A lot of people take it for granted that cars will drive themselves in the fairly near future.
And they may be right. Five years ago, I would have predicted much less advanced cars than the present state of the art is. Let me eat some preemptive crow: my skepticism could well be unwarranted.
But I do remain skeptical, on both technical and legal grounds.
The technology first: we’ve all heard that they’ve logged hundreds of thousands of miles, mostly without incident. But the devil is in the details.
First, while we know that the driverless cars have done some challenging driving from time to time, we don’t know how much of the hundreds of thousands of miles has been on empty streets, in well-lit, dry circumstances. We don’t know how often the test drivers have snatched back control from the car before entering a dangerous situation. And we shouldn’t lose sight of the fact that 300,000 miles without an accident is a reasonably impressive record, but by no means inhuman (if the average driver clocks up 12,000 miles a year, 300,000 miles is the equivalent of 25 years of experience without a crash — certainly a record many people can claim).
Second, we don’t know how many types of driving have been simply avoided by the projects so far. Do they even drive the cars in heavy rain or snow? Over icy bridges? Through crowded parking lots?
Third, even if the technology is excellent, there are some gaps in how the technology interacts with humans that remain questionable. When you cross a street on foot, you likely seek eye-contact with a driver, knowing that if you catch their gaze, they have seen you and will not hit you. When you’re a driver, you can read a pedestrian’s body language on the curb, and reliably guess whether they are waiting to cross the street or are just standing near the edge of the sidewalk, chatting. The best collision-avoidance system possible doesn’t bridge this gap.
But these problems seem surmountable. It would not surprise me at all if Google could produce an autonomous vehicle that is at least as safe as a human driver in the timeframe of five to ten years. (It also would not surprise me if they couldn’t).
The legal objections are much more problematic.
Let’s make a few assumptions:
- Autonomous vehicles won’t be perfect. Nothing we’ve made ever has been, and I don’t see that turning around soon. Sooner or later, a computer-controlled car will make a bad decision and injure or kill someone. In fact, it will probably happen pretty frequently: north of 30,000 people are killed, and millions are injured, every year. Even if automatic cars drop the incidence of death and injury by a factor of ten, that translates to thousands of deaths and hundreds of thousands of injuries.
- The artificial intelligence in any near-future autonomous vehicle will be different in kind from humans. We don’t know how to make human-like intelligence. It follows that the kinds of mistakes that the AIs make will be different than the kinds of mistakes that humans make.
- You can’t get around the problems of AIs by requiring a human behind a wheel. On a practical level, it just won’t work. People are not going to remain alert and in-control if for 99.99% of all trips, they don’t need to do anything.
So what happens when the grieving family of a victim of computer error takes the makers of the system to court?
Can you look someone in the eye and say, “Yes, it’s true that your child was killed by this vehicle. Yes, it’s true that if there had been a human behind the wheel, your child would most likely still have been alive. Yes, there is a real deficiency in the behavior of the artificial intelligence in question that could at least theoretically have been caught and fixed by more stringent standards. But we are none the less going to immunize Google from your wrongful death suit, because overall their cars are safer than human drivers, even if in the particular case that you’re interested in, they were less safe”?
Because you have to immunize the companies for this to hit market. Nobody, not even really big, wealthy companies like Google, can deal with thousands of wrongful death suits and hundreds of thousands of injury suits, per year. (Or even more: if autonomous cars are as awesome as we all hope they are, you could see people spending a lot more time on the road, increasing the number of accidents).
And should you? It’s troubling to me to say, “We’ll shield a corporation from liability for killing someone, because we think that the industry they’re in overall reduces deaths.” What if Schmoogle, Macrohard, and Pear all make car AIs, and Schmoogle’s is twice as unsafe as Pear’s and Macrohard’s? Even if Schmoogle’s software is overall safer than humans, it doesn’t sound like a good idea to say that they should bear no responsibility for deaths caused by their software when it seems like they ought to be able to do better.
If you have faith that we’ll thread this particularly thorny legal question with the kind of sensible, even-handed regulation that allows driverless cars while not creating the worst kind of perverse incentives, well… I’d suggest that you haven’t been paying a lot of attention.
Driverless cars are coming some day. But don’t be surprised if you can’t buy one twenty years from now.