With fully automated driving the number of crashes (and therefore injuries & deaths) will go down -dramatically-. So you can argue all you want about the ethics & morality of the ~100 deaths/year or so from automated car crashes. But that's far preferable to the current situation. Even if we still have 10-20k deaths/year from automated driving systems, that's still a large improvement.
This argument about the ethics of contrived car accident scenarios totally misses the boat. In my opinion the only ethical argument is to move over to fully automated driving systems as soon as -safely- possible.
But I don't. I don't care about 30k deaths a year if I'm killed by a floating point exception or an image reconnaissance glitch making my car think the truck in front of me is part of the sky.
If you die from a car crash it doesn't matter if it was from a drunk driver or an ai glitch. What matters is which is more likely- and humans are assholes. It will be orders of magnitude more likely to die by a human's hand than a random glitch.
> "But I don't"
I'm glad. Crashes still happen even when 1 party practices defensive driving techniques. Driving sucks.
At what threshold should one be willing to bet their lives in this probability every time they drive?
Right now we are speculating what the odds a self driving car has to make a mistake. But there aren’t any truly self driving car out there. So how do you determine the actual odds? If you replaced all the cars with self driving cars how many accidents would we have?
If 1,000 people a year die from software glitches, that's still a 30x+ improvement from the current situation.
I really don't see how people focus so much on "potential glitches" and not the current reality: drunk drivers, angry/hostile road-ragey people, people on the phone, people texting while driving, people on medication, people driving while sleepy, etc. A few glitches here and there is objectively superior to the current state of affairs.
This isn't true. The odds of me dying in a car crash is lower than the average as a result of precautions I take to be safe. For me 1,000 deaths/year might be a improvement (or it might not) but to say it's a 30x+ improvement just by looking at the total number of deaths is just false.
The focus on potential glitches is because it's something the driver has no control over. Is there a similar measure for number of completely accidental deaths (eg. someone swerving into your car or t-boning you at an intersection)?
Finally, until we have better laws that ensure companies are liable for their mistakes, companies won't take all the precautions to ensure vehicles are safer than a good driver (not just the average driver). Does anyone remember the Ford Pinto explosion issues because the company decided to use a cheaper gas tank and figured they would save more than they would pay from settling the few lawsuits that might emerge as a result?
I'm not specifically accusing you of this, but consider that more people than is numerically possible believe that they're better/safer than the average driver. There are a lot of people who believe they are much safer drivers than they actually are.
Regardless, just because you believe that you personally will be a safer driver than a computer, we should scrap the whole thing? What about all the people who aren't better drivers than the computer? Let's assume for a moment that you actually are safer than the eventual self-driving systems that are approved for general use -- which is by no means a certain assumption to make -- then maybe you just don't use or ride in a self-driving car? It's your choice, after all (especially in a place like the US, where I imagine manual-drive car ownership in a self-driving world will end up nearly as closely protected as firearm ownership). And sure, maybe someone else's self-driving car might hit you and kill you, but someone else's human-driven car might do the same. And if self-driving cars are doing that at lower rates than humans are, it's still a net win.
I think many people are taking this weird view that even though a self-driving car might make fewer mistakes (and cause fewer deaths) overall, it's somehow a worse situation that they'll likely make different mistakes than a human would; that is, a self-driving car might kill you in a situation where a human driver would save you. And that somehow makes the whole thing not worth it. I just find that line of reasoning to be flat-out wrong. It's an emotional appeal to some illusion of control. (Of course, unfortunately, logic doesn't write laws when it comes to contentious issues... emotion does.)
> The focus on potential glitches is because it's something the driver has no control over.
This is pretty short-sighted, because there are a ton of things that you have no control over when you drive your own car, and yet you've decided (in many cases likely unconsciously) that those things are acceptable risks.
I'm not saying you should ignore the possible risk of glitches, but focusing on a number that we don't even know yet, and immediately assuming that it will be too high for your risk tolerance is... a bit weird?
