Taking taxi is a 1000-times-per-year with low risk. Having a surgery is 1 per year with very high risk. Very different mental model here.

That calculus has a high dependency on skill of the driver. In the situation of an unskilled driver or surgeon you would worry either way.

The frequencies are also highly dependent on the subject. Some people never ride in a taxi but once a year. Some people require many surgeries a year. The frequency of the use is irrelevant.

The frequency of the procedure is the key and it’s based on the entity doing the procedure not the recipient. Waymo in effect has a single entity learning from all the drives it does. Likewise a reinforcement trained AI surgeon would learn from all the surgeries it’s trained with.

I think what you’re after here though is the consequence of any single mistake in the two procedures. Driving is actually fairly resilient. Waymo cars probably make lots of subtle errors. There are catastrophic errors of course but those can be classified and recovered from. If you’ve ridden in a Waymo you’ll notice it sometimes makes slightly jerky movements and hesitates and does things again etc. These are all errors and attempted recoveries.

In surgery small errors also happen (this is why you feel so much pain even from small procedures) but humans aren’t that resilient to the mistakes of errors and it’s hard to recover once one has been made. The consequences are high, margins of error are low, and the domain of actions and events really really high. Driving has a few possible actions all related to velocity in two dimensions. Surgery operates in three dimensions with a variety of actions and a complex space of events and eventualities. Even human anatomy is highly variable.

But I would also expect a robotic AI surgeon to undergo extreme QA beyond an autonomous vehicle. The regulatory barriers are extremely high. If one were made available commercially, I would absolutely trust it because I know it has been proven to out perform a surgeon alone. I would also expect it’s being supervised at all times by a skilled surgeon until the error rates are better than a supervised machine (note that human supervision can add its own errors).