The report is in partnership with the bureau of investigative journalism.
In an Amazon warehouse in California, squat blue robots carrying eight-foot yellow shelving units are dancing around each other as they make their way to human workers.
A man named Amari works 42 hours a week as a stower, putting products on the shelves. He said that cameras are trained on your station all the time. It is kind of degrading to have someone looking over your shoulder.
It's not the only people who are watching. If the camera system fails, a video is sent to a person thousands of miles away who can help improve the machine learning tools.
Workers review the videos in India. He said it was a very busy job. We should not blink our eyes while looking at a video. It's kind of painful that we have to be on screen for eight hours.
The two may work in different countries. Both perform mind-numbingly repetitive tasks whose output is strictly monitored, all of it serving to fine-tune the very system used by Amazon to closely monitor its own workers.
Viraj gets through up to 8000 videos a day, with their output ranked against that of their colleagues. Their work can take a toll on their body. $252 a month is the amount of money they are paid.
Video reviewers who were interviewed by the Bureau of Investigative Journalism reported headaches, eye pain, and even worse eyesight. They said they were made to hit punishing targets with tracking software logging any periods of inactive activity outside of break times. Some people said their targets did not allow for screen breaks and that the expert's recommendation was incorrect.
The Bureau interviewed 33 current and former Amazon employees, including 21 video reviewers, to learn more about the little-known outpost of Amazon.
A number of the allegations in the story were disputed by Steve Kelly, an Amazon spokesman.
Computer vision is a branch of machine learning that uses large numbers of images and learns to recognize patterns.
The cameras trained on Amari's station use computer vision to automatically register the location of products in its inventory. The technology has also been used inside the Amazon Go stores.
The system has an accuracy of 95 percent, according to Amazon. Every day millions of images and videos are sent to workers in India and Costa Rica, who decide if a product was successfully stowed and where it is located.
The system is accurate according to Amazon.
Video reviewers said their main role was stock management, but they can also record errors made by their colleagues overseas.
Their nonstop manual work helps to improve the computer vision system by learning from their responses and becoming more and more accurate. The people who teach Amazon's computers to see said their eyes have been damaged by the work.
Prisha said that they need to keep a watch on the videos as they will not be able to blink their eyes. I had a lot of health issues due to that. It makes the eyes dry because you stare at the screen all the time.
Reviewers said they could watch thousands of videos in a single day. Shifts can be increased to 11 hours in the run up to Christmas or Black Friday. Reviewers get around one and a half hours of break time, with any periods of inactive activity instantly tracked by tracking software.
The Costa Rican average was 514,000 colones, while the Indian average was 25,000 rupee.
Humans behind Amazon's all-seeing computer vision are closely monitored. When he was a reviewer in Costa Rica, Mateo would check to see if workers in the US were following covid protocols. There was something unnerving in his feed.
The chairs in the break room were very similar to the ones in his own building. He felt like he was being watched. He thinks someone else was watching him at the time.
Managers must maintain a high accuracy rate of between 95 and 99.5% if they want to keep track of reviewers. Nitara missed her accuracy targets and failed to pass her period. She stated that they were not allowed to make mistakes. It was hard for me to handle. I'm not a robot.
Someone else was probably watching me at the time.
They have a timer on their screen that counts how long it takes. The average time to get through a video will increase if they linger too long, and they may face losing their job.
Prisha said you can't do anything. You might land at the bottom if you give a little gap.
The TAKT ranking system for 25 employees is shown in a document passed to the Bureau by a reviewer in India. The top performer has a time of 5.7 seconds, while the bottom performer has 13 seconds.
Jiyan, a former reviewer based in India, said that the work was stillstressful. The monotony was something that bothered him the most. He said it was very boring. You are doing the same thing over and over again the entire day. There isn't anything new.
During the peak period of October to December, reviewers said their work increased significantly, with less time between videos, and one said bathroom breaks were more difficult to take. A worker in India said he couldn't take time off for the festival of lights.
Workers in India and Costa Rica are encouraged by the software they use to take short breaks throughout their shifts.
In the summer of 2020, Amazon rolled out a new system called "Nike" that relied on computer vision and manual checks from workers in India and Costa Rican.
The location of a product would be recorded with the help of three new cameras that were trained on the stowed stations. The time it would take for each item to be stowed away would be cut in half.
