I have heard a few discussions recently about the problem of how self-driving cars will make ethical decisions. I am completely unworried about this.
There are certainly things that we should worry about with self-driving cars. Chief amongst them are software bugs; there's an enormous scope for unexpected consequences in a piece of software controlling a vehicle at high speed in an open environment. Not far behind is security; who can control the car you are driving in (and the cars around it) and under what circumstances? And the question of how machines and software make ethical decisions is one we will need to grapple with as they become more powerful in what they control. I suspect we will find that it is not the machines they control that will be most important; it will be the social decisions. But cars, I am not so worried about.
For the rest of this discussion I will assume that self-driving cars (SDCs) are not subject to software bugs, and that they do pretty much what they were intended to do. That may not be true, but if so then it is bugs and not ethical decisions that are the problem.
In my lifetime I have been involved in four car accidents. All of them were low-speed collisions, caused by some clown not paying enough attention to what he was doing (yes, always he). In only one case was I the clown, and in my defence I was learning to parallel park at the time. Almost all of the car accidents I see the results of are clearly rear-enders in stop-start traffic; and many of the remainder are the result of someone going through an intersection when they don't have the right of way. SDCs will not suffer from attention lapses; continuous attention is something computers do well.
SDCs will not get drunk. They will not show off to their mates and/or girlfriend. They will not turn to yell at the kids in the back seat, or take their eyes off the road to text. They can be programmed not to speed, or drag race. They will not fall asleep. They will notice that the brakes are squishy, or that one of the wheels is rattling alarmingly.
The vast majority of accidents - no matter what the cause - do not involve ethical decisions of any kind. Even the ones that SDCs cannot eliminate (brake failures; bad decisions by other cars; rocks falling from cliffsides) will usually leave the car trying to avoid something either by stopping short of it or dodging into open space. The SDC is probably better at that than you. It does not panic and forget the proper braking technique; it does not hit the accelerator by mistake. It uses engine braking to its utmost, and shortens its stopping path as much as possible. When it cannot avoid hitting something, it can make sure the impact point is optimal for the airbag systems. It may even trigger airbags in advance of the impact, or tighten the seatbelts, or deploy all sorts of collision mitigation that depends on advance notice that we don't have yet and reflexes that only machines can achieve.
I expect that SDCs will quickly become safer than human driving for most people, even if you assume that the software is programmed to sacrifice the driver whenever there's an ethical decision to be made. And SDCs will keep getting better, where humans have to be retrained every generation.
But even in those limited cases where an ethical decision is involved, the SDC will quickly do a better job than a human. It will react faster to the situation; it will see the kid who runs out on the road quicker, and brake faster. It will be more likely to see the third option of not hitting anything, and it will be more accurate in taking it. Adrenaline is useful for many things, but precision maneuvering is not one of them. An SDC will be quicker to realize that there is an ethical decision to make, and it will be more capable of thinking through the possibilities in time to act on them. If nothing else, it will be able to make a definite decision and sacrifice one of the people involved, rather than acting too late and killing both of them. Even if it does the machine equivalent of flipping a coin, your chances of surviving such a crash are likely to be better than if you drove and chose to save yourself.
Greater capability means the ability to avoid ethical decisions; not by making someone else make them, but by finding a course of action that avoids harm to anyone. Humans have been using technology to get themselves better outcomes for as long as we've been human; and that has often meant no longer having to choose who lived and who died. Self driving cars are just another step along the (literal) road. I look forward to it.
There are certainly things that we should worry about with self-driving cars. Chief amongst them are software bugs; there's an enormous scope for unexpected consequences in a piece of software controlling a vehicle at high speed in an open environment. Not far behind is security; who can control the car you are driving in (and the cars around it) and under what circumstances? And the question of how machines and software make ethical decisions is one we will need to grapple with as they become more powerful in what they control. I suspect we will find that it is not the machines they control that will be most important; it will be the social decisions. But cars, I am not so worried about.
For the rest of this discussion I will assume that self-driving cars (SDCs) are not subject to software bugs, and that they do pretty much what they were intended to do. That may not be true, but if so then it is bugs and not ethical decisions that are the problem.
In my lifetime I have been involved in four car accidents. All of them were low-speed collisions, caused by some clown not paying enough attention to what he was doing (yes, always he). In only one case was I the clown, and in my defence I was learning to parallel park at the time. Almost all of the car accidents I see the results of are clearly rear-enders in stop-start traffic; and many of the remainder are the result of someone going through an intersection when they don't have the right of way. SDCs will not suffer from attention lapses; continuous attention is something computers do well.
SDCs will not get drunk. They will not show off to their mates and/or girlfriend. They will not turn to yell at the kids in the back seat, or take their eyes off the road to text. They can be programmed not to speed, or drag race. They will not fall asleep. They will notice that the brakes are squishy, or that one of the wheels is rattling alarmingly.
The vast majority of accidents - no matter what the cause - do not involve ethical decisions of any kind. Even the ones that SDCs cannot eliminate (brake failures; bad decisions by other cars; rocks falling from cliffsides) will usually leave the car trying to avoid something either by stopping short of it or dodging into open space. The SDC is probably better at that than you. It does not panic and forget the proper braking technique; it does not hit the accelerator by mistake. It uses engine braking to its utmost, and shortens its stopping path as much as possible. When it cannot avoid hitting something, it can make sure the impact point is optimal for the airbag systems. It may even trigger airbags in advance of the impact, or tighten the seatbelts, or deploy all sorts of collision mitigation that depends on advance notice that we don't have yet and reflexes that only machines can achieve.
I expect that SDCs will quickly become safer than human driving for most people, even if you assume that the software is programmed to sacrifice the driver whenever there's an ethical decision to be made. And SDCs will keep getting better, where humans have to be retrained every generation.
But even in those limited cases where an ethical decision is involved, the SDC will quickly do a better job than a human. It will react faster to the situation; it will see the kid who runs out on the road quicker, and brake faster. It will be more likely to see the third option of not hitting anything, and it will be more accurate in taking it. Adrenaline is useful for many things, but precision maneuvering is not one of them. An SDC will be quicker to realize that there is an ethical decision to make, and it will be more capable of thinking through the possibilities in time to act on them. If nothing else, it will be able to make a definite decision and sacrifice one of the people involved, rather than acting too late and killing both of them. Even if it does the machine equivalent of flipping a coin, your chances of surviving such a crash are likely to be better than if you drove and chose to save yourself.
Greater capability means the ability to avoid ethical decisions; not by making someone else make them, but by finding a course of action that avoids harm to anyone. Humans have been using technology to get themselves better outcomes for as long as we've been human; and that has often meant no longer having to choose who lived and who died. Self driving cars are just another step along the (literal) road. I look forward to it.
No comments:
Post a Comment