And that's the thing: I don't expect self-driving systems that have equal or worse crash records than humans do will be approved for use. And if they are, people will (rightly!) reject them. So any approved, accepted self-driving system will end up causing fewer deaths. Some of those deaths will be caused by outright bugs, and others will be caused by situations that a human driver would not be able to recover from either. All deaths are tragic, but fewer deaths overall is what we should be -- must be -- aiming for. Not playing games with control illusions. Not arbitrarily deciding that certain failure modes are somehow less acceptable than others when they cause the same (or even fewer!) deaths.
My position -- and what I believe to be the only logical, community minded position -- is that the glitch rate does not matter one bit. The only thing that matters is the overall death rate, and if self-driving cars have a lower death rate than human drivers, that should be enough. And if they don't, they should not be approved for use, and people will rightly reject them anyway.
I do agree with you that companies building self-driving systems need to be liable for mistakes and negligence to the same degree as human drivers are. Unfortunately that's harder to prove, but it's a necessary thing to figure out.
As much as I dislike the term "accident" when talking about car crashes (because, to me, the implication of the word is no one has to take responsibility), sometimes things just happen, because we are imperfect beings with imperfect nervous systems and imperfect perceptions and imperfect reaction times.
Self-driving cars will be better at a lot of things, but, yes, possibly worse at others. They have the potential to eliminate many causes of crashes, but might add a few new ones.
When you're on a plane, you're trusting not only the pilots, but a ton of complex avionics software. Why is that ok, while trusting self-driving isn't? I get that the two tasks are very different, and self-driving will require more sophisticated, nuanced software, but in both cases you're turning your safety over to a computer. That didn't work so well with the 737-MAX, but no one is talking about scrapping modern aviation because a bunch of people died due to bad software decisions.
The problem is the illusion of control. People think that driving their own car means they're in control of nearly every possible outcome, but in reality, they're not. Plenty of things can happen in a car that are out of the driver's control, even without another vehicle involved.
Another part of it is that people (Americans especially) can be excessively individualistic. Many people will balk at a solution that will result in (just making up numbers here) 25% fewer deaths if it means they personally will have a 0.001% greater chance of dying. Frankly, I find that mindset really worrying in a society, even if it can be understandable.
(Somewhat relatedly, I recall an episode of Star Trek TNG where one possible solution to the problem du jour was to give the computer full control over propulsion in order to save the ship. And we're talking about a futuristic computer that could probably flawlessly simultaneously self-drive every car currently on Earth without breaking a sweat. But in the end, blatantly pandering to our "human control is always superior" biases, the computer was found to be not good enough, and humans saved the day. Even more telling, I believe it was Captain Picard who took manual control; they didn't even have Data, the android, do it!)
Why do people (in particular people without robotics experience) keep assuming this will happen any time soon without any evidence whatsoever? "We must move over to self driving cars as soon as they are safer" is fair but it's almost tautological.
The only numbers we have so far show that despite only testing in near ideal conditions (e.g. Waymo choosing to test in suburban Arizona rather than Manhattan or rural snowy Midwestern areas), the accident rate for autonomous vehicles is actually significantly higher, not even counting "interventions" which every company measures in its own inscrutable way.
What I was discussing was the supposed ethics of fully automated driving vs human driving when crashes happen.
I'm tired of seeing these contrived, hypothetical examples about what an automated car may or may not do in a very specific scenario, that completely ignore the fact that crashes & deaths will go down by orders of magnitude.
Maybe crashes and deaths will go down by an order of magnitude one day. But my point is that that’s hardly an argument in favor of not worrying about the ethical implications of self driving cars today.
My view is that even if self-driving car corporations kill people with zero consequences (due to poor laws and poor oversight), as long as they're doing so at lower rates than human drivers are, that's still a net win. It's still not an ideal scenario, because those deaths could be further reduced under the threat of real consequences. But by and large, it'd still be better than what we have today.
If they're killing people during testing of experimental tech that hasn't been approved for general use (like the Uber case you mentioned), they need to be smacked down hard. If they're killing people at a higher rate than human drivers, they should not be approved for general use in the first place. If they're approved, but release an update that ends up killing more people, they again need to be smacked down hard, with the update immediately reverted.
