Tesla's Autopilot system, designed for hands-free driving, is under scrutiny following revelations about its data annotation practices. Employees from Tesla's data annotation team reported that the company sometimes disregarded traffic rules in training the system, aiming for a more 'human-like' driving style.
According to a report by Business Insider, workers were instructed not to teach Autopilot to recognize certain traffic signs, such as 'No Turn On Red' or 'No U-Turn.' A former employee described the approach as a 'driver-first mentality,' prioritizing human-like behavior over strict adherence to traffic laws.
Additionally, employees faced challenging working conditions, including viewing disturbing footage of accidents and being monitored through surveillance. Despite efforts to train the system, significant gaps remain in its ability to recognize various road hazards, as highlighted by a recent Wall Street Journal investigation into over 200 crashes involving Tesla vehicles.
Experts suggest that the reliance on cameras and machine learning contributes to these flaws, with many scenarios unaccounted for in the training data. Tesla CEO Elon Musk has dismissed the need for additional radar or LiDAR technology, which could enhance safety. This ongoing controversy raises concerns about the safety and efficacy of Tesla's autonomous driving technology.