen opens one up to skepticism. Gina Neff found similar patterns in how new media workers understood and performed distinct values as part of authentic “ways of being” within the industry (Neff, 2012). Neff found that participants in her interviews of Silicon Valley professionals would critique or denounce the values of their colleagues, often with a focus on colleagues’ perceived “authenticity,” in what Neff framed through Boltanski and Thévenot’s (2006) “worlds” framework as “denunciations” through which workers claim their own legitimacy as members of this world. Gig workers’ comments like “you must be new otherwise you would never fall for that,” are more than insults or even a matter of “toughening up” as claimed. By denouncing the authenticity of others that they perceive “don’t belong,” workers produce expressions of their own belonging to a particular set of values or world. In a process that Neff (2012) calls “defining values in the negative” (p. 71), workers perform belonging in their community of practice, by denigrating the authenticity of others’ membership as measured via their susceptibility to scam attacks.

Forum familiarity as legitimate membership

The second theme among replies to scam stories is the less pervasive, but illuminating, focus on the forum itself. Just as workers deride others for their lack of familiarity with scams, they also denigrate them for their lack of familiarity with the forum. Successfully searching and consulting the forum is another new skill expected of legitimate community members. The very act of posting a warning about a scam reveals to the community that the worker had not consulted the forum, to see whether stories of that scam had already been posted. Authors of these posts were openly rebuked for failing the so-called “credibility check” as we saw in the data.

Another thread displays a different pattern. The original commenter shares their story of an attack, but they also exhibit contrition with their opener, “After doing a search, I guess other members here have fallen for this same scam. . .” This narrative exhibits deference, by mentioning using the forums to search for information about scams before posting. The responses to this post are subtly but markedly different. While insults are posted, they are interspersed with expressions of solidarity largely absent from the other threads, with a higher rate of comments displaying empathy. The first reply is a reassuring “It happens” and “totally believable if you’re not aware or expecting it,” followed by “Damn scammers!”—positioning fault with the scammers. This is opposed to the usual pattern, where accountability is laid at the feet of the victim. In more evidence for this worker’s successful bid for legitimacy, they were also, unusually, praised for posting the story itself: “Posting this is good to help spread awareness on these thieves.” We can see here an interaction effect at play: even though the worker displays susceptibility for a scam, for which they would ordinarily be pilloried as an illegitimate member of the community, by concurrently displaying familiarity with the forum itself, the tone of the community response shifts to empathy.

One commenter attempted to make a centralized resource for scam information, in a post titled “SCAMS, Past and Present,” mimicking the title of a reference manual. The post included a survey, asking forum-goers if they’d ever been a victim of a scam (of those who responded, three, or 12.5% of the total, replied “yes”; six, or 25%, replied “no but someone tried”; and 15, or 62.5%, replied “no,” meaning that nine out of 24, or 37.5% of respondents, had been attacked at least once). The worker wrote that they’d been the victim of multiple attacks, and that they wanted to open an ongoing, centralized dialogue to get advice and support. They concluded their call to action with a request specifically addressing the popularity of the aforementioned patterns of derision:

I would appreciate it if this thread could be kept free of mocking behavior or putting others down if they have been victims. That is the worst thing for this community because it prevents those who have experiences from sharing them and building awareness.

This did not go as requested. Replies lambasted this, claiming that scam-related information was already readily available throughout the forum and that no centralized resource was needed. The commenter was disparaged for his opinion that information about scams even needed to be centralized. “We have dozens if not hundreds of posts every year from new people warning us,” and “Every week there’s a new one who comes here to tell us how dumb they are,” and “These forums have been around for years and now you’re here to save us?” were all comments from different users. These comments create a dichotomy between “new people” or “new workers,” versus “older, more experienced workers,” or “us,” reinforcing the barrier between “new people” and the semi-exclusive community of “us.” In one thread, someone else wrote in reply to a scam story, “This has been discussed here a thousand times. You’re new. The rest of us are not.” This is, as one comment intoned, “oftentimes an ugly business.”

