When people think of Artificial Intelligence, the image that often springs to mind is that of sentient machines or shiny metallic robots, a depiction heavily influenced by popular culture. This narrative, along with language around "magical" or "mythical" AI, tends to overshadow actual pressing ethical issues associated with AI development and usage. This post explores the exploitation of human labour in AI development, including low-paid workers used for categorising and labelling data, and the impact of the AI infrastructure on human workers.

In the ongoing arms race towards creating autonomous AI systems, multinational technology corporations are relying on a lot of "ghost work". This term, coined by anthropologist Mary L. Gray and computational social scientist Siddharth Suri, refers to labour carried out by a "global underclass" of precarious workers. Occupying roles such as content moderators, data labellers, and delivery drivers, these workers often come from economically disadvantaged backgrounds and perform critical tasks for the tech industry at low wages and under suboptimal working conditions.

The way AI functions currently leans heavily on methodologies like statistical machine learning and deep learning through artificial neural networks. Such methods necessitate vast quantities of data. To obtain this data economically, platforms like Amazon's Mechanical Turk have emerged, enabling "crowd work" which involves breaking down large tasks into smaller units that can be handled by numerous workers.

The emergence of such platforms and data-labelling companies, however, has resulted in workers being treated like parts in a machine, rather than individuals with rights and needs. These workers are often subjected to constant surveillance and repetitive tasks and face punitive measures for any deviation from assigned tasks. The mental and physical toll can be considerable, especially for content moderators who are continuously exposed to traumatic content without adequate support systems in place.

Case Study: OpenAI's Data Labelling

OpenAI, the company responsible for the enormously popular AI Large Language Model ChatGPT, has made great strides in the past 12 months with many forms of Artificial Intelligence. However, some of these achievements have raised significant ethical concerns regarding the exploitation of human labor and the handling of harmful content.

GPT-3 was designed to demonstrate exceptional linguistic abilities, stringing together sentences in a strikingly human-like manner. It was trained on hundreds of billions of words scraped from the internet, a vast corpus of human language. This method endowed GPT-3 with impressive language-processing skills but also became its largest setback, as it incorporated the internet's toxicity and bias into its output.

To tackle these challenges, OpenAI aimed to construct an AI-powered safety mechanism. The premise was straightforward: feed an AI with labelled examples of violence, hate speech, and abuse, and this tool could learn to identify and eliminate these forms of toxicity.

In November 2021, OpenAI began the process of creating this safety system. They sent tens of thousands of snippets of text to an outsourcing firm in Kenya, Sama. The text was pulled from various internet sources, including extremely harmful content describing graphic situations of abuse, murder, and self-harm. Sama, a San Francisco-based company, employs workers in Kenya, Uganda, and India to label data for Silicon Valley clients like Google, Meta, and Microsoft. While it brands itself as an "ethical AI" company, there are concerning elements surrounding its operations.

Sama's data labellers, who were contracted to work on behalf of OpenAI, earned a take-home wage of approximately $1.32 to $2 per hour depending on seniority and performance. This rate was for work that involved labouring over harmful, potentially traumatising content.

Anatomy of an AI System

"Anatomy of an AI System" is a large-scale map and long-form essay produced by Kate Crawford and Vladan Joler. The project aims to illustrate the complex network of resources, labour, and data required to create and operate a single AI system, in this case, the Amazon Echo.

The essay and map unravel the lifespan of an AI system, from resource extraction to its eventual disposal. Key themes include:

  1. Material resources: The extraction of Earth's minerals used in the manufacturing of devices, as well as the environmental implications of these processes.
  2. Labour: The often invisible human labour, such as precarious "ghost work", involved in the creation, maintenance, and disposal of AI systems.
  3. Data: The extensive amount of personal data that AI devices collect from users, which further trains and enhances the AI.

The main conclusion is to emphasise that AI systems are not intangible or magical but are deeply rooted in Earth's geology and underwritten by human labor and ingenuity.

Teaching AI Ethics

These "advanced level" AI Ethics activities use Project Zero thinking routines to explore the problem of human labour:

  1. Circle of Viewpoints Thinking Routine: Students assume different roles or perspectives in the context of AI ethics and labor exploitation. For example, they might adopt the viewpoint of a tech CEO, a data labeller, an AI ethicist, a delivery driver, a consumer of AI products, and a policy maker. Each student, from their character's viewpoint, discusses their perspective on the issue, its impacts, and possible solutions.
  2. Zoom In Routine for Unseen Labour: Present students with images that indirectly represent the impact of AI on human labour, such as a Google search results page, an Amazon product page, or a snapshot of an AI chatbot interaction. Introduce the concept of unseen labour in AI and explain how each image involves the contribution of countless unseen workers.
  3. Compass Points Routine on Proposed AI Regulations: Propose a hypothetical regulation or policy that aims to improve the working conditions of people in the AI industry. Use the Compass Points routine: E (Excited): positive outcomes; W (Worried): concerns and unintended consequences; N (Need to Know): additional information needed; S (Stance): individual viewpoints.