Image Source: The Verge
Tesla is increasingly relying on machines rather than people to create autonomous vehicles.
Tesla has laid off a sizable number of its data annotation specialists to reduce the number of salaried employees by 10%. Yet, for artificial intelligence systems to be able to undertake challenging tasks like safely navigating a city roadway, these specialists perform the grunt work that is essential.
Bloomberg first reported the layoffs on Tuesday, and CNN Business later verified them.
Data annotation specialists employ software tools to manually label items in video recordings gathered from Tesla vehicles. Lane markings, stop signs, traffic cones, curbs, and traffic signals are among the common roadside elements that experts label. The tagged data is used to train an artificial intelligence system to comprehend its surroundings appropriately. An AI system can perform better the more accurately labeled data it has.
In recent years, Tesla has created an automated method to complete some of this labeling work, allowing the company to reduce the number of employees it employs.
Tesla executives claim that automating data tagging has already sped up their development of self-driving cars.
At the AI Day event, Ashok Elluswamy, director of Autopilot software, said that Tesla was able to gather 10,000 video clips from its cars and automatically identify them in just one week. The majority of clips consist of 45 to 60-second video chunks together with pertinent GPS and odometer data.
Raj Rajkumar, a Carnegie Mellon University professor, specializing in autonomous vehicles, says there is no conclusive proof that manual, human labeling is more accurate than automated data classification. But, according to him, businesses like Tesla might continue to use certain humans to spot automatic labeling mistakes.
Tesla used to rely on an outside company to categorize its self-driving data about five years ago, but Andrej Karpathy, who oversees AI at Tesla, stated last year that the company has since brought its efforts in-house. Both San Mateo, California, and Buffalo, New York, have employed data labelers. He emphasized how important this is to raise the caliber of Tesla’s data. He stated during AI Day in 2021 that the business assembled a staff of more than 1,000 individuals.
Auto labeling and job losses do not make human work unnecessary. In reality, Tesla still makes some job postings for data annotators available to the public.
Karpathy claimed that Tesla wanted the auto labeling to be very precise, which may have slowed Tesla’s use of it.
Experts in artificial intelligence predict a decline in the demand for human annotators as more efficient methods is developed.
Adella Petrescu, a former supervisor of Tesla’s Autopilot data annotation, announced her termination on social media on Tuesday. In one year, Petrescu claimed she received two promotions and never experienced performance concerns.
A request for comment from Tesla was not immediately answered, and the company often avoids interacting with established news organizations.