Typically, the three components of work which are required for a group to constitute a community of practice are mutual engagement, joint enterprise, and shared repertoire (Wenger, 1998). The application of this framework to a novel space reveals variance in what might be considered “mutual engagement” and “shared repertoire.” As Wenger (1998) points out, a community of practice “defines itself in the doing,” (p. 4) and so we find that such meanings are redefined by “doing” on and through new sociotechnical contexts. Joint engagement is what these workers have in common. They are all engaged in doing the same tasks under the same algorithmic management system. Mutual engagement, however, and the “mutuality” of that engagement, is exactly what is in question when workers denounce each other’s values. These denunciations are found to be deeply concerned with the third component of a community of practice—a shared repertoire. Typically, a shared repertoire consists of communal resources for getting work done, including vocabularies and references. In the gig-worker forum, we see that familiarity with this repertoire, i.e. the forum itself as a resource for managing scam attacks, becomes a litmus test for legitimate membership. This research reveals that workers themselves, through their “doing,” and through their active, constant renegotiation of mutual engagement through reference to the shared repertoire, in particular, actively create and recreate the community of practice themselves, and that the particular sociotechnical context of this community, with its ecosystem of devices, workers, bad actors, communication channels, and platforms, gives rise to new ways of belonging.

Communities of practice fulfill a number of obligations to the organization (Wenger, 1998). RideHail Company, arguably, would not be sustainable without the extra labor these workers do on these platforms to provide information when systems break down. A feature of communities of practice is that they steward competencies, using identity as a filtering device to allocate attention and redirect what matters most in an evolving set of capabilities. This particular community insists on both (1) skills in navigating scams, and (2) skills in navigating forum-based materials, as new core capacities. It is possible that insisting on forum familiarity as a new skill is a sort of defense mechanism against scams, a self-defensive tonic against the fear of falling victim one’s self. As long as one is familiar with the forums and reads up, then the danger, perhaps, is lessened. Again, the ardency with which workers insist on the viability of this defense may be due to the precarity of their work and its arrangements, a lack of long-term security, and the proximity to scam attacks as a looming and unpredictable threat.

This research reveals that workers do form communities of practice, to discuss their interactions with algorithmic system and use what they’ve learned about the system through reverse engineering or “folk tradecraft” (Whittaker, 2020). However, the specific sociotechnical arrangement of the online forum, distanced digital participation, and the precarity, not just of the work itself necessarily, but of legitimate membership in the work community, drives a form of interaction that is distinct from how other types of communities coordinate and share support and information. In other words, while the technical material of the ride-hailing platform in concert with the online forum bring with them their own politics, this material is flexible, its meaning subject to some degree to the shaping influences of its social evocations (Winner, 1980). While the material of the ride-hailing platform may have been designed to foster extreme digital individuation, workers as seen here do form communities of practice, just differently than we might have anticipated. These communities are what I call “para-organizational” workers. “Para-organizational” workers labor for the same organization, but, as platform workers, their associations with, and membership of, that organization, are ambiguous, vague, and ill-defined—yet these associations and memberships are practiced, values are shared and culturally communicated. It’s simply that the mechanisms, taking place as they do in digital, disparate, distributed ways, are not yet cataloged. My findings reveal that these workers do engage in learning and knowledge-sharing practices reminiscent of strategies used by workers in traditional employment arrangements, but that this form of knowledge sharing follows distinct new patterns. This distinction may be due to the lack of coordination necessary among workers: they do their work individually, so each worker has more latitude in how they apply learning from the group to their own practices. The precarity that undergirds this lack of coordination may contribute to the vociferousness of these discussions. Workers on these forums do not only refer to each other’s acumen at handling their work, but often resort to personal attacks, and conflate ability to handle the work with innate intelligence and common sense. It should be said that online forums are not to be perceived as being generalizable to, or representative of, the wider population of gig workers. In fact they’re widely regarded as hosting the most incendiary, minority opinions, as such forums are often a site for workers to vent frustrations (Johnson et al., 2020). However, the focus of this work is not to paint a picture of the entirety of gig worker communications, but rather, to better understand the specific situated (Suchman, 2007) and local interactions within this novel sociotechnical and sociomaterial context of work. We see here that in this space, work activities and the material of the space reconstitute each other (Suchman, 1996) in ways that do not appear to conform to our definition of a community, yet do, except in unanticipated ways.

