Cars still don't drive themselves, and very few cars percentage-wise even have collision avoidance, which is the only sort of safety feature that can possibly significantly reduce the rate of collisions among people who truly aren't paying attention to the road in a dangerous way.
The reason cell phone use doesn't cause the huge number of crashes that were predicted is much more obvious: 99% of the time, a driver doesn't actually need to do anything, even without self-driving cars, because most of the time, the road is straight, and there isn't something in front of you. And unlike drunk driving, people don't typically use a phone continuously, so the probability of overlap is small. That makes the risk of cell phone use inherently very low even if it is technically a risky behavior. And if we assume that most people choose when to use the cell phone based on the conditions around them—at traffic lights, on long straight stretches with no visible cars at upcoming intersections, etc.—then the risk logically should drop to remarkably close to zero among otherwise good drivers. Any statistics that appear to show otherwise are highly suspect.