Self-Driving Vehicles: Programming vs. Training
Self-driving, driverless or autonomous—whatever you call it, hands-free driving, which was only recently a science fiction fantasy, has arrived. While we may be able to program cars to drive, however, can we train them to drive like humans?
While Google’s self-driving car is arguably the most well-known prototype, car manufacturers, such as BMW, Mercedes-Benz and Audi, are already rushing to get in on the emerging driveless car market. Simultaneous to the commercial development of driveless cars, military engineers also continue to explore the potential of driverless vehicles. Whether the goal is to free up driver’s hands to eat sushi or text friends while on the road or to take soldiers further away from the frontline action, however, the long-term consequences are significant. The arrival of driverless vehicles marks a new era in automation—an era when a growing number of tasks, which once required the responsiveness of humans, will no longer require human operators at all.
It goes without saying that some industries will be more impacted by the dawn of driveless vehicles than other industries. If the technology develops to the point where it represents a viable alternative to human-operated vehicles, gains widespread legalization, and finds an enthusiastic market, we can expect to see major changes in the transportation industry and in all industries that rely on moving goods from place to place. After all, if a driverless truck can deliver food from Florida to New York without stopping for a sleep break or to fuel up on coffee several times, why not go driverless?
In many respects, the driverless truck would come with distinct advantages. First, moving goods will cost less (after all, there will no longer be any need to pay a driver). Second, moving goods will take less time, since rest breaks will be eliminated. Finally, fewer variables will disrupt shipments (e.g., speeding tickets, traffic violations and weigh station violations will presumably be eliminated since the driverless vehicles will naturally be programmed to follow all the rules). In short, by eliminating human needs (e.g., sleep, water, food and companionship) and human errors (e.g., speeding), we will theoretically be able to ship goods for less money and more quickly.
While we may eventually be able to program cars, trucks, buses and other vehicles to operate without human operators, however, there are at least a few scenarios in which programmed vehicles will likely never be able to respond on par with well trained human vehicle operators. To appreciate why, consider the following scenarios.
: There are two trucks driving along an isolated stretch of the I87 north of Albany, New York. Both trucks pass a car with its flashing hazard lights on parked along the shoulder of the highway. Do they have the same response? Likely not. While the driveless vehicle may swerve slightly to avoid hitting the car parked along the shoulder of the highway, the human operated vehicle is far more likely to stop to offer assistance. Why? To the driverless vehicle, the car on the shoulder of this desolate stretch of highway is just an obstacle of avoid hitting. To the driver in the human-operated vehicle, the same car represents a stranded person or group of people in need of assistance.
: A driverless vehicle is about to drive through an intersection at a green light when a mother with a stroller enters the intersection. If the vehicle swerves, it will hit a large dump truck but if it doesn’t swerve, it will hit the mother and child. What will the driverless vehicle do? A human driver would naturally swerve away from the mother and child, but would a driverless vehicle do the same thing? It depends on whether or not the vehicle has been programmed to hit larger rather than smaller objects when faced with the choice of crashing into two different things. What can’t be programmed, of course, is how to make the most ethical decision in such scenarios.
This, of course, is where the difference between programming and training becomes visible. As more human tasks become automated, it is important to bear in mind that while machines can be programmed, they can’t necessarily be trained. Training, after all, is not the same thing as programming, since training necessarily entails both thinking and feeling. After all, while a car and human may both be equally able to be drive (if not now, in the near future), only human drivers can make ethical decisions while on the road. As more and more human tasks become automated, then, training will not be replaced by programming but it likely will change. Among other shifts, we can expect workplace training to become increasingly focused on affective and ethical issues as humans are called upon to step in precisely where machines fail.