Knowledge management under algorithmic competition

The emotional volatility of scam discussions reflects tensions between conceptions of individual-level accountability, versus responsibility of the platform to protect its drivers. Scorn and punishing commentary locating “fault” with the driver are the most popular of all thematic responses. At the same time, while less popular, a small number of drivers located accountability with Ridehail Company, saying that the company should provide more training and make their communications practices more clear to drivers. One asked, “Am I the only one who thinks RideHail should be on the hook for this? They don’t vet the riders!” However, they were immediately rebuked by another driver asking, “Does Target vet you before you go in there? Should Target be responsible if someone steals a credit card and uses it in their store? RideHail cannot prevent you from being stupid, nor are they responsible for your idiocy.” In one thread, after a scamming victim wrote that they would seek out restitution from RideHail Company for the funds they’d lost to the scammer, another driver scolded them harshly: “This [scam] isn’t a new thing around here. Why would [RideHail Company] owe you that money? Grow a backbone.” These frictions reflect the broader tensions in gig work between freedom and flexibility versus vulnerability and precarity. This positioning may be shaped by pressure, from the design of the platform system, on drivers to think of themselves as in competition with each other. Reflecting anxieties about potential over-supply of drivers in the market, one commenter referred to other drivers as “worker bees,” as in, “lots of worker bees will fall for [this scam] and that’s a good thing.” This same commenter had a habit of replying to other victims’ stories with nothing more than emojis of bees, flying in a row.

The prevailing sentiment is that scammers are unavoidable in the gig-work environment. Some commenters have even begun treating attacks as a rite of passage, with one writing, “I have 8k rides and still waiting for my scammer day to come,” and another, “I’ve been driving for years and still haven’t gotten a scammer, starting to feel left out.” Another wrote, “I must be a real jerk, no one has ever called me with a scam,” followed by a sad-faced emoji. Alongside this sentiment is evidence of the belief that it’s not the Ridehail system which makes drivers vulnerable, but rather, that the vulnerable drivers themselves are the ones responsible for attracting dangerous attention. In a similar fashion to how accountability for dealing with scammers is most often located with individual drivers, the mere presence of scammers is attributed to individuals, most often new drivers. One worker commented on the high churn of drivers contributing to the population’s vulnerability: “this is an old scam on these boards, it’s just new people still falling for it. There’s such a high turnover, always new suckers around.” Responsibility, not just for themselves but for the wider driver community, is laid again and again at the feet of drivers. One wrote in response to a victim, “scammers like dumb people like you. Now with [RideHail Company] they’ve found a concentration of dumb people—who will even drive right to them!” One claimed that other drivers were responsible, not just for falling victim to attacks, but for attracting the attackers in the first place (and again employing an “illegitimate drivers” versus “us” dichotomy), they wrote: “drivers like you are why passengers treat us badly.” Another sharply commented, “Stop falling for shit like this. That’s bad for all of us.” Again focusing on competition between drivers, it’s claimed that new drivers are responsible in some part for the presence of scammers.

