What if work meant play and working hard meant hardly working at all? This is the promise made by platforms such as Playment and Clickworker. Adorned with images of hip, young people relaxing on sofas idly browsing laptops, their sites suggest that if work still exists in our brave new economy, it is no less fun than playing a videogame or shopping for clothes. A sepia-toned snapshot of the remote work dream, as alluring as it is mendacious, gives lumpen piecework a glint of aspiration and glamour.
As if even the mention of “work” or “workers” might upset this gentle air of bonhomie, the sites refer only to “users,” “taskers” and “players.” Play now equals pay. “Brightly gamified compliance regimes” extend to the work process itself. On-screen rankings, nonmonetary rewards and access to new levels of accreditation — such as Mechanical Turk’s enigmatic “Masters qualification” — are used to gamify tasks in ways that blur the boundaries between work and play.
But when wages become “tokens” or “rewards,” leisure and ease soon look more like theft than fun. Linguistic flourishes such as “reward” hint at the fact that tasks represent a gamble, the wage less a contract and more a wager made by the worker when deciding to accept a task. And when not paid below subsistence levels — 90 percent of tasks on Mechanical Turk pay less than $0.10 per task — they do not pay at all. One of the largest surveys carried out across microwork sites found that 30 percent of workers regularly go unpaid. On Clickworker as much as 15 percent of all tasks go unpaid.
In other words, capital’s online infrastructure runs to a significant degree on unpaid labor.
This continues a longer trajectory, as Melinda Cooper notes: “Under post-Fordist conditions, the wage itself has become something of a speculative proposition… conditional on the achievement of performance indicators” and “unspecified hours of unpaid work readiness.” The worker thus operates increasingly in a quasi-magical economy of gambling and lottery. Microwork represents the grim apex of this trajectory, where the possibility of the next task being paid tempts workers time and again to return for more. Intricate reward schedules and contestable pricing gamify tasks and effectively repackage superfluity and precarity as new, exciting forms of work-cum-leisure.
When wage becomes wager, the status of the worker itself is brought into question. Wage and work are bound together as one under capital. This is not only of ontological interest; it has a vital political valence, for the coherence between worker and wage is the ground from which so much struggle against capital has emerged. Without a wage, one is not quite a worker but a slave, or else surplus — categories that are conceptually and, by extension, politically distinct.
It is for this reason that campaigns like Wages For Housework have demanded a wage be extended “behind the hidden abode” to domestic labor. As Sylvia Federici notes: “The unwaged condition of housework has been the most powerful weapon in reinforcing the common assumption that housework is not work.” That the wage so often goes missing in the case of microwork reveals a similar denial of “worker” status, a reminder that, as with care and domestic work, petty data tasks do not deserve formal recognition.
Microwork sites do of course promise pay, but because they give requesters carte blanche to act as they wish, vast numbers of tasks go unpaid. Structures that appear neutral are in fact often organized around systematic efforts to make the wage optional as opposed to obligatory. Even when paid, tasks are priced at such abysmally low rates that wages shed their reproductive function. On Mechanical Turk, the only platform where wages have been calculated, workers make less than $2 per hour.
Of the many ways these sites facilitate wage theft, the most effective is the form of payment itself. Looking past the shiny pomp of autonomous vehicles and delivery drones, one sees that among Silicon Valley’s most dazzling labor-saving devices is a throwback to 19th-century economy. Piece rates, which let employers pay by the finished product, are more than other forms of payment at eminent risk of wage theft. Let us not forget that it is for this very reason that Marx saw piece rates as “the form of wage most appropriate to the capitalist mode of production.”
A common feature of the Victorian model of capitalism, piece rates all but vanished in the Global North with 20th-century rationalization processes, which allowed tasks to be standardized and wages paid hourly. But they have remained the most common form of payment across the South’s gargantuan informal sector, a false lifeline to those forced to eke out a living at the system’s margin: rickshaw pullers and waste pickers, as well as the sweatshop workers who are subcontracted to work for domestic or global supply chains.
The return of piece rates to the US and Europe is premised on a crude solution to the service sector productivity puzzle. Without an easy automation solution to jobs like food, postal delivery and accountancy, sites like Deliveroo and Upwork have kindly reimagined Victorian capitalism for professional and precariat alike, introducing piecework across a range of once waged or salaried professions to push workers to merciless levels of intensity. This is no more the case than with sites like Mechanical Turk, where rigorous standards of quality are often of less importance than brute speed. AI can already do many tasks listed on microwork sites, but workers maintain the upper hand when it comes to pace. Given that a five-minute HIT (Human Intelligence Task) on Mechanical Turk can be paid as little as 20 cents, workers must work quickly simply to meet their daily needs.
