The Tyranny of the Task
The “microwork” that makes AI run is just piecework with a digital makeover
Marco Verch o
r
d
F
a
c
t
o
r
y
Work Without the Worker: Labour in the Age of Platform Capitalismby Phil Jones. Verso, 144 pages.
For many workers, being paid for the tasks they complete, rather than the amount of time they remain present at a desk, sounds like a straightforward way to dramatically improve their daily lives. They aren’t the only ones: human resource types and Harvard Business School experts have long heralded the idea of breaking down work into tasks as productivity-enhancing. It’s become a management book cliché. The promise of a “results only work environment” was taken up, at least for a limited period of time, at Best Buy corporate HQ all the way back in 2005, where office workers were allowed to decide their own hours as long as they got their work done.
Excessively long hours may still be common in finance and tech industries, but even there, workers have been chafing against the expectations of their profession. These upwardly ascendent workers have to contend with a lack of sleep, declining mental health, and managers’ enthusiasm for violating “protected weekends” policies, all in the service of addressing typos on PowerPoint presentations or column widths in Excel spreadsheets. An unscientific survey of finance meme accounts on Instagram uncovers many posts decrying long hours from the gilded towers of downtown Manhattan, while even more generic “hustle porn” accounts suggest it’s working smarter, not harder, that promises the highest reward. Working hours of overtime is no longer the badge of honor it used to be: the modern entrepreneur and the upwardly mobile professional are meant to approach each task with a sense of productive innovation.
For those toiling in less well-remunerated salaried work, organizing the workday around tasks and projects is a way to cut through useless meetings and busywork—the misery of what the late David Graeber called “bullshit jobs.” In public-facing service work, as well as jobs in the digital economy, many have become quietly adept at pretending to work when there’s no work to be done. A crucial selling point of the four-day workweek, which has had trial runs in Iceland, Spain, Scotland, and elsewhere, hinges on the idea that productivity will stay the same even while the working week decreases by 20 percent. In other words, the idea that tasks, not time, are the essence of work.
While a return to what is essentially piecework in the place of waged labor seems to promise a salve against white-collar malaise, it opens up another set of problems. Work Without the Worker, a new book by labor researcher Phil Jones, tries to make sense of the rise of “microwork,” best exemplified by platforms like Amazon’s Mechanical Turk (MTurk) and content moderation carried out for large social media networks. Reducing work to tasks, it turns out, can be a step backward for workers.?
As Jones argues, microwork is the labor that makes AI and algorithmic feeds appear seamless and all powerful. Almost invisible, it remains integral to both the everyday functioning and razzle dazzle of the tech industry. On microwork platforms, workers complete brief, monotonous jobs for whatever employer might request them—typically any operation that a computer would struggle with. Employers can request jobs and expect on-demand efficiency from workers even as there are no real ties between them, and the tasks are so fragmented that employers can justify paying pennies for this labor. Task-workers have very little in terms of bargaining rights or decision-making power; their jobs are presented as a benevolent favor on the part of unstoppable technological hegemons. Razzle dazzle indeed.
While some microwork companies outsource for Big Tech—like Appen, which provides task work for Microsoft and Google Cloud—others are actively brought into a company’s fold, as in the case of Uber’s acquisition of Mighty AI to help transform driver data into autonomous vehicle training sets. Task work also helps to form the bridge between tech companies and the public sector: in Finland, prison workers are offered the chance to perform task work as “a kind of prison reform” through a program orchestrated by an overzealous recruitment company seeking to uplift the country’s fledgling tech startups. Stateside, Alphabet employed microwork for Project Maven, a U.S. Department of Defense project that drew widespread protest at the company, eventually forcing Alphabet to declare that they would no longer work on military contracts. As part of Project Maven, workers were tasked with labeling urban environments so that the military could identify targets in battlefield situations. Just this month it was reported that Google was actively seeking new Pentagon contracts, reneging on the company’s promise.
There is very little distinction between training for the new digital economy and being brutally employed by it.
The conditions of this task-based work exemplify the worst of digital labor: it is freelance, poorly paid, and highly informal. Images, texts, and annotated videos flash before the eyes of microwork laborers, each task thoroughly detached from the product that it builds toward. As Jones writes, this “thins the aperture of knowledge to a tiny sliver of light, divesting workers of the capacity to know what they are doing and to what end.” In addition to being monotonous and alienating, content moderation in particular is psychologically damaging. While the breadth of task work spans far beyond this one sector, it all remains stubbornly dull. Twitter uses Amazon’s MTurk workers to identify trending topics in real time. Warehouse robotics and drone systems, chatbots, speech recognition tools, financial services, and health care all rely on specialist crowdsourcing firms. Self-driving cars, which rank among the most over-hyped AI products, are another technology that requires significant amounts of human click-work. In 2018, 75 percent of the crowdsourced work enlisted for Tesla’s autonomous vehicle training—for example, identifying and labeling objects in an urban environment—came from Venezuela, a country then struggling with a currency crisis.