On gig-work platforms, workers operate in an environment of algorithmic competition. Platforms both operate in and inculcate a brutal economy of attention, where only the “top” few options are usually ever seen by viewers. These “top” options are usually decided algorithmically, according to rankings along a variety of metrics (Stark and Pais, 2021), metrics which are often invisible to the people contributing to the platform (Burrell, 2016; Cameron et al., 2021). Algorithmic competition introduces another layer of labor into platform work: not just the labor assigned by the platform, but the work of being on the platform. In a stark example of worker competition in the ridehail industry, in one discussion a commenter mentioned what they called the “Airport Mafia.” They described how at airports (a valuable territory with a continuous flow of potential passengers), groups of drivers are known to coordinate their login times to drive down the price of rides to push other drivers out of the area. Once other drivers have gone, some in the coordinated group will then log back out, driving the price back up for their friends. On care-work platforms, workers must perform “entrepreneurship” to ensure what Ticona and Mateescu call “compelling individualized visibility,” which includes complying with platform background checks which, when passed, place a differentiating “badge” on careworkers’ profiles (Ticona and Mateescu, 2018), becoming a critical part of competing for the most visible and hire-able positions on the platform. On web-cam sex platforms workers are commodified, their faces displayed in thumbnails and ranked using opaque algorithms which propel a fierce winner-take-all dynamic (Velthuis and Van Doorn, 2020). Researchers conducting interviews with ride-hail workers to analyze their communication practices found that some workers report strategically withholding useful information from other workers on the forums, due specifically to their fear of algorithmic competition (Yao et al., 2021).

While algorithmic competition forces workers to take on more labor just to stay visible, algorithmic management also pushes to drivers to carry more risk—meaning that, not only do workers have to labor harder just to stay viable on the platform, but they carry the risk of potential failure. Algorithmic re-allocation happens across a variety of financial and liability risks. In the ridehail industry, it’s widely known that drivers own the cars and have no health coverage or benefits, and so carry the risks and costs of maintaining equipment, any accidents or damages, as well as lost wages due to missed work in the case of illness. Risk allocation happens in more insidious ways, as well. Moradi and Levy (2020) identify how algorithms help platforms shift risk to workers across several business processes, including staffing and scheduling, defining compensable work, and detecting and predicting loss and fraud. In forum discussions of scam attacks, positioning risk and accountability is carried out not just top-down, from platforms to laborers, but laterally as well, as workers scold each other for their weakness and vulnerability.

Why would workers reinforce this re-allocation of risk to each other? It’s possible that this scolding is done not punitively, but defensively. Viewed through Boltanski and Thévenot’s (2006) “worlds” framework, it’s possible that legitimate belonging may be wrapped up with practices ensuring safety and security, as well as beliefs about the effectiveness of those practices. By denouncing the authenticity and intelligence of others that they perceive “don’t belong,” workers produce expressions, not only of their own belonging, but of the reliability and protections afforded by their “legitimate” practices. Neff’s (2012) “defining values in the negative” (p. 71), may be seen here as “defining safety in the negative.” By denigrating the authenticity of others’ membership when they’re attacked, workers reinforce the viability of the forum itself, and the safety of its members. Remember the workers who insisted that it was due to their familiarity with the forum that they were inoculated from vulnerability: “I saw this scam on the boards so I knew what to do.” By paying attention to the forums and staying abreast of scam stories, workers may be telling themselves that they’re safe. This may be a more tolerable, sustainable position, then determining that nothing they can do can ever ensure their safety or security in a system like ridehail work.

Conclusion

New arrangements of devices and platforms, with attendant actors of passengers and scammers, create novel threat models for drivers. This research reveals some of the mechanisms by which these individuals can develop new kinds of algorithmic skills or competencies and by which groups recognize the legitimacy of those skills. Detecting and avoiding scam attacks as well as searching and consulting forums for useful information are both becoming core capacities for gig workers. Furthermore, demonstrating these skills has become a form of signaling legitimacy in a community of practice. Navigating scams has attained a level of practice intimately associated with this community, and one’s acumen at this task (or lack thereof) becomes fodder for other community members to affirm or deride membership. Workers assert their belonging to the community in the negative by disparaging those who to deftly consult forums and fall for scams, thereby asserting their own expertise (and perhaps their own safety). Generalizing up, this research has contributions for discourse on how organizations produce knowledge (Vertesi, 2020). I argue that the organization of gig work, and especially its structures of algorithmic competition, shape knowledge-sharing practices contingent on performing membership in the community of practice using denunciations, rebuke, and scorn. These practices in turn reinforce the re-allocation of risk through algorithmic means to the worker.

