J. and his wife will eat for free if the two couples are picked up at their houses and driven to the restaurant in a robot-driven car. But he has to pay if no such vehicle can be hired and/or it’s not legal in Santa Monica.
This is a great bet because the bettors get decades of fun out of it of arguing over who is going to win. Heck, I bring the bet up a couple of times per year in conversation and I’m not even involved.
With ten years to go, J. is quite confident. Google and a number of other companies have been making much-publicized progress toward self-driving robot automobiles. Tesla right now offers an autopilot package, albeit one that requires you to sit behind the wheel ready to take control at any instant.
The problem is it’ll give you back control when it encounters some problem, usually in an emergency, but even if there’s no real emergency, you’ll be totally caught with your pants down doing something else (or outright asleep) and all you’ll have is a fraction of a second to adjust because the car will be driving at 75mph. If you are in the middle of reading a book, not paying attention to the road, then how much time would you need to just make sure you now really have to drive the car (especially if the system was so good it only happened rarely), look around to see what needs to be done, and take action if needed, but avoid sudden action when no such action is needed? (Like suddenly and too forcefully turning the steering wheel or suddenly pressing the brakes with full force thus creating emergencies for other cars who will in turn stun their passengers into becoming drivers in a fraction of a second.)
It’s only useful if it never gives back control to you, which is only possible if the system running it is fully controlling the whole traffic on the highway, much the same way the whole railway traffic is controlled by the same system.
I suspect that with making robot cars safe going from 0% of the time to 90% of the time takes about as much work as going from 90% to 99%, which is as much work as going from 99% to 99.9% and so forth. (Warning: I just made up this particular learning curve. But it sounds pretty plausible.)
But there’s kind of a Zeno’s Arrow problem in this in that it may take a long time to get your self-driving car to the point where it would only kill, say, ten people a day if it were universal in America. That’s about an order of magnitude better than human driven vehicles at present, but it still sounds like a PR nightmare.
I suspect navigating through parking lots is a challenge for autonomous vehicles. Google use its mapping technology to understand roads, but parking lots are full of pedestrians wandering about. I know a supermarket parking lot is stressful for me as a human driver.
In a future, a lot of parking lots might be reconfigured to make robot car dropoffs at the door simpler and more routine, but right now they typically don’t work that way. Supermarket parking lots are typically not set up to deliver passengers to the door but are instead intended to give you a place to park somewhere short of the door.
Dropping your kids off and picking them up from school is another common trip that often gets complicated at the very end. It’s more like the kind of drop-off and pick-up service you’ll expect from a robot car than a trip to the supermarkets, but it must give Google people cold sweats about the possibility of a Google car running over a small child at a school.
A lot of schools have invested over the years in improving traffic flow, but that can take decades to get it right. For example, Campbell Hall, a private K-12 in North Hollywood, was notorious for how the dropoff and pickup flow through the middle of campus was a danger to the younger students who sometimes dashed in front of cars for reason of being six years old. Finally, the school raised a lot of money to reconfigure the entire campus to reduce the traffic stress. It appears to have been a success, but it took almost 70 years from the school’s opening in 1944 to the new arrangement in 2013.
Here’s a question: Have corporate jets’ autopilots improved to the point where most executives are willing to fly with just one pilot? Back in the 1980s and 1990s when I’d sometimes fly on the corporate plane, there were always a pilot and a copilot. That was expensive, especially considering that autopilot even then could get you most of the way. But the big guys seemed to like the idea that when it came to landing, there’d be a backup human available.
Cars, however, are easier to automate than planes in that if the driverless car senses, say, a child running out into the street after a ball, it can just slam on the brakes and the car won’t fall out of the sky. Two dimensions are simpler than three. There’s a reason that most American adults have driver’s licenses while very few have pilot’s licenses.