Second, though, I think commuting will be changed. The hard part of carpooling right now is finding fellow passengers. With rare exceptions, it's not practical to round up a new carpool every day, so you need to find one or two people who (a) live near you, (b) work near you, (c) all work regular hours, and (d) all work the same regular hours. That's pretty hard.
Once they reach critical mass, fleets of driverless cars completely transform this. When you need a car, you click a smartphone app that immediately starts searching a central database for matches. As long as there are lots of people looking for rides—and drive time is precisely when lots of people are looking for rides—you have a pretty good chance of finding a match anytime you look for one. What's more, because the car is driverless, it has more flexibility: a human would want everyone to have destinations really close to each other, because the driver doesn't want to spend tons of extra time dropping everyone off. A driverless car doesn't care. If it has to drive a few extra miles, it's no big deal.
This is obviously better for the driver, since she can now read the paper or play Angry Birds instead of driving. It's also better for the passengers, who don't have to worry about being precisely on time every day and also don't have to worry about whether the other passengers are precisely on time. If you're running a little late, no big deal. If you work flex time, no big deal. If you have a doctor's appointment and need to leave for work an hour later than usual, no big deal.
I think this is a stretch. While I'm willing to grant that algorithms can be gotten to the point where they can drive, the problem of repeatedly picking several strangers to share a small space before anyone has had their coffee, and to do it with a very low risk of an irritating social interaction, strikes me as extremely hard. And if the risk of an unpleasant carpool partner is more than trivial then we'll all keep taking our own cars, thank you.
This is not to say no-one is willing to sit next to strangers - many of us do it on planes, trains, and buses all the time. But I think it's worse in a small group, versus a crowd, and my bet is few people would do it if they have a ready alternative they are used to (their own car). It's noticeable on a train with open seating that people will invariably pick an empty seat group if one is available, and only start doubling and tripling up once they have no option. In fact, it's something of a breach of etiquette not to do this.
Again, I think it's very important to understand that the average American new car buyer is not stretching to buy basic transportation. The average new car in the US costs $30k, but basic cars can be manufactured for as little as $3k-$5k. The difference is going for some combination of comfort, convenience, safety, and status. Ie most of the cost of the car is for those things. If your driverless car service reduces those factors, it will not sell to the mainstream of the US new car market.
My commenters to my last post made several excellent points worth up-leveling. Commenter sunbeam argued that services like Zipcar would be made more convenient by having the car come pick you up at your house, and would allow you to use a different car than you normally own (eg a pickup). This is undoubtedly true. I can imagine this could help the growth and profitability of Zipcar-like services. Still, Zipcar is a company with a few hundred million in revenue (growing at about 15%) in a several hundred billion dollar car market, so it's going to be a long time before they are anywhere close to transforming the situation.
Commenter Greg makes the excellent point that self-driving cars will allow parents to stop having to chauffeur kids to endless lessons, sports games, etc. This might actually cause some people to live further out of town (and thus increase VMT). Clearly this won't be true in the early stages of self-driving, in which the system will be basically an autopilot but a driver will still be legally required to be at the wheel. But I imagine that restriction will be dropped as the technology matures, and then kids will indeed be able to go anywhere without the parent dropping them off. This will indeed allow families to spread out, potentially. I would also argue that the modern over-scheduling of kids is basically the middle-class solution to the problem of kid demands for addictive TV/video-games. Rather than endlessly saying "NO", it's easier just to have the kids scheduled in constant improving activities. Self-driving cars will further enable this. They will solve the babysitting problem too since the kids can be driven off to some class/game or other on a Friday night!
Finally, on an aside, I was interested to read Kevin's car owning history. I also am a former Honda Prelude owner (2000-2010), and bought mine in favor of a Z3 because, while I could have afforded the Z3 at the time, I just couldn't justify to myself that I was getting an extra $15k worth of value. The Honda handled beautifully and was plenty fast enough. At the time I bought it, I was regularly driving back and forth between Arcata, CA and San Francisco for work - a four-five hour drive on gorgeous empty Northern California roads so I really got to enjoy its driving qualities. I loved that car and my kids loved it even more - they still haven't forgiven me for selling it. Now I drive a Jetta-TDI Sportwagen which uses about half as much fuel as the Prelude, is more practical for bringing things back from the hardware store for honey-do projects, and is still pretty fun to drive with good handling and the low-end torque of the diesel. Again, I looked at Audi TDIs at the time I bought the Jetta and just felt the value wasn't there - I was basically paying an additional $10k for leather seats and to show off an Audi badge. My next car will be an electric or plugin-hybrid.