While AI was evolving, Tesla’s lockout hit the autopilot team
WASHINGTON, DC (CNN) – Tesla (TSLA) is turning to machines instead of humans as it seeks to develop autonomous vehicles.
As part of Tesla’s plan to cut 10% of its salaried employees, the company has significantly reduced the number of data commentators. These specialists play a vital role in enabling artificial intelligence systems to handle complex tasks, such as safe driving on city streets.
The layoffs were first reported by Bloomberg on Tuesday and confirmed by CNN Business.
Data commentators use software tools to manually label objects in video clips collected from Tesla vehicles. Specialists label common road objects such as lane lines, stop signs, traffic cones, curbs, and traffic signals. The labeled data is inserted into the Artificial Intelligence System so that it learns to know its environment accurately. The more properly labeled the data in an AI system, the better.
Tesla has in recent years developed an automated way of doing some of these labeling work, allowing the automaker to streamline its workforce.
“I don’t think we can solve the self-driving problem without auto labeling,” Tesla CEO Elon Musk said at the company’s AI Day in August 2021.
Tesla officials have suggested that automated data labeling has already accelerated its work on self-driving vehicles.
Ashok Eluswamy, director of Autopilot Software, said at the AI Day event that Tesla is capable of collecting 10,000 video clips from his car and labeling them automatically in a week. Clips typically contain 45 to 60-second video sections, as well as related GPS and odometer data.
“It would have taken months for humans to label each clip,” Eluswamy said. He also described plans to create millions of auto-label clips “to really crush this problem.”
According to Raj Rajkumar, a professor at Carnegie Mellon University who studies autonomous vehicles, there is no clear answer as to whether manual, human labeling or automatic data labeling is more accurate. Companies like Tesla can engage some people to catch the flaws of automated labeling, he said.
“If you do low-income economics, it’s an economic victory,” Prince said.
About five years ago, Tesla relied on outside business to label its self-driving data, but in recent years tried at home, Andrej Karpathi, who heads AI at Tesla, said last year. Data Labelers has worked in San Mateo, California and Buffalo, New York. He described it as important to improve the quality of Tesla’s data. The company has built a team of more than 1,000 people, he said at AI Day in 2021.
Job cuts and auto labeling do not eliminate the need for human labor. In fact, Tesla is publicly posting some job opportunities for data comment employees.
“It’s becoming another story for us, ‘How do humans and computers actually collaborate to create these vector space data sets?'” Carpathi said at AI Day.
Carpathi said Tesla wanted auto labeling to be extremely accurate, which would have affected how fast Tesla turned.
Artificial intelligence experts say that there will be less need for human commentators in the future because techniques have been developed that do not require expensive work.
“The future of data commentary is low,” Pedro Domingos, a professor of computer science at the University of Washington, told CNN Business. He cited the example of AI systems for languages that learn from large amounts of text data.
Adela Petrescu, a former autopilot data commentary supervisor at Tesla, posted on social media on Tuesday that she had been fired. Petrescu said she received promotions twice a year and never had a performance problem.
“For the past [two] years I have worked 50 [more] hours per week, many weekends and many weeks with 16 hour days and for all the purpose for which I truly believed, I still do – [a] good future for our future generations , “Petrescu said.
Tesla did not immediately respond to a request for comment and is not generally engaged in commercial news media.