I generally agree with the view that, sometime in the next 5-15 years, on-demand self-driving cars are going to change everything. So I don't have much to add to the conversation, except this prediction: at some point (soon), getting a ride will be the real-world equivalent of using google for search.
Here's my rationale, which is based on the parallels between a search and a ride. The marginal cost of providing search is negligible. Google (and Microsoft, and specialized search engines like Yelp) are able to monetize search in other ways -- primarily by selling ads against search results, and by using the information derived from searches in other ways. The same pattern goes for most of the other "free" services we get online: a product is provided for free to an end user, and then the provider advertises against it, or finds other ways to monetize it (eg, "freemium").
Providing a ride will (eventually) follow the same dynamic.
The first key trend is that the marginal cost of a ride is dropping, dramatically: per Evans' back-of-the-envelope math, above, an electric self-driving car today costs roughly between $.15 and $.20 per mile. (See Note 1, below.) As solar energy gets cheaper and cars get more efficient, we can expect the cost of driverless car trips to keep dropping .
The second key trend is that there are fantastic opportunities for driverless car networks to monetize the rides that they provide, including: (a) in-car advertisements; (b) in-car entertainment; (c) "premium" services like "upgrades" to bigger or fancier cars, or cars with, say, booze or food; (d) "finders' fees" for directing riders to certain stores / restaurants / entertainment venues; (e) delivery or logistics services; (f) metadata regarding who is going where, when, with who else; and so forth. Rides are an even richer opportunity for monetization than searches or email ever were.
At some point, the two trends are going to intersect: the opportunity to monetize a ride is going to exceed the cost of providing that ride, and driverless car networks will offer rides for free. It's going to happen, it's just a matter of when. Just like Google's search engine is the default starting-point for navigating the internet, pulling out your phone and requesting a free ride from a driverless car network (operated by, say, Google or Tesla or Uber) will be the default starting point for navigating the physical world. Incidentally, this would be a great service for a small- or medium-sized city that lacks a good public-transit infrastructure to offer, but that's a separate post.
I've thought a lot about the privacy implications of this; at first, it seemed pretty frightening. But then it occurred to me that our cell phone providers are already tracking us basically 24-7. It's not much different to have a Google car drive me to the porn store (or ACLU meeting) than to just bring my cell phone with me while I drive myself. I do think it will present an opportunity for a driving car network-equivalent of DuckDuckGo (the search engine that doesn't track its users).
More concerning, I think, are the questions about how law enforcement will interact with self-driving cars. If there's a warrant for your arrest, and the police know you're in a self-driving Uber, can the police order that Uber to take a detour to the police station?