Movies give us a glimpse of what the future holds. From smart phones, tablets, facial recognition, virtual reality and now self-driving cars. The idea of a vehicle that can run itself seems like something straight from science fiction. Fortunately for us, this is a reality but have you ever thought of how driverless cars are made possible?
Elon Musk recently announced Tesla will have fully autonomous vehicles by 2020. This is good news but what he left out of the speech is the manual aspect that makes autonomous driving possible. Big tech companies prefer not to talk about this, they prefer to say is “machine learning” magic. Truth is, the road to fully autonomous vehicles requires a lot of human input.
Companies dealing with autonomous driving technology employ hundreds if not thousands of people. They work in offshore outsourcing centers in India, Philippines and Kenya which is emerging as an African giant in the BPO industry. The workers job is to basically teach the robo-cars how to recognize pedestrians, cyclists, poles and other obstacles. This is done by manually marking and “labeling” thousands of hours of footage, often frame by frame. These videos are taken from vehicle prototypes driving around testbeds in Silicon Valley, Pittsburgh and Phoenix.
Given the dynamic nature of the outdoor environment, companies should ensure their team remains intact for some time. Despite the huge leaps forward in Artificial Intelligence and sensor quality that serve as the foundation of the driverless revolution, humans will still be needed for many years to come. Drawing boxes around new obstacles, highlighting road signs will always require human input to keep the systems fresh.
Machine learning is a myth, computers “learn” by being fed an array of manually labelled information which they then use as a model to identify objects and patterns when they encounter them again. The annotation process is a very hidden part in autonomous vehicle development that companies never mention. For the workers, annotation is a super painful and tiresome process given the level of accuracy required in autonomous cars.
Having seen the amount of work that comes with image tagging and annotation, it’s a worthy reminder that fully self-driving cars will be a plus for society. With self-driving cars, we can envision accidents being a thing of the past. Computers don’t get drunk, check social media feed or fall asleep while on the wheel.