Ghost Work: How to Stop Silicon Valley from Building a New Global Underclass

By Mary L. Gray, Siddharth Suri. 2019

956 citations recorded by the source corpus.

Subfields: AI Ethics, Fairness & Society, Social Sciences & Humanities

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In AI Ethics, Fairness & Society

In Social Sciences & Humanities

Citations in the corpus

Ethical and social risks of harm from Language Models (2021, background): “This includes concerns about the working conditions of data annotators or “ghost workers” (Gray and Suri, 2019), the ethics of supply chains of hardware on which LM computations are run (Crawford, 2021), or environmental costs of training such models (Bender et al., 2021; Patterson et al., 2021;…”
Evaluating the Social Impact of Generative AI Systems in Systems and Society (2023, background): “This phenomenon, the "paradox of automation’s last mile", is a self-propelling cycle in which every solution to automation problems creates new problems to be automated, and hence new demands for ghost workers [93].”
CrowdWorkSheets: Accounting for Individual and Collective Identities Underlying Crowdsourced Dataset Annotation (2022, background): “For example, as [22] points out, the remote nature of crowdwork differentially attracts workers along gender lines, such as mothers who do crowdwork because it allows for an easier balance of childcare in comparison to other work.”
Do Datasets Have Politics? Disciplinary Values in Computer Vision Dataset Development (2021, neutral): “…datasets, including descriptions of data annotators [49, 61], data availability for research purposes [12, 83], properties of data subjects and categories [98], and descriptions of decisions more generally which typically are obviated or obscured in the data collection process [10, 43, 44].”
Toward Trustworthy AI Development: Mechanisms for Supporting Verifiable Claims (2020, background): “AI development has raised concerns about amplification of bias [6], loss of privacy [7], digital addictions [8], social harms associated with facial recognition and criminal risk assessment [9], disinformation [10], and harmful changes to the quality [11] and availability of gainful employment [12].”
Large image datasets: A pyrrhic win for computer vision? (2020, background): “Recent works such as [37] has explored the intentionally hidden labour, which they have termed as Ghost Work , behind such tasks.”
Bringing the People Back In: Contesting Benchmark Machine Learning Datasets (2020, background): “…systems has focused on the importance of analyzing interrogating unspoken work practices of technical experts (Passi & Barocas, 2019; Seaver, 2019), subject matter experts in acts of data repair (Sachs, 2019), and crowd laborers (Irani & Silberman, 2013; Salehi et al., 2015; Gray & Suri, 2019).”
Decolonial AI: Decolonial Theory as Sociotechnical Foresight in Artificial Intelligence (2020, methodology): “The people who do this labelling for a living, the so-called “ghost workers” (Gray and Suri 2019), do this work in remote settings, distributed across the world using online annotation platforms or within dedicated annotation companies.”
On the genealogy of machine learning datasets: A critical history of ImageNet (2021, background): “, 2020), and data labor often goes undocumented (Irani and Silberman, 2013; Gray and Suri, 2019; Miceli et al., 2020).”
Taxonomy of Risks posed by Language Models (2022, background): “For work on upstream risks such as social concerns associated with working conditions of data annotators, see Gray and Suri [65]; on the ethics of supply chains for hardware to run LM computations, see Crawford [39]; and on environmental costs of training, see Bender et al.”
Power to the People? Opportunities and Challenges for Participatory AI (2022, background): “general data enrichment and moderation paradigm, but the abuses and concentration of these forms of “Ghost Work" in low income countries and peoples have also been extensively documented in recent years [40, 46, 65, 74].”
Algorithmic management in a work context (2021, background): “Insights from the use of algorithmic management in the gig economy reflect the same findings where workers are considered replaceable “human computers” and may engage in (ghost) work that is repetitive, does not result in any new learning, and only intends to train AI systems (Gray and Suri, 2019).”
What Do Platforms Do? Understanding the Gig Economy (2020, background): “…encompasses a large, highly varied group of workers living in both the Global North and South, tasks are priced under extremely competitive conditions, and the ability to earn a living wage in wealthy countries through microtasking is limited (Berg 2016, Berg & de Stefano 2018, Gray & Suri 2019).”
Harms from Increasingly Agentic Algorithmic Systems (2023, background): “…sociotechnical blindness [Johnson and Verdicchio 2017], wherein people “believe AI systems got to be the way they are without human intervention”, and obscuring of the (often exploitative) human labour which enables AI systems to exist [Gray and Suri 2019; Pasquinelli and Joler 2021; Perrigo 2023].”
ORES (2019, background): “ober 2020. ORES 148:21 more decoupled than in most content moderation models for commercial social media platforms, where corporate managers hire contractors to label data using detailed instructions [43,87]. While ORES’s patterns around the tasks of model training, development, and deployment are s”
Decolonizing Content Moderation: Does Uniform Global Community Standard Resemble Utopian Equality or Western Power Hegemony? (2023, background): “These fndings point to algorithmic exploitation enabled by digital colonialism where Facebook and other social media platforms based in the West appropriate the cheap labor of the moderators based in the Global South to train their underlying algorithms while doing little to empower the moderators d”
Hidden transcripts of the gig economy: labour agency and the new art of resistance among African gig workers (2019, background): “, 2018), and as a way to reduce overhead costs, perform work and develop social bonds and support to make work manageable (Gray and Suri, 2019).”
Between Subjectivity and Imposition (2020, background): “ get inscribed in datasets and systems [30]. We propose four (interconnected) reasons for integrating such a perspective into research: First, this perspective could contribute to making work visible [30, 44, 75]. Especially in the case of machine learning systems where the enthusiasm of technologis”