The "ghost work" of AI is done by low-paid workers across developing countries.
The hidden ghost work of data cleaning, image labeling, text processing, and content moderation is being performed by back-end workers across developing economies. Much has been written about the wage differentials in digital labor and the insufficient remuneration being paid to surveilled Global South workers doing the “janitorial" labor that keeps digital platforms healthy and productive. Not only this, it has also come to light that many AI assistants are in fact fully powered by real human agents working in low- and middle-income countries. When massive data sets are used to train AI systems, the individual images and videos involved are commonly tagged and labeled. There are millions of low-paid tech workers doing this back-end work.
Related Issues by Justice Area
Medicare Advantage insurance plans use AI to determine what care it will cover for its 31 million elderly subscribers in the United States.
Journalists at Stat found that companies are specifically using AI systems to deny coverage for care. The massive problem: the algorithms are a black box that can’t be peered into, making it nearly impossible for patients to fight for health care coverage when they don’t know why they were denied it in the first place.
Community Health and Collective Security Disability Justice Economic Justice Issue 2024AI hiring algorithms come complete with dangerous bias.
About 70 percent of companies (and 99 percent of Fortune 500 companies) around the world use AI-powered software to make hiring decisions and track employee productivity. The problem? The tools work by identifying and replicating patterns around who was previously hired, which means they perpetuate the bias embedded in the system, locking marginalized populations out of employment. This is particularly tough for disabled people, people of color, and disabled people of color, who are often subject to employment discrimination.
Disability Justice Economic Justice Issue 2020In June 2020, the north Indian state of Haryana began using a new algorithmic system, called the Family Identity Data Repository or the Parivar Pehchan Patra (PPP) database, to determine eligibility for people requesting financial assistance because they are unable to earn enough to survive for various reasons.
It works by pulling demographic and socioeconomic information for families via several linked government databases, using data on birth, death, marriage, employment, property ownership, income tax payments and more. But it is incorrectly marking residents as dead and blocking them from much-needed funding. Meanwhile, human workers don't have the power to override bad decisions made by the system. A 2023 government study on the system found that "it stopped the pensions of 277,115 elderly citizens and 52,479 widows in a span of three years because they were 'dead.' However, several thousands of these beneficiaries were actually alive and had been wrongfully declared dead either due to incorrect data fed into the PPP database or wrong predictions made by the algorithm.”
Disability Justice Economic Justice Issue 2023The release of large language models (LLM) to the public in 2023 reinvigorated a debate about the use of copyrighted data for model training and questions of appropriate credit and compensation.
Copyright concerns centered on appropriating creators' work and not compensating them for work used in training models, which might generate art that competes with theirs and threatens their livelihoods. Meanwhile, some argue that AI could boost creativity. The debate about the appropriate way to pay and credit creators for derivatives of their work centers the interpretation of fair use. Three artists filed a class-action copyright lawsuit against Stable Diffusion and Midjourney, as did Getty.
Economic Justice