Slashdot Mirror
Drivers Prefer Autonomous Cars That Don't Kill Them (hothardware.com)
"A new study shows that most people prefer that self-driving cars be programmed to save the most people in the event of an accident, even if it kills the driver," reports Information Week. "Unless they are the drivers." Slashdot reader MojoKid quotes an article from Hot Hardware about the new study, which was published by Science magazine.
So if there is just one passenger aboard a car, and the lives of 10 pedestrians are at stake, the survey participants were perfectly fine with a self-driving car "killing" its passenger to save many more lives in return. But on the flip side, these same participants said that if they were shopping for a car to purchase or were a passenger, they would prefer to be within a vehicle that would protect their lives by any means necessary. Participants also balked at the notion of the government stepping in to regulate the "morality brain" of self-driving cars.
The article warns about a future where "a harsh AI reality may whittle the worth of our very existence down to simple, unemotional percentages in a computer's brain." MIT's Media Lab is now letting users judge for themselves, in a free online game called "Moral Machine" simulating the difficult decisions that might someday have to be made by an autonomous self-driving car.
The article warns about a future where "a harsh AI reality may whittle the worth of our very existence down to simple, unemotional percentages in a computer's brain." MIT's Media Lab is now letting users judge for themselves, in a free online game called "Moral Machine" simulating the difficult decisions that might someday have to be made by an autonomous self-driving car.
People value their own lives..
No one will ever program a autonomous vehicle to choose one life over another. That's a lawsuit waiting to happen, if not an outright murder charge.
Save the environment, reduce carbon emissions, save water, reduce debt... unless it affects me financially.
thegodmovie.com - watch it
they're passengers. the drivers can't be killed because there are none.
If we could get an AI that can kill for a parking space, I'd be fine with that.
Have gnu, will travel.
These scenarios are just a little bit contrived... I can't fathom any real life scenario where any of these situations would occur with the odds of both options being equal, which is the point where the software would be called upon to exhibit a preference of one option over another.
Exactly. Why don't people discuss the millions of small decisions - "how quickly shall I go through this stop sign?", "should I signal this turn or is it too much hassle?". Those are where the existing human software is causing bad consequences on a daily basis.
No, let's discuss the one in a billion corner case instead.
I'm just waiting for the next movie where the main character is being chased down either by a draconian government or some super hacker. The main character clearly knows the risk, so he's driving a 1969 Mustang, but suddenly, all the cars on the freeway start chasing him down and trying to run him off the road.
Looking for a computer support specialist for your small business? Check out
Self driving cars will transfer the liability from the owner of the car to the manufacturer of the car. This is already happening. Otherwise, they could never sell a car to anyone. But if the liability is held by the manufacturer, you can be sure the crash algorithm will be one that minimizes total casualties (and thus total liability).
And notice that this is the same issue behind the Will Smith film, "I, Robot". Will's character is rescued from drowning by a robot that lets a little girl drown instead. The robot had calculated the chances of saving each and Will won the AI lottery.
"He took a duck in the face at 250 knots." -- William Gibson, Pattern Recognition
Wait a minute... Aren't the republicans the pro-life party?
I prefer driving my own car.
Hahaha. It's even simpler than that. Everyone seems to be making the assumption that the cars will be such driving geniuses. That's not going to happen for quite a long while.
0) We all know that stopping in the middle of the highway is dangerous, BUT the way the laws are written in most countries, it's practically always your fault if you drive into the rear of another vehicle especially if it didn't swerve into your path and merely braked suddenly, or worse was stationary for some time.
1) Thus for legal and liability reasons the robot cars will be strictly obeying all convincing posted speed limits (even if they are stupidly slow by some mistake, or by some prankster), and will stick to speeds where they would be able to brake in time to avoid collisions or at least fatal collisions. Whichever is slower.
2) In most danger situations the robot cars will brake and try to come to a stop ASAP all while turning on its hazard lights. Which shouldn't be too difficult at those said speeds.
3) If people die because of tailgating it's the tailgater's fault. Same if the driver behind doesn't stop.
4) There are hardware/software failures then it's some vendors fault.
5) If braking won't avoid the problem even at "tortoise speeds", in most cases fancy moves wouldn't either. In the fringe cases where fancy moves would have helped but braking wouldn't AND it would be the robot car's fault if it braked, the insurance companies would be more than willing to take those bets.
The odds of the car being designed to do fancier moves to save lives are practically zero. If I was designing the car I wouldn't do it - imagine if the car got confused and did some fancy moves to "avoid collision" and killed some little kids. In contrast if it got confused and came to stop ASAP if any little kids are killed it would more likely be someone else's fault.
If you are a human driver/cyclist/motorcyclist you better not tailgate such cars.
Look at the Google car accident history, most of the accidents were due to other drivers. Perhaps I'm wrong but my guess is it's because of "tailgating". Those drivers might still believe the AI car was doing it wrong but the law wouldn't be on their side.
You'd be idiotic to purchase a car which might sacrifice your life or health in ANY circumstances.
If everyone followed that logic, the only vehicles sold would be SUVs, with a speed limiter to 20mph.
