That seems like arguing about metrics for the sake of it. If we change the definition to be x% reduction in the rate of accidents relative to the expected rate

That seems like arguing about metrics for the sake of it. If we change the definition to be x% reduction in the rate of accidents relative to the expected rate if a non autonomous car would be use otherwise... you would get a similar rate of impact, normalized to a smaller population, limited primarily by rapidly diminishing economic barriers that don’t seem especially relevant to the argument at hanf