Globally, twenty million workers take part in microwork, a decent proportion of whom live in the Global South. In this part of the world, microwork has been smoothly integrated into the promise of “development.” Task-based work has been brought into the Dadaab refugee camp in Kenya and the Shatila camp in Lebanon, where refugees help clean data to train artificial intelligence programs. A World Bank–sponsored initiative in Palestine aims to address unemployment in the country with microwork, while the NGO Lifelong (backed by the startup Deepen AI) focuses on signing up Syrian refugees for click-work. The nonprofit Samasource operates as a middleman, funneling Ugandans, Kenyans, and Indians to MTurk. There are obvious connections between microwork and similar development schemes: “microloans” and “micro-entrepreneurship.” Posited as a miraculous fix to devastating unemployment, it only serves to further casualization.
There is very little distinction between training for the new digital economy and being brutally employed by it. Compensation for each task is meager, and many microworkers living in countries outside of the Global North are paid by tokens or Amazon gift cards rather than bank transfer. Any time spent between tasks is completely unremunerated, and those commissioning the tasks are endowed with despotic authority. On Amazon’s MTurk, employers can access a completed task, decide that the work is low quality, and reject it without paying. This is perhaps most shocking in the aggregate: on one site, Clickworker, 15 percent of tasks on the site are completed without compensation.
If task-based work is an employment rights slight-of-hand, its role in the development of artificial intelligence is a Criss Angel MINDFREAK Las Vegas residency. Designed to target automation’s bottlenecks, task-based work provides, as Jeff Bezos has described it, “humans-as-a-service.”? The tasks that comprise microwork are resistant to automation for the same reason most service jobs are resistant to automation—they require, for better or for worse, the human touch. Data is not just data; in order for it to be usable by a machine, it needs to be labored over and made legible.
A particularly satisfying appraisal of artificial intelligence’s limits comes from one of the engineers behind the Roomba, the massively popular robotic vacuum cleaner. For Rodney Brooks, who co-founded Roomba’s parent company and helped to bring dot-com bubble optimism to dusty floors around the world, there are only two scenarios in which today’s AI really lives up to the hype: if “it has a person somewhere in the loop, or the cost of failure, should the system blunder, is very low.” The stakes are low when a Roomba gets stuck in a corner. But more often, there’s a human fulfilling some task, somewhere in the world, that ensures artificial intelligence is doing what it claims to do.
The business model that’s emerged around AI is nothing if not speculative: as much an investment in a supposed future as a piece of technology itself. For Amazon, Jones tells us, what makes their AI venture worthwhile is the fact that they gain access to all of the data that’s involved in the tasks completed via MTurk, operating a kind of data rentierism that enables them to automate future tasks. Another platform, Playment, has a data sharing agreement with certain third parties, including Facebook. And beyond extracting data from the tasks themselves, how workers go about completing those tasks is useful behavioral data in its own right. Ultimately, microworkers are instrumental in the training of AI that seeks to replace them.
Microwork is an ominous warning for progressive ideas about post-work.
Is task-based work a culmination of something, or a temporary stopover? Is this where all work might be heading, or just an awkward teenage phase? Will we someday look back at microwork as an obsolete method for organizing the reserve army of labor?
As Jones notes, these platforms, and their links to development projects, blur the lines between employment and unemployment, pushing the idea of freelance over the edge. If there is anything to be recuperated from task-based work, it may be the breakdown of the idea of an occupation itself. To this end, Jones looks to one of Marx and Engels’s more sentimental quotations: “to hunt in the morning, fish in the afternoon, rear cattle in the evening, criticize after dinner, just as I have a mind, without ever becoming hunter, fisherman, herdsman or critic.”
But even then, microwork is an ominous warning for progressive ideas about post-work: liberty in the workplace for some often means squalor for others. For those working long hours in more conventional jobs, thinking about their work as a series of tasks might signal the arbitrariness and the deep contradictions of the wage-form. But in the dream of abolishing the wage, we should aim to do more than add an entrepreneurial sheen to what is essentially good old-fashioned piecework. While the home-working arrangements adopted during the pandemic have proven that productivity can remain constant (or even increase) when workers are away from the site of production, productivity itself can’t be the basis of an emancipatory politics of work. The tyranny of the task is no different than the tyranny of the clock.