If you think I voted for Trump because of this post, you're wrong. I voted for Dr. Jill Stein of the Green Party. Again.
What kind of crazy, concocted scenario are you coming up with where the AI controlling the car has to make a Boolean decision that kills people?
My car is driving down a busy road at a safe and steady 30mph. There is traffic in the opposite direction travelling at 40mph. The sidewalk alongside is crowded with people.
A child suddenly runs onto the road 4 feet in front of the car. There is nothing my vehicle can do to stop in that distance. It is mechanically not possible. However, it can swerve left or swerve right. One direction means a head-on collision, the other means mowing down a dozen pedestrians. Or maybe it does nothing and strikes the child. Whatever decision it makes will result in injuries, perhaps serious, and possibly deaths.
Which should it chose? Maybe my car reckons its safety systems will protect its passengers from the head-on. But unless it instantly enters into a split-second negotiation with the head-on traffic, how does it know what the outcome will be for it? What if the head-on traffic can react and avoid the collision? Or maybe it has actually got ancient, slow, AI and will not avoid the crash? What if it's a model with far superior AI, and has already calculated a path that is optimal for its passengers, but really bad for you?
This is not a crazy or unlikely scenario and involves the AI making several life/death decisions, perhaps in competition with other AIs doing the same.
The answer is always the same. Brake.
Doesn't matter if it's a car, or a crate full of little girls with cute bunnies.
Swerving anywhere is never the right answer.
If you (or the computer) had enough time to consider whether or not to swerve into some direction, all you did was waste time you could have spent slowing down.
Also, in that situation, your car should already be slowing down the moment it detects another car cutting you off.
Bad example. Your car is going 44 fps, so it has 0.09 seconds to do anything about this problem. In that time, it can't stop, and it can't swerve any meaningful amount.
So it doesn't matter what it's programmed to do, it's going to hit that child.
"I do not agree with what you say, but I will defend to the death your right to say it"
This is the same as the Trolley Problem, a famous philosophical dilemma, first proposed in 1967: https://en.wikipedia.org/wiki/...
Basically, a runaway trolley is going to kill five people. You can either do nothing and let the trolley kill them, or pull a lever to switch it to another track on which it will kill only one person. There are many variations, including one in which you push a fat man onto the tracks to stop the trolley. Philosophers have written a LOT about it. Here are some humorous variations:
http://existentialcomics.com/c...
https://xkcd.com/1455/
http://www.mcsweeneys.net/arti...
Driving involves making life an death decisions every day. People don't like to think about it.
The emergency instruction you have provided "Stop as quickly as possible to avoid hitting thing" will work in 70%-80% of all emergencies. In many cases "Stop as quickly as possible" will not solve the problem.
The next option could be "Swerve around impediment/obstacle" so the car drives around objects that you can't stop for (moving debris, running people, swerving cars/bikes/motorcycles). The swerve option may be safer for the traffic pattern than a crash stop. You need to prevent being rear ended as well.
These two directives alone can probably take care of 95% of the issues out there. Stop or go around the problem. The issues arise in the last 5% where the two directives conflict with each other. Easy scenario: There is oncoming traffic in the oncoming lane, a person/child steps out from behind a visual obstruction (signage, truck, etc). The person has stepped into your path of travel and is inside your braking distance. If you continue straight you will hit the person, If you swerve left you will hit an oncoming car, If you swerve right you will swerve into an obstruction (car, lamp post, etc).
I don't expect you have an answer for this no-win-scenario. People have to recognize that 5% where something bad is going to happen; and there is some choice/action that can be taken that will affect the outcome (number of people injured, types of injuries, etc). This is the problem that people are trying to wrestle with. You are presented with an ugly-no-win-scenario. Make the best of it and decide who gets killed, injured, maimed or saved - Yourself, the pedestrian, another driver?
Architectural plans are like computer source code with a couple of differences: You only compile once.
Given the current status of things the least worst solution may be to have divided lanes (think express lanes on freeways) just for AI vehicles. When all of them are 'thinking' the same thing then the chances of problems decreases exponentially. Most of these issues seem to come up when there's a mix of AI and meat sacks.
Sure this will limit their use but that's what you do when new behaviour is introduced into an established system. Continue doing testing in a mixed environment but create the programs for a controlled environment to get things started.
A computer should serve its owner's interests with absolute priority over the interests of all other parties. Period. If it's my computer -- my agent -- then I am #1. By default (without my interaction) it should allow a million children to slowly burn to death if it means that I get to skip an ad. (That's a ludicrous example, but if people want to explore the edge cases of the policy I'm advocating, then there you go.)
You're going to find that this strongly favors protecting other people anyway. The "someone must die, pick who" scenario is extremely rare to the point of non-existent, compared to the routine "avoid having any collision at all, so that no damage or injury happens" scenario. (Stop smoking before you drive yourself crazy with fear of being hit in the head by a meteorite!)
That's not a global policy; that's just the policy for my computer. I don't mean I'm more important than you; I mean that to my computer I am more important that you. And your computer should serve you, too!
As copyright owner of this comment, I authorize everyone to defeat any technological measure which limits access to it.