These emergent patterns have stark implications for the safety and security of these community members. Examining how security information is shared among non-expert end users is a key topic of concern for security analysts. Lay users tend to assemble their knowledge “haphazardly,” from a variety of sources, with one study finding that 40% of respondents had gleaned security knowledge from online forums (Redmiles et al., 2016). If workers are discouraged from sharing information about scams, then less information will be available to people looking for guidance, as one commenter in the dataset pointed out. Some scams do follow well-trod patterns, so advice for those will have a longer shelf life. However, it’s inevitable that new forms of attack will emerge as the sociotechnical ecosystem evolves. People who have seen others mocked for sharing their stories may be reticent to share their own, even if those stories are of truly novel attack types not yet recorded on the forum. This may damage the general level of security awareness in the entire community. In the age of mis/disinformation, one cannot help but speculate on the political economy of information sharing within particular communities of practice, the relationship of that political economy to identity, and the broader implications of these relationships for knowledge-management practices. Future research could examine how the design of such platforms, and the tradeoffs they force for workers attempting to navigate complex sociotechnical systems, could be more supportive and cognizant of the articulation and membership labor which underpins the continued working order of such systems.

Footnotes

Acknowledgements

The author would like to thank Winifred Poster, Alex Rosenblat, David Stark, Ingrid Erickson, C.W. Anderson, Michael Schudson, and Josh Whitford, as well as the anonymous reviewers, for their invaluable feedback.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: The author would like to thank the Columbia University Ph.D. Program in Communications for their support, as well as the Columbia Institute for Social and Economic Research and Policy.

ORCID iD

Elizabeth Anne Watkins

Notes

Author biography

Elizabeth Anne Watkins is a social scientist analyzing intelligent systems. She researches AI, work, and governance using a sociotechnical lens, analyzing how people use, navigate around, and are harmed by intelligent systems in the real world, with a focus on safety, risk, and labor. Her training is in human-computer interaction, STS, usable security, and organizational sociology. She works as a Postdoctoral Research Associate at the Center for Information Technology Policy, as well as the Human-Computer Interaction group, both at Princeton University. She is an affiliate with the AI on the Ground group at the Data & Society Research Institute in NYC.

References

Ajunwa

Greene

(2019) Platforms at work: automated hiring platforms and other new intermediaries in the organization of work. In: Vallas

Kovalainen

(eds) Work and Labor in the Digital Age. Bingley: Emerald Publishing, pp. 61–91.

Antonacci

Colladon

Stefanini

, et al. (2017) It is rotating leaders who build the swarm: social network determinants of growth for healthcare virtual communities of practice. Journal of Knowledge Management 21: 1218–1239.

Baym

(1999) Tune In, Log On: Soaps, Fandom, and Online Community. Newbury Park, CA: SAGE.

Bechky

(2006) Talking about machines, thick description, and knowledge work. Organization Studies 27(12): 1757–1768.

Bishop

(2019) Managing visibility on YouTube through algorithmic gossip. New Media & Society 21(11–12): 2589–2606.

Boczkowski

(1999) Mutual shaping of users and technologies in a national virtual community. Journal of Communication 49(2): 86–108.

Boltanski

Thévenot

(2006) On Justification: Economies of Worth, vol. 27. Princeton, NJ: Princeton University Press.

Bowker

Star

(1999) Sorting Things Out: Classification and Its Consequences. Cambridge, MA: MIT Press.

