Artificial intelligence is not self-taught. Behind them, there is a group of hard-working teachers.
The Translation Bureau is a compilation team that focuses on technology, business, workplace, life and other fields, focusing on foreign new technologies, new ideas, and new trends.
Editor’s note: There is no doubt that the field of artificial intelligence is undergoing rapid development and progress every day. However, in many countries around the world, including China, there is an emerging profession that you may not know. They are responsible for teaching artificial intelligence systems. They are the most lovely people in the group. Some people call them digital Foxconn “first-line workers.” This translation from the New York Times article, the original title is A.I. Is Learning From Humans. Many Humans, author Cade Metz introduced the story of this industry. This is the first part of this series, mainly through a company based in India, which introduces the fact that artificial intelligence is not self-study, and there is a group of hard-working “teachers” behind them.
Recommended reading: 500,000 patients per year Endoscope, why is it so difficult to disinfect?
iMerit is based in Kolkata, India. Image source: Rebecca Conway for The New York Times
Bhubaneshwar, the capital of Orissa, India, is about 40 miles (65 km) from the Bay of Bengal. In the city centre of Bhubaneswar, Namita Pradhan is sitting in front of his workbench, watching a video on the screen in front of him. This video was recorded in a hospital on the other side of the globe.
The content of this video is the picture inside the colon of a patient. The work that Pradden did was to find polyps through video. Polyps are the neoplasms that protrude into the intestines of the large intestine. It may cause cancer. It looks a bit like a small pustule.
When Pradane is in the videoAfter finding the polyp, she will immediately mark the digital “border line” around the small pustule through the mouse and keyboard.
Pradan is not a medical professional, but her work is to teach the “teacher” of the artificial intelligence system until the artificial intelligence system graduates and can complete the doctor’s job duties.
Puradan’s “classroom” is located on the fourth floor of a low-rise building. In this “classroom”, in addition to Pradesh, there are dozens of other Indian youths who are doing the same job. In the training before they officially took up their posts, they learned how to label various digital images, and can accurately mark any content on the image, such as traffic signs and pedestrians in real streetscapes, and satellite maps. Factory and tank trucks, etc.
Many people in the technology industry may tell you that artificial intelligence is bound to be the future of this industry. Thanks to the application of machine learning in development, today’s artificial intelligence is undergoing rapid changes.
However, those who have the right to speak in the field of technology rarely mention the labor-intensive processes behind these rapid developments. In fact, artificial intelligence is not self-taught, they are taught, and the people who teach them can even be described as countless.
Before the AI system begins to learn, you must first enter the data that has been manually labeled, as in the previous example of artificially delineating colon polyps. This work is crucial, whether it is in the field of driverless, surveillance systems or automated medical and health, the development of artificial intelligence is inseparable from this work.
Basically all technology companies are avoiding this work. However, because they store a large amount of personal privacy information, and even share this data with external third parties, more and more active people who are concerned about privacy have come out.
In the first half of this year, through hard work, I won an opportunity to see the story behind it. Such a rare opportunity is usually unsuccessful in Silicon Valley.
From the beginning of Bhubaneswar in Orissa, India, to the last New Orleans, Louisiana, I have visited five such offices. In these offices, all the staff are doing endless repetitive work – teaching artificial intelligence systems to learn various annotation data. And their “teaching” system comes from a company called iMerit.
In these offices, there are many colon inspectors like Pradang. Of course, there are also “teachers” who are responsible for other “teaching” tasks, such as real streetscape recognizers, or language and symbolists.
Is this a pedestrian? Is this a double yellow line or a dotted white solid line? One day, no one will driveCars must know these differences.
IMerit employees must have data tagging skills to be employed. Image source: Rebecca Conway for The New York Times
What I saw in these offices was not the one that saw the future. The buildings in which these offices are located were originally telephone call centers or payment processing centers.
One of the offices is located in the low-income residential area at the western end of the city of Calcutta, the capital of West Bengal, India, and the building where the office is located is a dilapidated building that was once used as an apartment. The surrounding area of the building is bustling pedestrians, vendors selling on the street, and shuttle cars.
Other cities in India also have the same offices as Bhubaneswar. In addition, these offices exist in China, Nepal, the Philippines, the United States, and several countries in eastern Africa. In these offices, thousands of “teachers” sit. They work at the card, the only job responsibility is to teach artificial intelligence systems to learn.
In addition, there are thousands of telecommuting “teachers” who complete the “teaching” work at home, and also complete the image data annotation work. In most cases, they are powered by Amazon Mechanical Turk (AMT), an Amazon labor crowdsourcing platform. On the AMT platform, anyone can initiate digital tasks. No matter which country the “teacher” comes from, you can claim and complete the work alone. However, the reward for labeling data information is not very high.
IMerit, headquartered in India, serves many giant clients in the technology and automotive industries. At first, because of the confidentiality agreement, they had rumored to reject the request to disclose their customer information. But recently, according to iMerit, more than 2,000 employees in nine of its global subsidiaries are struggling to contribute to Amazon’s online data tagging service called SageMaker Ground Truth. Before that, they also publicly revealed that Microsoft is also their customer.
In iMerit, employee art and motivational text are attached to the office wall. Image source: Rebecca Conway for The New York Times
One day in the future, but we still don’t know what day it is, artificial intelligence will definitely hollow out the entire talent market. But for now, it brings a relatively low-paying job opportunity.
In 2018, the entire data labeling market exceeded $500 million. According to market research firm Cognilytica, by 2023, the market will exceed $1.2 billion. In terms of data labeling alone, this work accounts for 80% of the time it takes to develop artificial intelligence technology.
So, is this work exploitative? The answer to this question depends on your region and city, and the work you do. In India, this job will allow you to enter a medium-level life; in New Orleans, it can be counted as a decent job. For an “individual contractor”, it is more like a dead end.
Image Source: iMerit
To be qualified for the job, you must also learn the necessary skills in advance. For example, to identify signs of a disease through video or medical scans, or to maintain a constant hand strength when delineating a digital circle for an element of the picture. In some cases, it may involve medical surgery videos as well as pornographic or exposed images, which may be uncomfortable.
“When you first see this content, you will feel very upset. You may even have the idea of giving up the job.” Christie, who has been involved in data tagging on the Amazon AMT platform for several years. According to Kristy Milland.
“But some people may lose their income if they give up the job. So they have to endure silently.. “Milan added.
Before leaving for India, I tried to tag images through a crowdsourcing service platform. I have tried to draw a digital borderline around the Nike logo and I have tried to identify the “Not Safe for Work (NSFW)” image. I found that I might be really hard to do the job.
Before I started working, I had to pass a test. Almost on this test, it almost collapsed, and I failed three times in a row. The whole experience process is very frustrating. I need to label all kinds of information in different pictures so that people can immediately search the relevant retail products online. In addition, in the process, I also spent a lot of time to identify offensive content such as naked women and sex toys in the picture, and labeled them “NSFW.”
For AI researchers, they want to be able to build a system that can learn from a small amount of data. But from the foreseeable future, human labor is still an unattainable factor.
“This is a world of rapid expansion, a world hidden behind technology.” Anthropologist at Microsoft, Mary Gray, who is the author of Ghost Work. Say, “If you want to exclude humans, it is not realistic.”
Translator: Ishii Junichi