Skip to main content
CampaignsEqualityHousingEnvironmentGeneral ElectionSupport Our WorkFixing BritainMigrationEducationRaceCultureWorkGlobal

The ghosts behind AI

Far from succeeding humans, machine learning desperately needs humans to succeed. But the emerging market for educating robots is a dark one. 

February 03 2023, 10.51am

What is climate change?

Who is history’s best boxer?

How can I write a great essay?

These are all questions you can ask ChatGPT, and receive an elaborate answer that goes beyond what Google has ever offered. The chatbot, which was launched by OpenAI in November last year, hit a staggering million-plus users signing up in the first five days after its release. Its popularity provoked stark questions about the future of AI, with many asking whether AI will eventually be so intelligent that it replaces us. The simple answer to this is no: AI won’t ever replace human labour, because it depends on it.

Digital labour by human beings is needed to train AI systems. Most profit-maximising algorithms require deep learning, a technique that depends on scores of labelled examples to work. From Tesla’s self-driving cars to ChatGPT’s essay answers, these smart tech have to learn their skills somehow. So, who makes intelligence intelligent? Meet ‘microworkers’.

While most of us might have never met a microworker, an estimated 20 million people do this type of work, with the majority located in the Global South. Anthropologist Mary Gray and computational social scientist Siddarth Suri term this invisible and exploited labour “ghost work”, due to these workers operating behind closed doors. Microworkers complete short data tasks which are hosted in digital platforms, who sit between the contractor and the microworker. Contractors tend to be the tech companies we know and love - Facebook, Google, Amazon - and the microworkers are often invisibilised workers who depend on these tiny tasks to earn a living.

For the most part, our understanding of technological innovation and artificial intelligence is largely perceived through an individualistic framework, with CEOs positioned as ‘brilliant minds’ who spend their time thinking up new inventions that will change our world. In reality, microworkers are the invisible workforce powering technological development, what Phil Jones calls the “hidden abode of automation.” The scale of the microworking industry is huge; its value is projected to reach $13.7 billion by 2030. Despite this, those who do this work often earn as little as $1.46 an hour after tax, with some tasks on Amazon’s Mechanical Turk platform paying just $0.01.

From completing surveys, to clicking on images flashing on screens, to short translation tasks, to coding data, microwork, also known as crowdwork, is a growing form of employment in the digital age. These monotonous tasks require workers to hyperfocus and meet arbitrary targets, creating inevitably stressful working conditions. Mary Gray, co-author of Ghostwork: How to Stop Silicon Valley from Building a New Underclass, believes the term microwork is insulting in itself because “there’s nothing small about having to focus so intently on an information task.”

The performance of these invisible workers is surveilled by the platforms they work for, with workers penalised with poor ratings if they produce bad results, making it harder for them to find work. Not only this, but the psychological stress of being monitored and dependent on fast task completion and keeping yourself constantly available to spot new tasks when they become available makes this work increasingly stressful, with workers being required to be constantly seeking more work. The fragmented and hidden nature of microwork makes collective action harder; workers experience high rates of alienation due to working in solitude.

Platforms like Amazon’s Mechanical Turk, Clickworker, Appen and Sama, pay workers piece wages - meaning they get paid per completion - to complete these tedious tasks. In true dystopian form, microworkers compete across the world to access these tasks, bidding to complete them for cheaper prices. The median hourly rate for an Amazon Mechanical Turk worker is $1.38, but tasks often pay as little as 1 cent. Last year, a report by thinktank Autonomy found that 95% of microworkers in the UK earn below minimum wage, with almost two in three earning less than £4 an hour.

Microwork or marginalised work?

But this work isn’t simply poorly paid and mind-numbingly repetitive. The tasks that microworkers rely on to earn a living can also be traumatic. A recent investigation by TIME, found that Open AI, the creator of ChatGPT, outsourced Kenyan workers paid less than $2 an hour in an attempt to make it less toxic. Three employees reported they were expected to read between 150 to 200 passages of text in each 9 hour shift. Workers were to label text, including graphic descriptions of child sexual abuse, bestiality, murder, suicide, torture and incest, in order to train the app to be able to detect these in practice, and prevent it from using such examples with a user. 

The microworking company that made this exploitation possible was Sama, a San-Francisco based firm whose clients include Walmart, Google, and Microsoft. This is not the first scandal involving the company - last year, a South African content moderator sued Facebook for a toxic work environment, under his employment by Sama in Nairobi.