Naomi found it hard to adjust to her new job. She said it was quite nitpicky, the way you had to stand, the way you moved. You couldn't have your own freedom.
While you are doing it, your brain dies slowly.
stowers were told to use clean and straight movements when putting an item away to maximize the computer's chance of success. Video footage was sent to video reviewers when the system failed.
At the California facility, where Jade was a former supervisor, the new system was rolled out. She said there were many different rules about how things had to be stowed.
The new camera system caused many of these reports to be generated.
Amazon doesn't agree with the idea that the automated system was used to discipline workers. In order to identify strengths and opportunities for growth, metrics are shared with employees and managers. Nike cameras are only used for inventory purposes.
Jade would check to see if workers were doing well in other metrics, such as how long they were inactive outside of break times, or how many units they stowed per hour.
A write-up was given to a former stower at a Michigan warehouse after he accumulated four minutes of time off. He was unwell and went to the bathroom at the end of his break. He received a written warning even though he explained.
Amazon insists that Isaac's experience isn'trepresentative. Kelly told us that employees are told to take short breaks throughout the day.
Jade would check to see if workers were doing well in other metrics, such as rate, the number of units they stowed per hour, and time off task.
She thought the productivity targets were reasonable and people could hit them. She said that the work was "mind-numbing."
She said that the brain dies slowly while it is being used. People in Costa Rica are doing the mind-numbing brain side. You are treated like a machine.
When covid protocols were in place across the world, the company said it would retool its machine learning systems to help enforce social distance in its warehouses.
The six-foot green circle around the workers' feet was displayed on large screens. The ring would turn red if they got too close to a coworker. Images were sent abroad in cases where the computer was not sure how close they were.
In Costa Rica, Thiago worked on the Proxemics team. I don't think it's the best one I've ever done.
It was difficult to look away from the screen if you wanted high metrics. He said his eyes would cry at the start. You get at least one image every six seconds. It was really bad.
If you look to the side, you have already had 10 seconds on your screen.
Video reviewers in India have become the shop assistant's replacement.
Video reviewers are at risk of dry eyes, impaired vision, and headaches due to digital eye strain. There is a chance of permanent eye damage if the resolution is low.
Screen breaks should be every 20 to 30 minutes, according to him. The Bureau interviewed some workers who said it was hard to take breaks outside of their allotted time.
Thiago said he was given an extra 10 minutes to use the bathroom when he wasn't on break.
He was given access to the employee monitoring software when he was promoted to leader. He said that they could tell how long you hadn't touched the screen or mouse.
Reviewers have worked on footage from Amazon Go grocery stores, which use computer vision to detect what a customer has purchased, billing their card automatically.
The futuristic novelty of an unstaffed store is the focus of the marketing. Video reviewers in India have taken over the role of shop assistant.
Ishan, who worked on the Amazon Go team, said he was given just four minutes a day to use the bathroom outside of his scheduled breaks. Some will survive and others will have to leave. An employee is expendable for the company.
The pay was very low for the high level of stress.
The accounts given to the Bureau don't represent the majority of those working on the team, according to Amazon.
The people being filmed by Amazon don't know much about the people watching them. Only one of the nine workers interviewed said they knew that footage from their station could be sent to other countries.
Some workers in India and Costa Rica don't know how Amazon uses the fruits of their labor. One worker said they didn't know where the data was going. We didn't know what was happening in the back end.
We didn't know where the data was going.
There is a new law in California aimed at addressing the conditions of Amazon workers. Assembly Bill 701 came into effect at the start of this year and prohibits performance targets from being set at a level that could pose a safety risk. Warehouse workers can request three months of their own productivity data.
The bill was written by Gonzalez when she was a state lawmaker. She thought that giving workers access to their own data would be the first step in resisting management.
Gonzalez said that they wanted to make sure that they had the right to that information.
According to the stower in California, the law has made a noticeable difference in the warehouse and managers are no longer disciplining workers for falling behind on productivity.
The work of a stower in California is very different from the work of a video reviewer in India. They are constantly generating new data points to make the tools that monitor and discipline them better.
Gonzalez said that the only way to fight back was by refusing to go quicker. An individual can't do that on their own.
Unless workers come together and organize as a collective against the kind of speed-up that happens with computer management, they're just going to require people to go faster and faster.
All Rights Reserved © 2024