With tasks paid by the piece, workers experience long fallow periods as they search for new jobs, which often means longer hours to make ends meet. Like others made fugitive to the market, more time is spent hunting for jobs than actually completing them. A Mechanical Turk worker from a former Mining town in Appalachia describes a typical day on the platform:
If I work 12–16 hours a day, I’ll make maybe $5/hour. That’s when there is work, but when you’re sitting in between jobs and you consider that time, when you’re just looking for work, then the hourly wage falls dramatically. There are so many of us now, and fewer quality jobs. Sometimes I wake up in the middle of the night just to see if I can grab some good requests. Most HITs are gone if you don’t click right away.
As with other sections of the informal economy, microwork sites stage a “chronic super abundance of labor.” This oversupply and a lack of employment options elsewhere compels workers to spend their nights hunting for tasks paying little more than a few cents. Unlike the spontaneous surpluses swamping cities like Mumbai and Kinshasa, such abundance in the digital realm is strategically planned. Microwork sites are organized to attract greater numbers of workers than there are tasks available, to ramp up productivity and drive down wages, meaning all must accept poor conditions such as long hours and working through the night. Accounts such as the one above are not uncommon. A large study on microwork in Sub-Saharan Africa found Kenyan workers regularly putting in 78-hour weeks.
Under the pressure of accelerated pace and ever longer working hours, accuracy tends to suffer. Yet, with tasks paid so little, mistakes are of little concern to requesters, who distribute vast numbers of similar tasks to multiple workers with the knowledge that a great many “finished products” will be unusable. All that matters to requesters is that enough passably decent tasks are fulfilled in a short time frame.
To make sure pace is maintained, most sites allow requesters to place specific time limits on tasks, which, if broken, result in docked pay. On Leapforce (acquired by Appen in 2017), a typical task was limited to anything between 30 seconds and 15 minutes, often outsourced from the site’s largest client — Google Raterhub. Yet, despite hosting such reputable clients, Leapforce’s platform was clunky and would frequently lag. Oftentimes a task would take more time to load than the time allotted to complete it. One can see the problem here: the requester — in this case, Google — still receives the completed task but can lawfully retract payment due to delayed completion.
Even on sites more sophisticated than Leapforce, workers remain susceptible to the caprices of server volatility, poor connectivity and hostile requesters. On Mechanical Turk, time restrictions are only indicators of how long a task should take, but because the restrictions are defined by requesters, who are eager to cut costs, a task might be marketed as one dollar for 15 minutes but actually take closer to 30 to complete, a reality that a worker might remain unaware of until they are 10 minutes into the task. Once under way, backing out means surrendering payment.
Even tasks completed in the allotted time frame frequently go unpaid. Those deemed “bad quality” by requesters are more often than not simply rejected out of hand. In their study of Amazon Mechanical Turk’s payment system, M. Six Silberman and Lilly Irani found that a photo tagging task might be posted twice for two workers to complete. If the two workers produce the same answers, the requester’s software can pay both workers. If they produce different answers, the software can post the task a third time… In this workflow, the workers in the “majority” are paid; the “dissenter” is assumed to be incorrect and is not paid.
In this specific case, only one out of three workers loses their wages. But this might happen on a far larger scale with hundreds of workers completing the same tasks — say, with 60 workers paid, 30 unpaid. Because the requester can easily claim that any finished task is unsatisfactory — no matter how low their standards may actually be — and subsequently withhold payment, the system readily blurs the boundaries between paid and unpaid labor, commodification and decommodification.
The sweating system, then, returns with a vengeance, with digital sweatshops far better suited to wage theft than their Victorian counterparts. Delphic software architectures turn a quantitative change in the amount of wage theft into a qualitative shift, whereby daylight robbery, now on a systemic scale, pushes the wage to its farcical conclusion as a discretionary reward. In a 19th-century textile workshop, a certain amount of reliability regarding who pays the wages, as well as where and when, meant that workers could at least identify a thief — a necessary step to taking strike or legal action. Even today, rates are usually paid by a single, often familiar employer, a relatively easy target for worker pushback — think here of the 1934 general strike of US textile workers, who fought against diminished piece rates by walking out. On microwork sites, there are no workplaces. “Employers” are multiple over the course of a single day and remain entirely anonymous — hidden behind opaque interfaces — leaving the worker with no idea whom they are working for.
So-called “bad” requesters would not be able to refuse payment if platforms were not organized so as to encourage infringements of the wage contract. To protect their status as intermediaries, platforms operate under the guise of “neutrality” and refuse to enter disputes between workers and requesters. The partisan neutrality of free-market doctrine is here ramped up to absurd levels of prejudice — a neutrality that allows requesters to withdraw wages for unusable tasks, yet still grants requesters full intellectual property rights; that offers requesters total anonymity but makes details about workers publicly known; that allows requesters to come and go as they please but traps workers via payment holding periods.
Curated crowd sites like Appen and Lionsbridge attract long-term clients, but these clients are under no obligation to remain on the platform. This means requesters can easily vanish without paying, while workers are forced to wait until they can cash in their wages, sometimes for as long as 30 days after joining the site or until their payment balance reaches a specific sum.
