* If a driving algorithm is a little more accident-prone than the average human driver at a given speed, that deficiency could be rectified by forcing it to observe lower speed limits.
* On the other hand a driving algorithm that proves to be two orders of magnitude less accident-prone than the average human driver at a given speed, should be granted higher speed limits. (Not so much higher as to erase all or most of its safety advantage. But higher.)
Why would you assume faster is better? If the car drives itself, it has no need for a driver. Thus, it could be completely unattended, and take advantage of the fact that optimal fuel economy tends to occur at about 35-55 MPH, where wind resistance is too low to be problematic.
Thanks to exponential nature of inertia, doubling speed generally causes four times the wind resistance. It doesn't take long for that ratio to get stupid, and that's why we don't have planes that fly 5,000 MPH.
In general aviation, it's commonly understood that a more powerful engine will help you climb faster, but typically doesn't speed the plane up much except at the extreme lower end of the power/weight curve - that's mostly a function of wind resistance.