Ghost Work: the Labor that Powers AI

On-demand jobs will continue to grow in the post-pandemic economy — will wages and benefits rise, too?

Apr 11 · 4 min read

By Paula Klein

The inequities faced by on-demand workers compared to full-time employees are familiar although recent legal challenges have met with mixed results. Uber drivers in the U.K. successfully petitioned the courts to count 70,000 of them as full-time employees so they can gain important health benefits and wages. In California, on the other hand, ride-sharing companies — Uber and Lyft — prevailed over workers in similar disputes. Around the globe, fair compensation for online contract workers is in the spotlight.

Behind the scenes is a less-known digital economy worker. In 2019 Siddharth Suri, Senior Principal Researcher at Microsoft Research, wrote a book, Ghost Work, with co-author and anthropologist Mary L. Gray, to shed light on an invisible gig economy workforce spread across the world.

At a March 18 seminar hosted by the MIT IDE, Suri talked about the “human labor that powers modern AI systems” — not only to do web search checks, but also for the important work of content moderation, image recognition, spam detection, and a host of other tasks.

Who are these workers and what is their contribution to the digital economy?

Online Work Transactions

Suri noted that increasingly tasks are “sourced, managed, delivered, and billed through an API” in online work transactions. Workers test algorithms, edit pages, take surveys, offer feedback, and ensure the accuracy and appropriateness of online content. Suri said it’s important to understand the role of these resourceful workers so that they are not further marginalized and also to realize the amount of human input that’s involved in online and AI processes.

Seemingly automated responses to web searches and purchases are actually screened and tested by thousands of human “digital assembly line” workers. These workers are industrious and are on call continuously, according the book.

The authors explain how services delivered by companies like Amazon, Google, Microsoft, and Uber function smoothly thanks to the judgment and experience of this vast workforce that “performs often difficult, high-tech piecework: proofreading, designing engine parts, and much more.”

And in a post-pandemic economy marked by rising unemployment and underemployment among workers at all levels, these jobs will continue to be coveted despite poor pay and lack of benefits.

Suri estimates that in the year 2015, 8 percent of working-age Americans had done some form of work through an API; and that number is growing.

At the seminar Suri explained that on-demand task work results from two-sided platform networks. Increasingly, he said, platforms come between the workers and task requesters. Apps use those platforms. In particular, he studied the rise of MTurk, the widely used task-hiring platform first developed by Amazon in 2005 when it needed temporary labor to help tag online content.

Today, MTurk is ubiquitous and the model of matching workers and tasks over an API is ramping up.

McKinsey estimates that by 2025 on-demand work platforms could add $2.7 trillion or 2% to the total GDP, Suri said. On-demand hiring spans sectors; “it’s a unified model and new way to do work,” he said.

Suri and Gray interviewed hundreds of workers to describe the people behind the work. The authors asked why workers chose the work, how they were compensated, and the nature of the work itself. Findings show a wide range of responses.

Flexible Work, Skill Building

The primary reason workers of all ages and backgrounds do this type of work is to earn money. “But they also do this type of work to control their own schedules, to work from anywhere, and to build their skills,” Suri said.

A 2018 research paper led by Kotaro Hara of Carnegie Mellon University, studied “crowd work” — tasks posted on MTurk. The researchers found the median hourly wages for this type of work to be approximately $2, and only 4% earned more than $7.25 per hour at the time.

Variables such as time spent looking for work, and which jobs are most often posted, also were considered. While the average requester pays more than $11 per hour, according to the report, lower-paying tasks are more plentiful.

For his book Suri monitored employee forums where workers help each other find work, make referrals, and discuss jobs. The forums are important networking tools and differ from unionization — a hot topic among Amazon workers and other gig employees today.

“Unions would help workers get more power in the market where requesters currently have the overwhelming majority of the power. Online forums allow workers to get social support from one another. One isn’t a substitute for the other. They solve different problems,” Suri said.

As recent legal challenges indicate, on-demand work is still proving its value in an ever-more digital economy. Gray and Suri would like to see new opportunities arise that emphasize the value of these workers. “The purpose of the book was to shine a light on these workers so that the world can see how valuable their contributions are,” Suri said. “They made major contributions to the AI revolution. The world should know that.”

MIT Initiative on the Digital Economy

The IDE explores how people and businesses work, interact…

MIT Initiative on the Digital Economy

The IDE explores how people and businesses work, interact, and prosper in an era of profound digital transformation. We are leading the discussion on the digital economy.


Written by


Addressing one of the most critical issues of our time: the impact of digital technology on businesses, the economy, and society.

MIT Initiative on the Digital Economy

The IDE explores how people and businesses work, interact, and prosper in an era of profound digital transformation. We are leading the discussion on the digital economy.