Are Human Drivers Finally Obsolete?

What It Covers

PJ Vogt traces the 20-year development of autonomous vehicles from DARPA's 2004 desert robot race through Google's secret California road tests to Waymo's current 10-city robotaxi rollout, examining safety data showing 80% fewer injury-causing crashes than human drivers, while previewing the political battle over 4.8 million American driving jobs now under threat.

Key Questions Answered

• •Safety Data Benchmark: Waymo's published crash data across 127 million miles shows 80% fewer airbag-deploying crashes and 90% fewer injury-causing crashes compared to human drivers. Independent researchers broadly validate this methodology. The fatal crash comparison remains statistically inconclusive — academics estimate 300 million miles are needed for confidence — but current results favor autonomous performance over human drivers.
• •Consumer Confidence Gap: JD Power data reveals a stark perception divide: only 20% of people who have never ridden in a robotaxi express confidence in the technology, while that figure jumps to 76% among actual riders. This suggests public resistance to autonomous vehicles is driven primarily by unfamiliarity rather than evidence, meaning direct exposure is the most effective trust-building mechanism.
• •Technology Readiness Divergence: At the time of Uber's fatal 2018 Arizona crash, Waymo safety drivers intervened once every 5,600 miles, while Uber's required intervention more than once every 13 miles — a 430-fold performance gap. Despite this disparity, Uber reduced its safety crew from two humans to one, five months before the crash, over internal employee objections.
• •AI Training Scale Effect: Neural network performance for autonomous driving improves non-linearly with data volume. Sebastian Thrun describes feeding 100 million documents producing adequate results, but 100 billion producing dramatically superior outcomes. This scale threshold explains why Waymo's continuous road mileage accumulation functions as a compounding competitive advantage that newer entrants cannot quickly replicate.
• •Contextual Physics in Autonomous Driving: Human driving comfort is not governed by fixed physical tolerances but by situational context. Research by Google engineer Don Burnett found acceptable lateral acceleration on highway on-ramps measures 2.0 meters per second squared, but drops to 0.75 on residential cul-de-sacs — nearly three times lower — despite identical physical forces. Autonomous systems must encode this contextual awareness, not just raw physics limits.
• •Job Displacement Scale: Approximately 4.8 million Americans currently drive professionally, making it one of the most common occupations in the country. Historical parallels — lamplighters in Belgium organized violent strikes before losing to electrification — suggest organized resistance is likely. Current political organizing in cities like Boston represents early-stage friction that could significantly slow autonomous vehicle deployment timelines regardless of technological readiness.