Melting Labour: Asking about My Work’s Future

This Place May Be a Factory Growing Human “Parts” to Increase the Accuracy of the Machinery

2020.01.29 22:51
Sim Yoon-ji

Yi Seung-tae, a manager at CrowdWorks explains the concept of machine learning and artificial intelligence (AI) training data using a picture of the street captured by a closed circuit camera. For AI to analyze the image, people must process the image to facilitate AI’s learning by bounding and labeling each object, such as “car” and “person.” When such training data is sufficiently collected, AI can identify cars and people by looking at a closed circuit image on its own.

Yi Seung-tae, a manager at CrowdWorks explains the concept of machine learning and artificial intelligence (AI) training data using a picture of the street captured by a closed circuit camera. For AI to analyze the image, people must process the image to facilitate AI’s learning by bounding and labeling each object, such as “car” and “person.” When such training data is sufficiently collected, AI can identify cars and people by looking at a closed circuit image on its own.

“The picture of an exit with a fire hydrant, you can all see it, right? All you have to do is type in the letters in this picture without leaving a single letter out. As for the space between the words, you bound the word and then type in the letters. In the case of the word ‘hydrant’ here, the space between the letters is quite wide, so you’ll have to type it in as ‘h y d r a n t.’”

The hands busily maneuvered the mouses. Some people pressed their faces close to the monitor in case they missed anything. Other than the occasional question, there was no talking. The task was simple. All they had to do was bound each word and then type in the letters so that the computer could read the words in the picture. It was the same as teaching a child learning to read Korean for the first time--you point to each letter one by one and teach her how to pronounce the letters.

The data created in this manner will be used to train artificial intelligence (AI). The ingredients listed in the food label, the notes taken during a world history class, the KakaoTalk conversation with a colleague at work…. As the data collected and processed from various environments pile up, the computer will be able to smoothly read the words inside pictures it sees for the first time. The application that translates menus that users capture with their cell phone cameras on overseas trips was also created based on this data.

The Kyunghyang Shinmun dropped by a training session for new “CrowdWorks” staff conducted in Gangnam, Seoul on December 10. CrowdWorks, established in April 2017, is a company that provides a platform for processing training data for AI. It split up the task of collecting training data, a challenge for all AI firms, into small tasks and then solved the problem by handing the tasks out to the public through crowdsourcing. In a way, it took the collection of AI training data from Amazon’s crowdsourcing platform, the Mechanical Turk, and relocated it to South Korea.

In an interview, CrowdWorks CEO Park Min-woo said, “We can split 70% of our work into unit tasks. That practically means that the task can be handled at home using a platform.” He predicted an end to an age where people worked from nine to six at the workplace and said, “A time will come when more than 50% of the working population will be freelancers, contract workers, and part-timers.” Is this the future of labour?

■ Digital Manual Labour in the 21stCentury

“Sewing eyes onto dolls in the age of the Fourth Industrial Revolution.” This is the nickname given to CrowdWorks. For the simple task of typing in Korean, workers receive 20 won per letter. If they type in all the letters in a picture, they receive 1,000 to 5000 won. Extracting the shape of a fingernail, accurately capturing the location of joints in the hand, creating a Q&A text according to the five W’s and one H after reading a newspaper article…. The fee for all the tasks that the staff learned during the training is only some dozens or hundreds of won each. Yi Seung-tae, the manager in charge of the training said, “Many people wonder if they can actually earn money doing such simple tasks. According to the accumulated income of top workers on the CrowdWorks website as of November 25, there was a person who earned more than three million won a month.” Yi continued his explanation as if familiar with the doubt, “How can a person earn three million won by getting paid 20 won per letter?”

“The high-income staff works more than ten hours a day, six days a week. Just because they earned that much this month, doesn’t mean they will receive that much the next month, and just because someone else earns that much doesn’t mean that you can earn that much either. But if you work consistently for a long time and are good at the job, you can definitely earn more than the minimum wage. If you do it several times, you’ll begin to smell the money.” Many of those present said they came after watching a YouTube video of a person who said, “I earned 80,000 won a day with CrowdWorks.”

