I'm trying to figure out a way of saying this in the nicest way without sounding like a jerk, but I really think you need to not be close minded and do some research and see what the system is capable of. The drives do not need to be in specific locations or in ideal conditions. It's capable of making on-the-fly decisions in complex scenarios, even if it's not a situation the model has been explicitly trained for. In many, many cases it can see, react, and decide scenarios significantly further and faster than a human.
Again, there's a ton of work to be done, but this tech exists today and is improving rapidly.
I am fully aware of what the system is capable of, and it's an error to say it's "making decisions." It's incapable of making decisions, which is a huge problem when the end goal is autonomous driving. There are too many variables that require real-time decision making that this AI model will ALWAYS be unable to perform.