Ghost work

Ghost work is the term Mary L. Gray and Siddharth Suri coined in their 7 May 2019 book for the task-based human labor that AI systems and digital platforms structurally require but systematically conceal — routed through APIs and algorithmic interfaces that render the worker invisible to the end user while presenting the transaction as human–machine interaction. The framing's central claim is that work is not disappearing in the age of AI; it is being hidden. What Silicon Valley markets as automation is, for a structurally irreducible class of tasks, an underpaid distributed workforce performing judgment-intensive work that algorithms cannot reliably perform. Gray's formulation of the mechanism, given to MIT Technology Review in May 2019: ghost work is "any work that could be—at least in part—sourced, scheduled, managed, shipped, and built through an application programming interface" — and it becomes ghost work specifically when companies "market the output as fully automated while relying on hidden human labor." The framing's political-economy charge follows directly: the invisibility is not incidental but designed in, because the product Silicon Valley is selling is "AI" — not "AI plus a large underpaid workforce."

Origin: the May 2019 book

Ghost Work: How to Stop Silicon Valley from Building a New Global Underclass was based on a five-year ethnographic study across the United States and India — the two primary labor markets for English-language ghost work on platforms including Amazon Mechanical Turk (MTurk) and CrowdFlower (later Figure Eight). Gray and Suri documented a workforce including young mothers, early retirees, recent graduates, and people excluded from conventional employment, working piece-rate with earnings frequently below legal minimums, no health benefits, no employer-relationship protections, and at-will termination as a structural feature. The isolation was designed in: workers operated "in isolation, amid great uncertainty, without feedback or benefits," unable to see one another, unable to organize collectively within the platform's architecture, rendered individually interchangeable and collectively invisible to the consumers of their work.

The specific companies the book documented as structurally reliant on ghost work included Amazon, Google, Microsoft, and Uber — companies whose products Gray and Suri showed could function smoothly only through the labor of a vast invisible workforce performing content moderation, data labeling, image tagging, proofreading, and edge-case resolution. The statistic the book placed in general circulation: approximately 8% of Americans have at some point participated in the ghost economy. The book won the 2019 McGannon Center Book Prize at Fordham University and was named a Financial Times 2019 Critic's Pick; Mary L. Gray was named a 2020 MacArthur Fellow for her investigation of "how labor, identity, and human rights are transformed by the digital economy."

Automation's last mile

The book's animating analytical concept is what Gray and Suri called automation's last mile: the irreducible remainder of tasks that AI systems cannot yet fully automate — tasks requiring human judgment, cultural context, and the resolution of edge cases — offloaded onto a distributed workforce through API interfaces that render the labor invisible. The paired claim is structural: every AI automation project produces a last-mile residue, because human judgment cannot currently be removed from any large-scale AI system; the ghost workforce fills that residue, but the platform architecture ensures it fills it invisibly. As Gray framed it in the MIT Technology Review interview: "we've never quite had industries so completely sell contract labor as automation...to say that there's really not a person working here at all." The shift she named was not merely a labor-conditions failure but a structural reorientation: companies were no longer using contingent labor to fill gaps in an otherwise conventional employment model; they were building entire business models on contingent labor and marketing the result as the output of automation — "the dismantlement of employment itself."

The framing locates itself distinctly among the corpus's labor-conditions messages. Bossware names the algorithmic surveillance of workers — the worker's existence is acknowledged and the question is how they are watched. Algorithmic management names the algorithmic direction of workers — the worker's existence is given and the question is who or what gives them their instructions. Ghost work names the prior move: the worker's existence is itself what is denied. The invisibility is not a byproduct of platform design; it is the platform design's most load-bearing feature. Lilly Irani, cited by the Wikipedia ghost-work article, observed that tech companies cultivate a perception of "technological magic" while obscuring the human labor; Gray and Suri's contribution was to give that observation a name, a five-year evidentiary base, and a structural explanation.

The global supply chain dimension

The original Ghost Work study anchored in the US and India. Subsequent research by Brookings Institution analysts extended the global supply chain dimension the book had identified. The Brookings analysis documented the ghost workforce extending across sub-Saharan Africa, South Asia, and Southeast Asia — workers performing tasks for US and European demand sources at $1–$8 per hour, with US employers as the largest single demand source and with "no basic labor protections, such as minimum wages, safety regulations, and clarity regarding taxation regimes." Approximately 10,000 new tasks are published and 7,500 completed per hour on MTurk alone. A separate 2025 analysis of 76 workers across Colombia, Ghana, and Kenya documented 60 independent incidents of psychological harm — anxiety, depression, PTSD, panic attacks, substance dependence — from content-moderation ghost work; one former moderator described reading "up to 700 sexually explicit and violent pieces of text per day."