This frequently means wages disappear before the worker has a chance to withdraw them. One of the more draconian measures taken by microwork sites is to shut down the accounts of those who protest or act in ways deemed to undermine a site’s rules, often without so much as a word from the platform, which can mean all wages stored during the holding period are lost. Expulsion from the platform more often than not is the result of more innocent activity — whether glitches in the software or so-called “errors” on the part of the worker, such as changing address or bank details — often seen as red flags for malfeasance.
The story of capitalism is, in no small part, the story of individuals gradually coming to terms with the disciplinary framework of waged life, even as gainful work itself is eroded. As E. P. Thompson already noted in his 1967 paper “Time, Work-Discipline and Industrial Capitalism”: “in all these ways — by the division of labor; the supervision of labor; fines; bells and clocks; money incentives; preachings and schoolings; the suppression of fairs and sports — new labor habits were formed.”
To these techniques, intended to forge habits conducive to orderly labor, we may now add account closures and public score systems. Effectively allowing “employers” to sack workers without so much as a warning, they return the world of work to a place that resembles Victorian England, only now with the objective pretenses of algorithmic decision-making.
Score systems offer a veneer of objectivity, allowing requesters to numerically measure worker performance. But they are in their own way as partisan as account closures. How they are handled differs from platform to platform, though they tend to entail each worker receiving an aggregate score of ratings previously given by requesters, made public on the site so that other requesters can identify good or potentially bad performers. A rare few ascend the ranks to the heady heights of Master rating and the better paid tasks such ratings bring. For most, however, there is only paralysis or a downward staircase on which workers can only watch as their ratings decline. Metrics thus take on a near tyrannical quality, often representing the difference between finding work again and disenfranchisement.
On the site Microworkers, an approval rating (“temporary success rate”) that drops below 75 percent prevents access to jobs for up to 30 days. As this example shows, if a worker’s ratings suffer at the hands of a particularly strict or hostile requester their reputation plunges and opportunities to find more work shrink.
A whole architecture thus exists to divest the wage of its already tenuous contractual status, granting the likes of Google and Microsoft excessive powers of anonymity and mobility, allowing them to withhold payment for tasks that fail to meet unreasonable time limits, while effectively making workers inert, transparent and, in many respects, powerless to protest.
Alongside these stealth tactics are cruder forms of wage debasement. Many platforms, for instance, pay exclusively in nonmonetary “rewards.” On Picoworkers, wages become Amazon gift cards and cryptocurrency; on Swagbucks, Walmart coupons and Starbucks vouchers; and on InstaGC, the worker chooses between a variety of gift cards for high street brands. In an interview, a founder of Crowdflower casually revealed that the company “paid workers in points for various online reward programs and videogame credits.”
While vouchers and tokens are technically commodities (so avoid the total decommodification of payment) they can hardly be described as money. There are good reasons why the wage comes in monetary form. Money, as the numéraire of the capitalist system, is exchangeable for all other commodities. And while Amazon may claim to be the “Everything Store” — a kind of universal equivalent in corporate form — its gift cards are not as universal as the dollar. A voucher restricts the realm of exchange to the products or services sold by a specific firm and therefore reduces the means by which a worker’s daily needs can be met. One cannot live off Starbucks alone.
Mechanical Turk is perhaps the most interesting in this respect, considering its size and geographical scope. Workers from across the world use the platform, but only a limited number have access to payments via bank transfer. The majority, in the words of Amazon, must “leverage gift cards for … rewards.” This is a highly racialized system: the platform offers most European countries the option of bank transfers, while workers in counties from the Global South — such as Botswana, Qatar and South Africa — only receive gift card points. In these countries, the platform comes to resemble a kind of digital company town, where tasks are completed for tokens only spendable on services and goods provided by Amazon. Even those lucky enough to receive cash for their labor may, ultimately, have little access to it.
Before changes were made to Mechanical Turk’s payment system in 2019, many Indian workers were paid via cheques. These were often lost en route, or were else not cashable — not least because slums and remote villages often have limited postal services and no access to banking facilities.
The wages of microwork are nearly always less than the universal contractual form so often invoked by capital’s fabulists and boosters. They certainly do not add up to the $40,000 figure plucked as if from thin air by the World Bank. This is a significant but unremarked feature of platform capitalism: the workers turning masses of data into the valuable information that sustains the system are waged only in the loosest sense. Microwork sites allow large platforms to hide this reality or at least to make it seem acceptable.
The workforces of Google and Microsoft exist behind a marketing mirage that sustains a sense of microwork as not quite work, the microworker not quite a worker. Far too often, the requesters receive all of the work without having to pay a worker. Only for those actually doing the work does the sentiment ring hollow: getting paid to not do much really means doing a lot to not get paid.