Now heralded as the Sub-Saharan hub for microwork, Nairobi has become a perfect fit for profitable tech billionaires looking to maximise their profits, where they can access and exploit English speakers who are paid pitiful wages. The outsourcing of these types of technological work to countries across the Global South, where labour regulations are often weaker, is intentional, and is done to boost profits and keep wages low. Adrienne Williams, Research Fellow at the Distributed Artificial Intelligence Research Institute believes the experiences of microworkers follows a historical trend of exploitation and marginalisation. “That is pretty much a tale as old as time, which is why it is being referred to as digital colonialism. Find people with the least resources, pay them as little as you can get away with and dispose of them as necessary.” says Adrienne. The reproduction of hierarchies of devalued labour across the Global South and wealthy consumers’ reveals the way that racial capitalism continues to operate within the global economy today. “Tech companies are very aware of which workers reside in countries nobody cares about. Governments do not have open door policies that lend an ear or instant protections to vulnerable populations when multinational corporations are stealing their wages, or violating health and safety policies, or wage and hour laws. They make poor people fight for every inch or die trying.” 

Contrary to John Maynard Keynes’ prediction that technological development would reduce our working hours, technological development has in fact opened up new opportunities for the continued extraction of our time and exploitation of our labour. 

Microwork comprises a large part of the explosion of the gig economy in an increasingly precarious and neoliberal global economy, whereby workers are desperate to make ends meet and where Big Tech are capitalising on their suffering. For many of these workers, they’ve long been excluded from standard wage employment - from being incarcerated, or chronically ill, to undocumented migrants or those living in countries undergoing economic crisis - microworkers are a growing group of marginalised workers. In Lebanon’s Shatila Refugee Camp, Syrian refugees conduct freelance programming, labelling footage of urban areas, from within their makeshift tents.

Venezuela is one source of microworkers, a country which has become a hub of underpaid tech labour. In 2018, more than 75% of Tesla’s data needed for its driverless vehicles was labelled by Venezuelans. Following the country’s economic collapse, employment suddenly rose, leaving workers in need of new forms of employment. Suddenly jobless, Venezuelan workers turned to microwork platforms to label urban environments, often for less than a dollar an hour.

Microworkers are the epitome of the entrenchment of inequality in a globalised neoliberal economy run by free market fundamentalism. As Phil Jones writes in Work Without the Worker, “in the hour it takes Amazon CEO Jeff Bezos to make $13 million, a refugee earns mere cents teaching his algorithms to spot a car.”This is no coincidence. Research by the Oxford Internet Institute last year found that these ghost workers often face low pay, risky and exploitative conditions, and little to no bargaining power. One of the reasons for this is their classification as independent contractors and self-employed, excluding them from both the protection of the tech clients dependent on their work and international labour regulations. This often results in microworkers not being paid at all, like microworker Rafael Perez, who earned $180 in 15 days, but never received the money from the platform, and has no one to raise this with. Speaking out against unfair treatment is often prevented, with many microworking platforms using non-disclosure agreements to silence workers, and possessing the ability to shut down worker accounts with no explanation, depriving them of vital income.

Despite criticisms around working conditions and platforms such as Mechanical Turk and ChatGPT growing at an immense rate, this type of labour has even been implemented into non-profit initiatives under the guise of enabling economic development, such as Save the Children’s Microworks initiative, which uses cryptocurrency to pay unemployed young people for completing short tasks, and the World Bank’s 2work project, which is powered by Palestinians. The settler-colonial occupation of Palestine makes it a hotspot for microwork, where marginalised workers are in need of income, and come to depend on a neocolonial network of tasks being outsourced to them.

The cognitive dissonance that permeates modern understandings of artificial intelligence as being attributed to the imaginative talents of CEOs or shiny machines is fuelling the continued exploitation of microworkers. Utopian narratives around AI portray a pretty picture of innovation that obscures the dark reality of its sudden expansion, and fuel the erasure of the workers that are making our digital world possible.

The digital age is creating new opportunities for exploitation and erasure. The Big Tech empires that surround us put microworkers on the lowest rung, devaluing their labour and raising their work. Popular representations of AI as some autonomous and omnipotent machine are inaccurate - artificial intelligence gets its intelligence from the labour of armies of exploited workers across the globe. In the continued race to the bottom, it’s vital we uplift workers, and call for the provision of decent wages and working conditions.

Mary Gray believes that reframing our understanding of what work is, and how it takes place. “We must reimagine a social safety net detached from the hours and places that we work. As platform economies regularly upend what counts as “skilled” labor. Ultimately, we need to build systems so that the worker, rather than the API, controls her employment opportunities. We should penalize employers for misclassifying, delaying, or failing to pay workers, which remains one of the greatest injustices against freelance workers today.”