Self-driving technology depends on more than recognising a lane or detecting another vehicle. An automated system has to understand its surroundings, predict what other road users may do, plan a safe path and control the vehicle. AI contributes to several of these steps, but the reliability of the complete system matters more than a striking demonstration in ideal conditions.
Four stages of automated driving
The source material describes perception, prediction, planning and control, using data from cameras, radar or lidar depending on the system. Waymo’s Level 4 robotaxi service operates within defined areas and conditions. Tesla’s Full Self-Driving (Supervised) is a Level 2 driver-assistance system that requires an attentive human driver. Mobileye and other suppliers provide technologies that automakers can integrate into their own vehicles.
Level 2 and Level 4 are not the same
The source notes reports of more than 250,000 paid Waymo rides per week in 2025 and references large volumes of Tesla supervised-driving miles. Those measures are not directly comparable. Ride counts, miles, crash rates and intervention data describe different things, and company-reported figures need a clear time period and methodology.
Why mileage and ride counts need context
AI may improve detection and response, but road safety also depends on weather, construction, unusual objects, human behaviour, mapping, vehicle maintenance and the conditions in which a system is allowed to operate. Regulators and fleet operators need transparent incident reporting and careful comparisons against human driving under similar conditions.
The conditions shape safety
AI can make driving systems more capable, but a claim that a vehicle is safer than a person needs robust evidence across real-world exposure. Check the automation level and the operational limits before treating any system as autonomous.
Evidence matters more than demonstrations
A meaningful safety comparison should state the operating domain: roads, speeds, weather, mapping conditions and whether a human is supervising. It should also include the severity of incidents and the miles travelled, not only the number of reported events. Regulators and independent researchers need access to consistent data before broad claims can be assessed. A system that performs well in one city may not be ready for every road.