Can AI drive a car more safely than the average human on public roads ?

The question asks whether autonomous vehicles can navigate public roads with a lower collision risk than the average human driver. Recent advances in driver-assistance and fully autonomous systems suggest a narrowing safety gap—but how robust is the evidence, and where do the systems still fall short?

Background

Autonomous driving systems have reached an inflection point where AI can navigate complex traffic scenarios with fewer human interventions than typical drivers. Advanced driver-assistance systems (ADAS) now handle highway driving, lane changes, and even city intersections with remarkable precision. Some AI stacks have logged millions of miles without a single at-fault accident, outperforming human benchmarks in reaction time and error rates. The technology leverages deep learning models trained on diverse real-world driving data and real-time sensor fusion.

Today’s leading self-driving systems (such as Waymo Driver, Cruise, and Mobileye) can operate on public roads in geofenced areas under normal conditions, matching or exceeding the safety performance of human drivers in terms of collision rates per mile, according to internal safety reports. However, these systems struggle with rare or unpredictable edge cases (e.g., sudden construction zones, unusual pedestrian behavior) and still require remote human oversight in many deployments. General availability as a mainstream consumer product remains limited and constrained by regulatory, technical, and insurance hurdles. Significant progress is continuous, but full autonomy under all weather and traffic conditions is not yet realized.
— Enriched May 12, 2026 · Source: National Highway Traffic Safety Administration (NHTSA)