If we focus exclusively on the risks we are most concerned about, we can miss the other risks that obtain should our demands be met.
Tesla provides an autopilot that allows its cars to drive themselves in certain circumstances. Recently, while on autopilot, a Tesla car crashed and killed the driver. The autopilot apparently did not distinguish between a white tractor-trailer and a brightly lit sky in the background of the trailer. The driver is rumored to have been looking at a movie, against the express mandate that a driver using the autopilot keep his hands on the steering wheel.
The National Highway Traffic Safety Administration (NHTSA) is investigating the crash, and the government is now considering how to regulate autopilots. I fear that if regulators aren’t careful, they may kill more people than they save. The basic problem is that first recognized by the great French economist Frederic Bastiat. People too often consider effects that can be seen, but not those that are invisible. Here the focus is likely to be on lives lost by the autopilot, often in fiery crashes that get attention. But lives may be saved as well by its introduction and these lives will receive almost no attention. Statically, the current autopilot itself may save some lives. Dynamically, permitting autopilots may lead to faster improvement in self-driving cars that may save more lives in the future. There are a lot of such lives to be saved. More than 30,000 people die each year in car crashes in the United States, and most such crashes are caused by driver error.
This problem is compounded by the prism though which government bureaucrats view regulation. Many politicians will blame them for accidents caused by autopilots that they did not prohibit. Many fewer will blame them for the lives lost by failure to innovate.
Moreover, the current rationale for regulating for safety is inapplicable to regulating autopilots. The best argument for safety regulation is that car buyers may not being willing to spend on safety features that raise the cost of the car, because they will not appreciate their virtues, particularly as almost everyone believes himself an above-average driver. But the issue here is not to mandate an autopilot, but to determine what kind of autopilots to permit. And even without government regulation, drivers are likely to make the sensible choices about what kind of autopilots to buy, because they bear the additional cost. And tort law will police unreasonably dangerous autopilots by holding companies liable. To be sure, the market backed by tort law may not make perfect decisions, but it seems likely to make better ones that the government, burdened both by bureaucratic (mis)incentives and by the general problem for regulators of evaluating unseen benefits.
I would make one exception to this hands-off attitude to regulation. The government should prevent autopilots from favoring the lives of those in the car as opposed to pedestrians or occupants of other cars. Car buyers may have perverse incentives to buy such autopilots even though they would not be in the public interest. But I would limit federal regulation so as to encourage swift innovation in the most promising development for motoring since Henry Ford invented the Model T.