Flock uses overseas gig workers to build its surveillance AI

Herd, automatic The license plate reader and AI camera company uses foreign workers from Upwork to train its machine learning algorithms, with training materials that tell workers how to review and categorize footage, including images of people and vehicles, in the United States, according to material reviewed by 404 Media that was accidentally leaked by the company.

The findings raise questions about who exactly has access to the footage collected by Flock’s surveillance cameras and where the people reviewing the footage are located. Flock has become a ubiquitous technology in the United States, with cameras in thousands of communities that police use every day to investigate things like car theft. Local police have also conducted multiple searches for ICE in the system.

Companies using artificial intelligence or machine learning often turn to overseas workers to train their algorithms, often because the labor is cheaper than hiring domestically. But the nature of Flock’s business—creating a surveillance system that continuously monitors the movements of US residents—means the film may be more sensitive than other AI training businesses.

Flock’s cameras continuously scan the license plate, color, make and model of all vehicles in transit. Law enforcement can then search cameras across the country to see where a vehicle has been driven. Officials typically review this data without a warrant, and the American Civil Liberties Union and the Electronic Frontier Foundation recently sued a city covered in nearly 500 Flock cameras.

In general, Flock uses artificial intelligence, or machine learning, to automatically recognize license plates, vehicles, and people, including what they’re wearing, from camera images. A Flock patent also refers to cameras that recognize “race.”

Several guides pointed 404 Media to an online panel exhibit that demonstrated various metrics related to Flock’s AI training.

It included figures on “annotations completed” and “annotation jobs remaining in the queue,” with annotations workers add to reviewed videos to help train AI algorithms. Tasks include categorizing cars, colors, and types, transcribing license plates, and “audio tasks.” Flock recently started promoting a feature that detects “shouting”. The panel shows that workers sometimes complete thousands of annotations in two-day periods.

The panel on display included a list of people tasked with annotating Flock’s videos. Taking these names into consideration, 404 Media found that some of them are located in the Philippines, according to their LinkedIn and other online profiles.

According to the leaked material, many of these people were hired through Upwork. According to Upwork’s website, Upwork is a freelance and gig platform where companies can hire designers and writers or pay for “artificial intelligence services.”

Facilitators also pointed to several publicly available Flock presentations that explained in more detail how workers categorize films. It’s unclear what specific camera footage Flock’s AI staff is reviewing. But screenshots in the worker’s guide show several images of vehicles with US license plates, including in New York, Michigan, Florida, New Jersey, and California. Other images include road signs that clearly indicate the footage was shot inside the United States, and one image contains an advertisement for a certain law firm in Atlanta.