[quatrevingts]

A 3 mile cab fare in Berkeley is $20+. The all-in cost of driving a basic private car is around 55 cents a mile (or 90 cents for a BMW). That only leaves license cost, so unless you need a $500,000 medallion, the rest goes to the driver. (And as long as the driver works for himself, the medallion is a capital investment which he can sell later.)

[brc]

But the cost of taxi licences is significant in many cities. I doubt any local/state/federal government is going to let the possibility of driverless taxis proliferate without similar licensing schemes.

I agree that having a car running privately is a lower per-mile cost, but that discounts the number of 'dead' miles that a for-hire car drives - the private car always takes the person where they want to go, while ordering one up via the internet is going to at a minimum double the per-mile cost, because only half the miles (as an average) will be utilised. The car will also not always be in use because demand will never be that smooth, so the in-demand times have to be priced higher to make up for the quiet times. Then there has to be a higher insurance cost, plus a margin to cater for damage to the cab, booking systems, marketing, etc etc.

It's not that hard to see where a cost goes from the 50c/mile to the ~$4/mile which isn't far under the taxi cost.

Taxi drivers make just over minimum wage in most places. While the taxi (plate) owner does make a decent return if run well, it's not like there are a pile of taxi billionaires around. I'm just not convinced removing the driver is going to change the cost by an order of magnitude.

[quatrevingts]

Well, like you said, the fare has to make up for the time the driver is idle, so that problem goes away. Licensing is done to ensure passenger safety, and sometimes to restrict supply. The computer is not going to drive drunk or kidnap you, and excess unproductive machinery can just drive itself to a city where there is more demand.

And the dead miles problem is not a big factor in places like Manhattan where utilization of taxis is high. If this system were pervasive, you could simply maintain a car or two per block.

The biggest issue would be dealing with the peak demand. Fortunately, heavy rail is complementary here, and it's only marginally less convenient when you have quick on-demand service to and from the train station. Demand-based taxi pricing would sort things out.