Tesla seeks to develop autonomous cars; Tesla is increasingly looking to machines instead of humans. Data annotation specialists use software tools to label objects in Tesla vehicles‘ video clips manually. Specialists label everyday road objects such as stop signs, lane lines, traffic cones, and curbs. The AI system then uses the tagged information to learn how to perceive its environment accurately. The better an AI system is, the more correctly it labels data.
Elon Musk, Tesla CEO, stated at the August 2021 AI Day that “without the Auto-Labeling, I think it would be impossible to solve the Self-Driving Problem.”
Tesla executives claim that the company’s efforts to automate data Labeling have already promoted its pursuit of self-driving automobiles. Ashok Elluswamy was the director of Autopilot software. He stated at AI Day that Tesla had been able to capture 10,000 video clips from their cars and then automatically label them all within one week. Clips can be 45 to 60 seconds long and include related GPS or odometer data.
He explained that “this would have taken several weeks with humans Labeling every clip.” He also mentioned plans to produce one million auto-labeled clips to ” crush that problem.”
Raj Rajkumar (a Carnegie Mellon University professor studying autonomous cars) says there isn’t a clear answer about whether automated data or manual Labeling is better. He suggested that Tesla could keep some people involved to find automated labeling flaws. He stated, “If you do an economics of fewer human beings, it’s financial gain.”
For five years, Tesla had used an outside firm to label its self-driving data. However, the company has recently moved to the house, Andrej Karpathy said last year. Data labelers were employed in San Mateo, California, and Buffalo, New York. He said this is crucial for improving the quality of Tesla’s data. He said that the company now has more than 1,000 employees.
The human need to work is not eliminated by job cuts or Auto-Labeling. Tesla has continued to post job openings in data annotation for data workers.
“For us, it’s becoming more about the story of “How can humans and machines collaborate to create these vector spaces data sets?” Karpathy, speaking at AI Day.
Karpathy stated that Tesla wanted accurate Auto-Labeling, which may have affected Tesla’s speed in obtaining it.
Artificial intelligence experts predict that human annotators will become less important as technology is developed that doesn’t require costly work.