Burrell

(2016) How the machine ‘thinks’: understanding opacity in machine learning algorithms. Big Data & Society 3(1): 622512.

10.

Cameron

(2020) Allies or adversaries? Making meaning of work in the “new” gig employment relationship. In: Guclu

(ed.) Proceedings of the Eightieth Annual Meeting of the Academy of Management. Available at: https://journals.aom.org/doi/10.5465/AMBPP.2020.67

11.

Cameron

Christin

DeVito

, et al. (2021) This seems to work: designing technological systems with the algorithmic imaginations of those who labor. In: Proceedings of the CHI’21, Yokohama, Japan, 8–13 May. New York: ACM.

12.

Chan

(2019) “Becoming an expert in driving for Uber”: Uber driver/bloggers’ performance of expertise and self-presentation on YouTube. New Media & Society 21(9): 2048–2067.

13.

Christin

(2020) Algorithmic ethnography, during and after COVID-19. Communication and the Public 5: 108–111.

14.

Comunian

(2017) Temporary clusters and communities of practice in the creative economy: festivals as temporary knowledge networks. Space and Culture 20(3): 329–343.

15.

Da Cunha

Orlikowski

(2008) Performing catharsis: the use of online discussion forums in organizational change. Information and Organization 18(2): 132–156.

16.

Gillespie

(2018) Custodians of the Internet: Platforms, Content Moderation, and the Hidden Decisions that Shape Social Media. New Haven, CT: Yale University Press.

17.

Graham

Lehdonvirta

Wood

, et al. (2017) The risks and rewards of online gig work at the global margins. Available at: http://pinguet.free.fr/gigwork17.pdf

18.

Gray

Suri

Ali

, et al. (2016) The crowd is a collaborative network. In: Proceedings of the 19th ACM conference on computer-supported cooperative work & social computing (CSCW’16), San Francisco, CA, pp. 134–147. New York: ACM.

19.

Gray

Suri

(2019) Ghost work: How to stop Silicon Valley from building a new global underclass. Eamon Dolan Books.

20.

Green

Viljoen

(2020) Algorithmic realism: expanding the boundaries of algorithmic thought. In: Proceedings of the ACM conference on fairness, accountability, and transparency (FAT*). Available at: https://dl.acm.org/doi/abs/10.1145/3351095.3372840

21.

Greene

Shilton

(2018) Platform privacies: governance, collaboration, and the different meanings of “privacy” in iOS and Android development. New Media & Society 20(4): 1640–1657.

22.

Hargittai

Gruber

Djukaric

, et al. (2020) Black box measures? How to study people’s algorithm skills. Information, Communication & Society 23: 764–775.

23.

Holtz

Kronberger

Wagner

(2012) Analyzing internet forums. Journal of Media Psychology 22: 55–66.

24.

Irani

Silberman

(2013) Turkopticon: interrupting worker invisibility in Amazon Mechanical Turk. In: Proceedings of the SIGCHI conference on human factors in computing systems, Paris, 27 April, pp. 611–620. New York: ACM.

25.

Jackson

(2014) Rethinking repair: breakdown, maintenance and repair in media and technology studies today. In: Gillespie T, Boczkowski P and Foot K (eds) Essays on Communication, Materiality, and Society. Cambridge, MA: MIT Press, pp. 221–239.

26.

Jarrahi

Sutherland

(2019) Algorithmic management and algorithmic competencies: understanding and appropriating algorithms in gig work. In: Proceedings of the International Conference on Information, Washington, DC, 31 March–3 April, pp. 578–589. Cham: Springer.

27.

Johnson

Black

Hayes

(2020) Roles in the discussion: an analysis of social support in an online forum for people with dementia. Proceedings of the ACM on Human-Computer Interaction 4: 127.

28.

Johnston

Land-Kazlauskas

(2018) Organizing on-demand: representation, voice, and collective bargaining in the gig economy. Conditions of work and employment series no. 94. Geneva: ILO.

