It's probably a bit misguided in the sense that, if the end-result further down the line is the "more machines than humans driving in the streets" (let alone zero humans), then it might have made more sense to iteratively change the infrastructure instead, rather than have machines try to work under infrastructure that is designed to be good for humans but bad for machines, and then try to "iteratively improve" under conditions that carry significant risk of injury.
Brings Ford's quote on "faster horses" to mind. Imagine if we still required roads to be horse friendly, and expected our cars to go through mud and jump fences.
I agree with GP: automating everything and removing humans from the equation would be a much better solution. Unfortunately it's not practical because it's not profitable.
Getting humans to agree on something at scale larger than half a dozen is like herding cats. Money is catnip - it works well at aligning masses towards common goal.
Except iteratively changing the infrastructure requires political capital along with actual capital in the form of tax dollars. Self-driving cars can be built outside of existing infrastructure and paid for entirely with private funding.
Changing infrastructure is a nice idea in theory, but the practicality of getting anything done in SF means that if you want progress, you should look to other options.
There is nothing misguided in switching out a human for a machine that can do a job better, while keeping the job unchanged. Iterative improvement is the only thing anything ever gets done, and yes, that means we might first need to get drivers out of the streets before we're able to make more substantial redesign of the transportation fabric.