Most crashes aren't in difficult situations.

https://crashstats.nhtsa.dot.gov/Api/Public/ViewPublication/...

Current self driving stats in general are meaningless because all that driving is done under constant human supervision. But if the car drives itself for 20Km and then needs intervention to avoid a fatal crash it will still count as another 20Km of self driving with no incidents. The human saves don't make it into the publicized stats.

There are 2 stats that could prove meaningful for this: some average of distance driven between human interventions, and percentage of driving conditions covered. Then compare this to human stats, I'd assume the average human driver might go 10000Km without incident while covering 90% of driving conditions (like everything except extreme weather or road conditions). A self driving car would have to match that.

In other words AP only has 5 deaths because those were the only drivers who left it unsupervised. I'd guess that if every car in the world today would suddenly be driven by a computer we'd see worse numbers than with human drivers one way or another (efficiency, deaths, etc.). Self driving today is great as an assist, with the human still fully in control.

People need to lookup the definition of autopilot. It is not self driving.

Simpler option: don't call it autopilot if it is shown that it misleads people.

In my 2015 Model S it is called Autosteer in the settings, the manual, and in the instrument cluster when messages appear related to that function.