This is getting into traffic research territory - i.e. given the entire dataset (current and historic) can you shorten average trip time vs individual cars just doing whats best for them. Either way, having a real time data infrastructure seems only positive.

Traffic lights are already remotely controlled by city planning departments, it's the software that's missing. My whole point is that driverless cars, where the city enforces data sharing and maybe even some routing, can have a real impact on throughput times.

The best part is that most of this is just software. And not ~$2B to make a 1.7 mile subway line or 10+years to make a bus lane (see SF)