The geographic supply-demand asymmetry structures the ghost work critique as simultaneously a labor-conditions critique and a global-equity critique. Workers in Kenya earn approximately $2 per hour for data-labeling work identical in content to work performed at $20 per hour by US contractors. They train algorithms for products — self-driving cars, medical diagnostics, large language models — whose benefits flow entirely to Global North markets and companies, while the workers building them have no access to the products, no share of the profits, and no formal legal relationship with the platform companies that profit from their labor. This colonial geometry of the AI supply chain connects the ghost work framing to the modern-slavery framing that African content moderators have advanced through litigation and union organizing, while remaining analytically distinct: ghost work is the supply-chain-level critique of an AI industry mode of production; the modern-slavery framing is the worker-side naming of that supply chain's specific labor conditions from a Kenyan legal-pleading and organizing context.

Propagation into worker organizing

The ghost work critique's path into worker organizing had a precursor a decade before the book. Turkopticon, founded in 2009 by Lilly Irani and Six Silberman at the University of California, Irvine, was built to address the information asymmetry that ghost-work invisibility creates for Amazon Mechanical Turk workers: a reputation and mutual-aid system that let workers rate requesters, flag exploitative tasks, and share information on wage-theft practices. By 2021 Turkopticon had transitioned from an academic tool to a worker-led project, with MTurk worker Sherry Stanley taking the lead-organizer role — the progression Gray and Suri had identified in the book as "the very beginning of a labor movement among independent workers." Ghost Work, published a decade after Turkopticon's founding, named the phenomenon Turkopticon had been responding to before a name existed for it.

The framing's clearest organizational adoption in African labor organizing is the Data Labellers Association (DLA), launched in Nairobi on 13 February 2025 under the theme "Empowering the People Powering AI" by Joan Kinyua — a former Sama data labeller with over five years of annotation work experience spanning self-driving cars, medical diagnostics, and content moderation. Kinyua's launch-day frame — "AI does not exist on its own. Behind every algorithm, every dataset, and every technological advancement, there is invisible labor" — is the ghost work framing's central claim reproduced in an African worker-organizing voice. The DLA drew 339 members in its first week; its advocacy targets — fair compensation, mental health support, legal protections, collective bargaining — operationalize the ghost work critique into the specific conditions its membership faces on platforms including Sama, Majorel, and Teleperformance. Kinyua's interview register consistently carries the invisibility charge: "We are the labour behind AI but remain excluded from its profits."

The African Content Moderators Union, founded in Nairobi on 1 May 2023, organizes a workforce that is ghost labor in the precise sense of the book's definition: content moderators for Meta, TikTok, YouTube, and OpenAI at outsourced Nairobi operations — invisible to platform users, invisible in the AI narratives platforms propagate, and invisible in the legal structure through the outsourcer contractor layer. The ACMU and the DLA together constitute the 2023–2025 African organizing wave that carried the AI supply chain's ghost labor from structural invisibility into a collective-bargaining and litigation register. In the United States, the Alphabet Workers Union–CWA and TechEquity Collaborative published the 2025 report "Ghost Workers in the AI Machine" directly adopting the book's framing to document US-based data workers in opaque contractor networks, with a median wage of $15 per hour, 86% financial hardship, and 66% spending unpaid time waiting for tasks. The report title is the frame; the research operationalizes it.

Why the framing has carried

Three structural features have made the ghost work framing durable across the movement.

First, it names a concealment rather than a harm. Adjacent framings — bossware, algorithmic management, modern slavery — name conditions that workers experience as harmful: surveillance, algorithmic control, bondage. Ghost work names the premise that makes all those conditions possible: the denial that the workers exist at all. Because Silicon Valley's automation narrative is the ideological infrastructure on which the ghost labor economy rests — the AI product cannot be "AI" if it acknowledges human input — the ghost work framing attacks that infrastructure at its foundation rather than one of its downstream effects. The politics of the framing are partly epistemological: to name ghost work is to refuse the automation narrative, to insist on a fact (workers are here, doing this work) against a constructed fiction (this is automated).

Second, the framing operates at the supply-chain level rather than the workplace level. Ghost work is not about a specific employer's practices or a specific sector's conditions; it is about the architecture of a mode of production in which human labor is systematically routed through interfaces designed to make it invisible. This supply-chain frame is portable across continents, platforms, and sectors: the ghost work claim applies identically to an MTurk worker in Hyderabad labeling product images, a data annotator in Nairobi flagging violent content for OpenAI's training sets, and a Filipino BPO worker scoring AI-generated outputs for a US tech company. Different supply chains, one structure.

Third, the framing arrived with a specific program. Ghost Work included ten policy and technical recommendations — portable benefits, data portability, minimum-wage and overtime protections applicable to task-based work, and platform-design requirements giving workers visibility and collective-representation capacity. The framing is not merely diagnostic; it came paired with a reformist program whose specific demands the organizations that have adopted it — the DLA, the ACMU, CWA/AWU — can locate their own advocacy within. The question "what does this framing actually buy you?" has an answer in the book's recommendations, and each organizing vehicle that has adopted the framing connects its demands to one or more of them.