29.

Kellogg

Valentine

Christin

(2020) Algorithms at work: the new contested terrain of control. Academy of Management Annals 14(1): 366–410.

30.

Klawitter

Hargittai

(2018) “It’s like learning a whole other language”: The role of algorithmic skills in the curation of creative goods. International Journal of Communication 12: 3490–3510.

31.

Kunda

(2009) Engineering Culture: Control and Commitment in a High-tech Corporation. Philadelphia, PA: Temple University Press.

32.

Lave

Wenger

(1991) Situated Learning: Legitimate Peripheral Participation. Cambridge: Cambridge university press.

33.

Lehdonvirta

(2016) Algorithms that divide and unite: delocalisation, identity and collective action in “microwork.” In: Flecker

(ed.) Space, Place and Global Digital Work. London: Palgrave Macmillan, pp. 53–80.

34.

Mateescu

Nguyen

(2019) Algorithmic management in the workplace. Data & Society 8: 1030332.

35.

Matias

(2016) The civic labor of online moderators. In: Internet Politics and Policy Conference. Oxford. Available at: https://blogs.oii.ox.ac.uk/ipp-conference/sites/ipp/files/documents/JNM-The_Civic_Labor_of_Online_Moderators__Internet_Politics_Policy_.pdf

36.

Meadow

(2013) Studying each other: on agency, constraint, and positionality in the field. Journal of Contemporary Ethnography 42(4): 466–481.

37.

Meltzer

Martik

(2017) Journalists as communities of practice: advancing a theoretical framework for understanding journalism. Journal of Communication Inquiry 41(3): 207–226.

38.

Moradi

Levy

(2020) The future of work in the age of AI. In: The Oxford Handbook of Ethics of AI, p. 271. Available at: http://ageofaiwork.theatlantic.com/

39.

Neff

(2012) Venture Labor: Work and the Burden of Risk in Innovative Industries. Cambridge, MA: MIT Press.

40.

Postmes

Spears

Lea

(2000) The formation of group norms in computer-mediated communication. Human Communication Research 26(3): 341–371.

41.

Qadri

(2020) Algorithmized but not atomized? How digital platforms engender new forms of worker solidarity in Jakarta. In: Proceedings of the AAAI/ACM conference on AI, ethics, and society, New York: NY, 7–9 February. New York, NY: ACM.

42.

Redmiles

Kross

Mazurek

(2016) How I learned to be secure: a census-representative survey of security advice sources and behavior. In: Proceedings of the 2016 ACM SIGSAC conference on computer and communications security, pp. 666–677. Available at: http://users.umiacs.umd.edu/~mmazurek/papers/ccs2016-learned-secure.pdf

43.

Rosenblat

(2018) Uberland: How Algorithms Are Rewriting the Rules of Work. Berkeley, CA: University of California Press.

44.

Rosenblat

Stark

(2016) Algorithmic labor and information asymmetries: a case study of Uber’s drivers. International Journal of Communication 10: 27.

45.

Rosenblat

Kneese

Boyd

(2014) Workplace surveillance: open society foundations’ future of work commissioned research papers. Available at: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2536605

46.

Sachs

(2019) The algorithm at work? Explanation and repair in the enactment of similarity in art data. Information, Communication & Society 23: 1689–1705.

47.

Scott

(2004) Reflections on a half-century of organizational sociology. Annual Review of Sociology 30: 1–21.

48.

Seering

Wang

Yoon

, et al. (2019) Moderator engagement and community development in the age of algorithms. New Media & Society 21(7): 1417–1443.

49.

Selbst

Boyd

Friedler

, et al. (2019) Fairness and abstraction in sociotechnical systems. In. Proceedings of the Conference on Fairness, Accountability, and Transparency (Pp: 59–68.

50.

Schmitz Weiss

Domingo

(2010) Innovation processes in online newsrooms as actor-networks and communities of practice. New Media & Society 12(7): 1156–1171.