As someone who spends more time cycling than driving, watching videos like this[1] give me more confidence in today’s self-driving technology (assuming good conditions) than in most drivers on the road, even knowing that any internal video is bound to have a dose of propaganda. It’s not entirely even that drivers are to blame; our infrastructure puts cyclists at risk as does the human limitation of only having two eyes.
The outcomes of those choices are known and have causes which are preventable; they are by definition not accidental in nature.
I know it's not intentional but words matter. Crash is a better word 99% of the time than "accident".
This post addresses "The Semantics of Intention":
https://laist.com/2020/01/03/car_crash_accident_traffic_viol...
excerpt:
Drivers aren't out there aiming for pedestrians and cyclists, so how does intention factor in?
UCLA's Madeline Brozen argues it can be traced back to both failure to follow road safety laws and a lack of understanding about how dangerous driving a car is — especially since unsafe speed is the top contributing factor in L.A. traffic deaths.
Research shows that a pedestrian struck by a driver going 20 mph has an 80% chance of survival. If that driver accelerates to 40 mph and hits a pedestrian, the victim's chance of surviving drops to just 10%.
"The act of going above the speed limit or going fast [in unsafe] road conditions...that is an intention," Brozen said. "When someone is driving in a way that can kill someone, they are creating a risk."
According to John Yi, another "degree of intention" in traffic deaths falls on car-centric society and L.A.'s leaders, who are "intentional about what we're building and what we're not building."
City officials have stated clearly that L.A.'s mission to eliminate traffic deaths is informed by the fact that "underserved communities are disproportionately killed in traffic crashes." But Yi argues that the historic neglect of those communities can be viewed as intentional.
"To take that away, I think, is really not looking at some of the most disinvested communities and what they're going through," he said. "To put it squarely on the shoulders of drivers and say it's their fault and they're the ones who should be moderating behavior is overlooking the situation altogether."
If only the people who get outraged about self-driving cars were as outraged by this fact and applied pressure to improve street design and speed limit enforcement.
One time I lost control of my car while driving on the highway and hitting a patch of black ice. Luckily, nobody died. But that happens all the time.
Presumably, a self-driving car will drive slower in adverse weather than a human might, as well as being able to control the car if it hits an ice patch better than an untrained human can.
Also, the car can notify other cars on the same road of the ice patch, instantaneously.
...except the corporation being fined and/or sued in civil court, along with the executives/engineers responsible facing criminal charges.
Look at the GM ignition switch scandal where executives knew explicitly the likely consequences of their decision and yet no real punishment was enacted even after repeatedly lying about the death toll (initially by at least one order of magnitude) that eventually officially reached 124 deaths and likely much higher in reality.
Look at Boeing's 737MAX scandal where executives also knew the likely consequences and worked to go around rules, regulations, certification in order to pretend those consequences won't happen. Both Boeing and authorities either buried or turned a blind eye to reports that this happens. No real punishment here either.
Autonomous cars or not, as long as corporations pay for your laws you will always be on the lower rungs of the ladder.
https://www.bbc.com/news/technology-54175359
> Ms Vasquez was charged on 27 August, and made her first appearance in court on 15 September. The trial is now set for February next year.
> Despite the decision not to levy criminal charges against Uber itself, the company did not escape criticism.
> Days before the crash, an employee had warned his superiors that the vehicles were unsafe, were routinely in accidents, and raised concerns about the training of operators.
> Following the crash, authorities in Arizona suspended Uber's ability to test self-driving cars on the state's public roads, and Uber ended its tests in the state. It received permission to carry out tests in the state of California earlier this year.
So despite all the safety failures by the company, just the hired driver was charged...
>So despite all the safety failures by the company, just the hired driver was charged...
A thought experiment: if a municipality was warned about safety failures about its streets (high speed limits, poor lighting/signage, lack of pedestrian crossings), and some kid got killed in a car accident, should the municipality be liable? What if everything that uber/the municipality did was within the law, and the only thing they're guilty of is not taking additional safety measures? eg. dropping the speed limit to 20mph in suburbs will probably eliminate all pedestrian deaths, should the municipality be liable if it set the speed limit to 30mph and the kid died?
"Tragic. Thoughts and prayers. Software problem. Nothing we could do."