Ok but don't you have to weigh that against the potential of avoiding accidents by reacting quicker then you as a human can?

I have defiantly seen cases where FSD Beta stopped and the driver didn't understand instantly understand why.

Or simpler cases where the car follows a lane while the driver wasn't paying attention (adjusting the radio or whatever). Those can easily lead to an accident but are unlikely to if you are in FSD.

How do you make that calculation?

The calculation is easy, FSD does the wrong thing so often that there's no question a human driver is safer.

The question is not if FSD makes mistakes, its if FSD+a human driver monitoring is worse then a human driver.

All the available scientific evidence suggests these systems diminish driver focus and attention which definitely makes them less safe since these systems can't be operated safely without an attentive driver.

I have seen some evidence bases on some limited lane keeping systems that suggest this. But not on systems that are as advanced as FSD Beta for inner-city with the same amount of forcing the drives attention systems.