51.

Soper

(2020) Amazon drivers are hanging smartphones in trees to get more work. Bloomberg. Avilable at: https://www.bloomberg.com/news/articles/2020-09-01/amazon-drivers-are-hanging-smartphones-in-trees-to-get-more-work

52.

Stark

Pais

(2021) Algorithmic management in the platform economy. Sociologica 14(3): 47–72.

53.

Suchman

(1996) Constituting shared workspaces. In: Engeström

Middleton

(eds) Cognition and Communication at Work. Cambridge: Cambridge University Press, pp. 35–60.

54.

Suchman

(2007) Human-machine Reconfigurations: Plans and Situated Actions. Cambridge: Cambridge University Press.

55.

Star

Strauss

(1999) Layers of silence, arenas of voice: the ecology of visible and invisible work. Computer Supported Cooperative Work 8(1–2): 9–30.

56.

Sweeney

(2019) Uber, Lyft drivers manipulate fares at Reagan National causing artificial price surges. WJLA, 16 May. Available at: https://wjla.com/news/local/uber-and-lyft-drivers-fares-at-reagan-national

57.

Tavory

Timmermans

(2014) Abductive Analysis: Theorizing Qualitative Research. Chicago, IL: University of Chicago Press.

58.

Ticona

Mateescu

(2018) Trusted strangers: carework platforms’ cultural entrepreneurship in the on-demand economy. New Media & Society 20(11): 4384–4404.

59.

Ticona

Mateescu

Rosenblat

(2018) Beyond disruption: how tech shapes labor across domestic work and ridehailing. Data & Society. Available at: https://datasociety.net/library/beyond-disruption/

60.

Velthuis

Van Doorn

(2020) Weathering winner-take-all. In: The Performance Complex: Competition and Competitions in Social Life, p. 167. Available at: https://oxford.universitypressscholarship.com/view/10.1093/oso/9780198861669.001.0001/oso-9780198861669-chapter-8

61.

Vertesi

(2020) Shaping Science: Organizations, Decisions, and Culture on NASA’s Teams. Chicago, IL: University of Chicago Press.

62.

Watkins

(2020) Took a pic and got declined, vexed and perplexed: facial recognition in algorithmic management. In: Proceedings of the companion of the 2020 ACM conference on computer supported cooperative work and social computing, Virtual Event, 17–21 October. New York: ACM.

63.

Webb

Weick

(1979) Unobtrusive measures in organizational theory: a reminder. Administrative Science Quarterly 24(4): 650–659.

64.

Webster

(2016) Microworkers of the gig economy: separate and precarious. New Labor Forum 25(3): 56–64.

65.

Wenger

(1998) Communities of practice: learning as a social system. Systems Thinker 9(5). Available at: https://thesystemsthinker.com/communities-of-practice-learning-as-a-social-system/

66.

Whittaker

(2020) The folk tradecraft of workers observing and assessing opaque algorithmic management systems then modifying their behavior in response is both heartbreaking and impressive “people’s research.” Available at: https://twitter.com/mer__edith/status/1300878248059056134 (accessed 1 October 2020).

67.

Winner

(1980) Do artifacts have politics? Daedalus 109: 121–136.

68.

Wood

Lehdonvirta

Graham

(2018) Workers of the Internet unite? Online freelancer organisation among remote gig economy workers in six Asian and African countries. New Technology, Work and Employment 33(2): 95–112.

69.

Yao

Weden

Emerlyn

, et al. (2021) Together But Alone: Atomization and Peer Support among Gig Workers. Proceedings of the ACM on Human-Computer Interaction 5(CSCW2): 1–29.

70.

Yin

Gray

Suri

, et al. (2016) The communication network within the crowd. In: Proceedings of the 25th international world wide web conference (WWW), Montréal, QC, Canada, 11–15 April. New York: ACM.