Yi said, “I don’t recommend people to do CrowdWorks full time yet.” This was because in the entire labour market, task-based labour still remains a minority. But it seems clear that such labour is quickly increasing. CrowdWorks has over eighty client firms including large corporations, public agencies, and IT firms, such as Naver, Kakao, Samsung Electronics, Hyundai Card, and the Korea Labour Institute. It also hires full-time regular managers at the speed of one manager a week. The company was also selected as a supplier for the “Data Voucher Support Project”--a project in which the Ministry of Science and ICT funds AI companies--and is receiving state funding.

Human attention is still necessary in tasks that appear to be simple. When the size of the letters is different, one needs to decide whether to type them in as one word or not. If different letters overlap, it could get confusing as to how to handle the letters. On the CrowdWorks online bulletin, a stream of questions comes up concerning the rules of the task and the standards for returns. When AI becomes more sophisticated, maybe someday such simple tasks will disappear. In the case of Remember, an application that manages name cards, people had to manually input information on the name cards in 2014 when the app was first introduced, but now, the human workforce has been reduced to a tenth.

“They say the intelligence of AI is about that of an average eighth grader. That means the computer still needs to learn all the contents you see, hear, speak, read, and write.” But how long will that “still” last? When that time comes, will the human capability to “see, hear, speak and write” no longer be useful? Perhaps the training session was a factory growing human “parts” to increase the accuracy of the machinery. The thought gave this journalist goosebumps.

■ The End to Wage Labour

In the room where the training was in progress, over a dozen men and women in their twenties to fifties filled the seats. They had some things in common. They were either tired of being harassed by their boss at work from nine to six, didn’t earn a sufficient income, or had dropped out of such a life.

Bak Yun-ji (alias) was a freelance web designer in her thirties, who had had her career interrupted by the birth of her son, now six years old. She received work from the talent-sharing platform Kmong, but it was hard for her to win in a competition with other experts who designed websites full time. The alternative she chose was to seek out various task labour. From producing and selling smartphone applications and the Naver Smartstore to investing in the foreign exchange market, she said, “If you count all the jobs that I have done so far, I have had more than ten.” This wasn’t what she wanted. “I had no choice but to quit work, because I had no one to look after my son for me. When he was about three-months old, I wanted to return to work, so I learned coding for a year at a training center funded by the state. I swear I sent my resume to over a hundred places for two years, but less than 10% asked me to come in for an interview. I wasn’t able to get past those interviews when they asked, ‘How are you going to work overtime with a child?’”

For Gang Min-jeong (50, alias) from Ulsan, CrowdWorks was a part-time job she could do while studying at graduate school. She used to introduce an offshore plant firm in Ulsan to overseas clients. She fell ill due to excessive work and stress. She recovered and when she tried to find work again after five years, the offshore plant industry had declined. “The world is changing so fast, I thought the risk was too big to earn money the old way.”

There were also young people in their twenties, such as a student looking for a part-time job he could do at home to earn an allowance while continuing his studies overseas after graduating from college and an employee of a startup interested in a side job because of an irregular income. They all discovered this place while desperately seeking an answer to a situation where there aren’t enough regular jobs.

President Park Min-woo’s argument that “Now wages based on tasks, not hours, will be the most efficient form of labour” left an impression in their minds. “The intensity of labour may grow higher, but that will also be up to the individual. It is not a bad change even from the perspective of the worker, because she can concentrate more during working hours and do more work to earn money and enjoy the rest of her time.” Who could easily argue with these words by Park, even if it was apparent that the workers did not opt for this change?

추천기사

바로가기 링크 설명

화제의 추천 정보

    오늘의 인기 정보

      추천 이슈

      이 시각 포토 정보

      내 뉴스플리에 저장