Scale AI Teaches Electric Cars to Learn to “See” Faster

can be reached at meehna@gmail.com
can be reached at meehna@gmail.com
Share on facebook
Share on twitter
Share on pocket

A startup in San Francisco called Scale AI has plans to help electric cars process visual information more efficiently. Currently, Artificial Intelligence (AI) systems in vehicles learn how to “see” via a system called labeling. Tedious and time consuming, it relies upon actual people looking at driving footage and identifying and labeling images, such as a cyclist, car, roadblock or even a  tumbleweed. Once labeled, images are fed back into the AI software. The software assimilates the images, and learns over time to recognize objects on its own.

  • Artificial Intelligence (AI) systems learn to see via a process known as labeling, which takes an army of people working innumerable hours to categorize pictures. 
  • Scale AI has developed a software tool that automatically labels images, which is then checked over by a worker. 
  • By automating a big chunk of the process, Scale AI’s technology could cut costs and help get autonomous cars on the road faster. 

Scale AI believes that they can make the process easier for both teacher and student. Co-founded by 22-year-old wunderkind Alexandr Wang, the company has created software tools that markup images before passing them off to a network of about 30,000 contract workers who check and finalize the work. Apparently, the software can label most of the objects, which are then reviewed by a person for accuracy. To make a change requires just a quick mouse click on an object for the system to trace it.

 tumbleweed
Scale AI could identify this tumbleweed. (Photo: Getty Images)

The typical labeling system involves a worker viewing images on a computer and using a mouse to outline all the objects and categorize them: everything from buildings, traffic signals and pedestrians, to parking spaces and, yes, tumbleweeds. Depending on the complexity of the picture, it can take from 10 minutes to a couple hours for an individual to trace all those objects. Now multiply that by millions of images. It truly is an excruciating and expensive proposition.

Scale AI is already working with big players in the self-driving space, including Waymo, Cruise and Uber. Now it plans to license its technology to other companies.

“It takes billions or tens of billions of examples to get AI systems to human-level performance,” says Wang. “There is a really big gap between the handful of giant companies that can afford to do all this training and the many that can’t.”

Scale AI must be on to something because it gained the support from some prominent venture capitalists. This month it raised $100 million in a Series C funding round led by billionaire entrepreneur Peter Thiel’s Founders Fund, that pushed its valuation past $1 billion.

WHY THIS MATTERS

AI systems in vehicles learn through a labeling system that is laborious, expensive and slow. If Scale AI’s new technology works, it will slash the amount of time it takes, while also cutting costs. Most important, it could bring the day closer when you can actually relinquish the wheel to your car.


About the Author

can be reached at meehna@gmail.com
Close Menu