By Dr. Lance B. Eliot, the AI Insider for AI Trends and regular contributor.
One of the most commonly repeated claims for the advent of self-driving cars is the notion that we will be able to eliminate car related fatalities. An estimated 30,000 to 40,000 fatalities occur due to car incidents each year in the United States alone. Wouldn’t it be wonderful if we could just whisk away those potential future deaths via the miracle of self-driving cars? Presumably, self-driving cars won’t get drunk, they won’t fall asleep at the wheel, and otherwise won’t be subject to the same foibles as human drivers. Indeed, some of the major car makers are saying that with self-driving cars we will have zero fatalities. I say bunk. There is a zero chance that we’ll have zero fatalities due to self-driving cars.
My statement of there being a zero chance might be shocking to some of you. It certainly would be a shock to most of the major media outlets. They have bought into the zero fatalities moniker on a hook, line, and sinker basis. Regulators love the idea too. Self-driving car makers love the idea. Anyone that cares about the lives of people loves the idea. It just sounds catchy and something we all would welcome to have occur. Unfortunately, it is unrealistic and belies the facts.
Let’s take a closer look at the fatalities topic. According to the U.S. Department of Transportation (DOT), there were 35,092 fatalities in 2015 due to vehicle related incidents in the United States (DOT census numbers were released in November 2016 and represent the latest counts and statistics available on this topic; we’ll need to wait until November 2017 to see the 2016 numbers). They also estimate that it cost about $242 billion dollars for the aftermath recovery of the incidents and fatalities. Obviously, the toll on human lives is huge and so is the monetary cost. Some have argued that if we stopped driving altogether, we could save those 35,000 lives in the United States annually, plus many more globally. They tend to say that we should forget about the use of self-driving cars and just stopping driving cars at all. I’m not going to go down that path here, but you are welcome to look it up and see their position on this.
As a relative comparison, heart disease is the top killer in the United States and amounts to about 614,000 deaths annually. So, the number of car fatalities is relatively small in comparison, and in fact if you add-up the Top 10 means of death in the U.S, the number of car related fatalities amounts to approximately just 1% of that count (it’s not even in the Top 10 list). That being said, I want to emphasize that any deaths due to car fatalities is way too much. Anyone that has experienced a friend or family member killed in a car fatality knows the pain and agony associated with car fatalities.
The number of crashes involving those 35,092 fatalities was 32,166. Thus, there were about 1.1 fatalities per crashes involving fatalities. In other words, it tended towards one death, rather than say two or more deaths, on the average overall. The number of motor vehicles involved was 48,923. Thus, there were about 1.5 vehicles involved per crash. This generally seems to make sense, since we would have expected that the fatalities would tend to occur when two or more cars crash together. This is not the only way to have a fatality as it could also be that a car swerves off the road and crashes into a wall, killing someone as a result of the incident and not involving a crash into another vehicle.
We next get into even more interesting stats on the car related fatalities. Consider that a fatality could be the driver of the car, or maybe an occupant inside the car, or perhaps a pedestrian, or a motorcycle rider, or even a bicyclist. Can you guess what percentages each of those circumstances might be? Here’s your answer. About two-thirds or roughly 66% were occupants (which includes the driver), while the remaining one-third consisted of pedestrians (16%), motorcyclists (14%), bicyclists (2%), and large truck occupants (2%).
Being a pedestrian is a dicey thing, when it comes to car fatalities, as the number is high enough to realize the potential for getting killed while not actually being inside a car. What we don’t know is whether the pedestrians were killed because the driver was essentially at fault, or whether the pedestrian was at fault. In other words, if a pedestrian suddenly darted into the street and there was no reasonable way for the car driver to avoid hitting and killing the pedestrian, this kind of fatality is not particularly attributable to the car and more so to the pedestrian.
Of the occupants killed in the car related fatalities, nearly 52% of the drivers were not wearing their seat belts, while 57% of the passengers were not wearing their seat belts. We don’t know how many might have lived had they been wearing their seat belts, but it is generally believed that many, if not even most, would likely have survived the crash. Why is this important? Well, rather than looking toward self-driving cars as a savior to reduce car related deaths, just imagine if we simply got more drivers and occupants to wear their seat belts that we could dramatically likely cut down on the number of car related fatalities immensely.
This is important for another reason too. Let’s suppose we do have self-driving cars. The passengers inside the self-driving car should be wearing seat belts as a safety precaution, in case the self-driving car gets into a crash. But, I am willing to bet that people will become complacent and not want to wear their seat belts while in a self-driving car. They will act like they are in a limo or a bus that traditionally you don’t wear a seat belt as a passenger. People will tend to trust the self-driving car, over time, and opt to not wear their seat belts. As such, I am predicting that we might actually have a rise of a per capita deaths per self-driving car crash in comparison to non-self-driving car crashes, simply due to people being less likely to wear a seat belt in a self-driving car.
