Google Hired Gig Economy Workers to Improve Artificial Intelligence in Controversial Drone-Targeting Project

Millions of gig economy workers around the world now earn a living on so-called crowd worker websites — work that falls under the umbrella of crowdsourcing, or dividing up tasks into minuscule portions to spread over a large number of people. The sites pay as little as $1 an hour for individuals to perform short, repetitive tasks, such as identifying images seen in pictures and churning out product reviews.

Some of these crowd workers were unknowingly helping to build out the Pentagon’s battlefield drone capability.

The work was done as part of a Defense Department initiative called Project Maven. Last year, The Intercept reported that the Pentagon had quietly tapped Google as part of the project to develop an artificial intelligence program to help Air Force analysts swiftly sort the thousands of hours of drone video and choose targets on the battlefield.

Outsourced crowd workers were tasked with providing the initial image data labeling — correctly identifying parts of an image — that allowed Google’s artificial intelligence program to tell buildings, images, trees, and other objects apart.

The artificial intelligence program, however, must learn to be able to distinguish between objects on the videos — and in order for the program to learn, someone must teach it. Enter the outsourced crowd workers, who were tasked with providing the initial image data labeling — correctly identifying parts of an image — that allowed Google’s artificial intelligence program to tell buildings, images, trees, and other objects apart. The role of the crowd workers has garnered little attention.

Emails obtained by The Intercept show that shortly after Google inked the deal with the military, the tech giant began working to label a set of satellite images captured by a technology known as wide-area motion imagery. In October 2017, Google sent a company called CrowdFlower — which subsequently changed its name to Figure Eight — the raw images with instructions on data labeling. The data labeling project started showing results quickly over the following month, as engineers developed better guidelines for the crowd workers to teach the artificial intelligence to identify the objects.

Project Maven, with the help of the crowd workers, was designed to allow Pentagon officials to engage in “near-real time analysis” and to “click on a building and see everything associated with it,” including people and vehicles, according to leaked documents obtained by The Intercept.

Asked if crowd workers have continued to contribute to Project Maven, a spokesperson for Google referred our questions to Figure Eight, which did not respond to The Intercept’s inquiry.

Since 2007, Figure Eight has hosted one of the largest digital platforms that allows individuals to sign up to perform micro-tasks, such as data annotation. The “human-in-the-loop” service is marketed as a cost-effective way for companies to fine-tune large data sets to make algorithms more accurate. Other firms in the industry include Amazon’s Mechanical Turk, Upwork, and Clickworker.

Will Pleskow, an account executive at Figure Eight, confirmed his company’s role on the Project Maven initiative during a September 2018 interview with The Intercept at the AI Summit, a trade show for machine-learning companies. Pleskow said the workers performing the data labeling, known as “contributors,” did not know that they were working for Google or for the military, which is not an unusual arrangement.

“Our customers have the option of showing who they are. Most of the time, it’s kept anonymous,” said Pleskow. He added that his firm provides similar image labeling tasks to improve the quality of technology for autonomous vehicles, such as driverless cars.

Several Figure Eight workers told The Intercept that it is not out of the ordinary for workers to be left in the dark about how their assembly-line style of data entry is used.

“Contributors to the Figure Eight platform are not given who the data will benefit,” said one former crowd worker, who used an online username when The Intercept reached out over social media. “Usually, they are given a reason for why they are doing a task, like, ‘Draw boxes around a certain product to help machines recognize it,’ but they are not given the company that receives the data.”

Another former crowd worker said that Figure Eight only provided the identity of the end-user client in a handful of scenarios, typically when the platform was used to conduct academic studies.

“Workers absolutely should have the right to know what they are working on, and especially when moral or politically controversial activities are involved.”

The rise of the gig economy has presented a myriad of challenges for organized labor. Most gig economy firms, including virtually all crowd-worker platforms, classify their workers as contractors, which means that they do not qualify for benefits, minimum wage, or overtime.

The distributed network allows for a global workforce. Figure Eight has a large user base in countries such as Venezuela, Indonesia, and Russia, as well as the United States. The far-flung employee base and individualized tasks on an opaque platform provide few opportunities for questioning corporate decisions.

Pleskow said in the September 2018 interview that he was unaware of any discontent or ethical concerns from Figure Eight workers after the revelation of the Project Maven initiative last year.

At Google, however, an open rebellion broke out over the project. Several employees quit Google in protest, while others openly challenged the Silicon Valley giant’s leadership, claiming that the company had abandoned its “Don’t Be Evil” ethos. Employees demanded that the company swear off future “warfare technology” projects. Executives were later caught misleading workers, erroneously stating that the contract was merely worth $9 million, while internal documents revealed that Google expected Project Maven to ramp up to a $250 million contract.

Following the protest, Google executives announced that they would not renew work on Project Maven after the initial phase of the contract ran its course — though they did not promise to eschew military work in the future.

Other tech giants are reportedly interested in engaging the military as it continues to deploy artificial intelligence technology. Much larger machine-learning projects may require vastly new engagement from gig economy workers, who may unknowingly engage in the work.

“Workers absolutely should have the right to know what they are working on, and especially when moral or politically controversial activities are involved,” said Juliet Schor, a sociology professor at Boston College, in an email to The Intercept. “It’s a basic dimension of democracy, which should not stop at either the factory or the platform ‘door.’ For too long, the country has tolerated erosion of basic civil rights in the workplace, as corporations assume ever-more control over their workforces. It’s time to win them back.”