Chapter 84: Content Moderation: Twitter, October 27–30, 2022

This chapter examines Elon Musk's approach to autonomous driving technology at Tesla, focusing on his insistence on camera-only vision systems against engineer advice and his pattern of making repeatedly optimistic public predictions about Full Self-Driving capabilities that consistently failed to materialize.

Musk rejected Google's LiDAR-based approach to autonomous vehicles on principle and cost grounds, believing that since humans drive using only visual data, machines should too. Despite engineering pushback from team members like Dhaval Shroff and Drew Baglino, who advocated for radar and ultrasonic sensors as safety supplements, Musk pursued a camera-only vision system. Early challenges included a problematic curve on Interstate 405 where faded lane markings caused the Autopilot to swerve dangerously; the team ultimately solved it by having the highway lines repainted through a SpaceX tour arrangement. By 2016, Tesla equipped vehicles with eight cameras, twelve ultrasonic sensors, and forward-facing radar, though Musk remained committed to eventually eliminating radar. The chapter then documents fatal accidents involving Autopilot, including a May 2016 crash in Florida where a driver watching Harry Potter was killed when the system failed to detect a white tractor-trailer against a bright sky. Musk resisted public outcry over these deaths, arguing statistically that Autopilot prevented more accidents than it caused. He made increasingly bold predictions about Full Self-Driving: in October 2016 he promised a Los Angeles-to-New York autonomous drive within a year; at a 2017 TED Talk he claimed two years; in 2018 he said next year; in 2019 he declared feature completeness certain that year. Designer Franz von Holzhausen created Robotaxi mockups to maintain momentum, though full autonomy remained elusive. By the end of 2022, Musk acknowledged the challenge was greater than expected, recognizing that Full Self-Driving required solving real-world artificial intelligence—a problem far harder than his timeline predictions had suggested.