Now, some will say that self-driving cars aren’t going to crash. Somehow, magically, the AI in these self-driving cars will prevent the cars from crashing. Really? Let’s unpack that logic. If a pedestrian runs into the street and directly in front of a self-driving car, and if there was no practical way for the self-driving car to see or know that the pedestrian was darting into the street, the self-driving car is going to potentially kill that pedestrian (in one of my other columns, I discuss the ethics of self-driving cars and point out that the self-driving car will need to make a choice between harming the car occupants or the pedestrian).
My point is that no matter how good self-driving cars are, you are still going to have circumstances of pedestrians getting killed by accidentally or foolishly getting into the path of the car and when the car itself has no other way to proceed other than killing that person. The same is true of a bicycle rider that swerves in front of a self-driving car. Likewise, a motorcycle rider that goes into the path of the self-driving car. These are plain physics. The self-driving car is not going to magically leap into the air or go into instant reverse. Fatalities are going to happen.
Zero fatalities is zero chance.
Some say that we should focus our attention on engineering and roadway measures, which would separate pedestrians away from cars, any kind of cars, self-driving or human driven. The use of well-designed sidewalks and barriers can likely reduce the pedestrian deaths as much as can the use of self-driving cars.
Another claim about how wonderful self-driving cars will be about reducing fatalities is that self-driving cars don’t get drunk. Well, of the 32,166 crashes, there were an estimated 4,946 drunk drivers that were killed. We don’t know how many drivers overall were drunk during those crashes, and just know how many of the drivers that were killed had been drunk at the time of the crash. It is predicted that perhaps 6,973 of the deaths could have been avoided if all drivers that were drunk were kept off the roads. Thus, maybe about one-fifth or 20% of the car related fatalities were due to drunk drivers. That’s a significant amount, but much less than what is implied by the news media. Most of the general news media seem to think that if we had non-drunk self-driving cars that we’d be down to maybe a handful of fatalities, but as you can see, we’d still have 28,119 deaths. That’s a lot.
Of the car related fatalities, about 20% of the deaths are due to the vehicle going off the road and hitting an object, such as a tree, a telephone pole, or other traffic barriers. Why are cars swerving off the road? Could be due to being drunk, could be due to fatigue, could be due to inattention to the task of driving. Some say that we could reduce those deaths by being more careful about putting hardened objects near the roadway. A human driven car would presumably survive if there weren’t objects to be hit that would make a crash fatal. One can argue about this, and though some say we should clear the area around roadways or put breakaway objects in their place, it obviously is a rather large logistic problem to somehow make sure that roadways are designed in this manner. But, it is a factor worth considering.
It generally makes sense to believe that a self-driving car is not going to fall asleep or get drunk in any human-like way, but we also should not assume that the AI is perfect and at all times perfect. We all have experienced computer systems that have bugs in them, or where we have hardware fail. Self-driving cars will be no exception. You could be in your self-driving car and suddenly the tires blow, and no matter how good the AI is, the car might go off the road and hit a telephone pole. Or, the AI might encounter a “bug” in the software that causes the car to swerve into a truck next to the car (you might want to read my column about the Tesla car that did just that, though the claim is that the software worked as intended and was not a bug per se). Or, the sensors on the self-driving car might suddenly stop working, leaving the self-driving car “blind” to the roadway ahead and it might plow into another car without its sensors being active.
As you can see, there are lots of opportunities for a self-driving car to kill its occupants, or kill pedestrians, or kill bicyclists, or kill motorcyclists. It is going to happen.
We are also somewhat assuming in this false belief about the perfection of the self-driving cars that all cars on the road will be self-driving cars. That’s not going to happen for a very long time. We will gradually see self-driving cars emerge. The millions upon millions of human driven cars will exist for years and years. We cannot overnight economically swap out all human driven cars for self-driving cars. As such, you can expect that self-driving cars will be interacting with human driven cars. That interaction is definitely going to produce fatalities.
What then will occur with fatalities and self-driving cars? In some ways, yes, self-driving cars will reduce the fatalities. But, as discussed, in other ways it might keep those fatalities going, and even increase some classes of fatalities. One thing we can say for sure, we aren’t looking at zero fatalities simply due to the introduction of self-driving cars. Anyone that says that is living in some kind of science fiction novel. By the time the world gets toward a future of all self-driving cars and relatively perfected AI, we will probably be using our jet packs and maybe even doing Star Trek like beaming, so we might not have conventional car fatalities anymore, and instead have fatality stats on jet pack crashes and beaming problems. Zero fatalities, never. Sorry to break the news to you.
This content is original to AI Trends.