Hacker News Reader

Comment by zer00eyz | Hacker News Reader

"Gartner placed autonomous vehicles at the peak of inflated expectations"

Sounds about right. I think that we all want to see self driving cars, but I don't think were ready for software with morality. If I'm driving a car filled with kids and I'm faced with the choice of running over the old lady or driving the kids off the cliff I'm pretty confident in what choice I'm going to make. Is that the choice everyone would make? is that the choice the programer made? do I have a morality switch in the car? will it make value judgements about the occupants and the external world when forced to make a bad call?

Its going to be interesting times if these things end up on the road.

View on HN

Yeah, there's always someone with a contrived answer like yours. If 1/10th the people die because the computer isn't perfect, aren't we much better off? Deaths are actually increasing this year: http://www.newsweek.com/us-traffic-deaths-injuries-and-relat...

https://en.m.wikipedia.org/wiki/List_of_motor_vehicle_deaths...

"In 2010, there were an estimated 5,419,000 crashes (30,296 fatal crashes), killing 32,999 and injuring 2,239,000.[2] The 32,479 traffic fatalities in 2011 were the lowest in 62 years (1949). Records indicate that there has been a total of 3,551,332 motor vehicle deaths in the United States from 1899 to 2012."

Self driving cars are by no means the only solution to road deaths.

Probably true, but they're the most effective tool that would have the greatest singular impact available to us right now.

Self driving cars are not available to us right now.

That's a truism. What are these solutions, how much do they cost, what are the relative pros/cons compared to self driving cars?

That's for self driving car advocates to state, rather than simply raising the bogeyman of road deaths every time there is any commentary that detracts from their position.

The easiest answer to your argument is for self driving car advocates to claim that it's the ONLY solution to reduce road deaths. Note that you are the one who claimed there are more than one solution.

Except that that is self-evidently false, and we're not talking about some meaningless argument in a vacuum, but one between humans who have experience of the world.

> That's for self driving car advocates to state

Uhh what? You said:

> Self driving cars are by no means the only solution to road deaths.

I asked you "what other solutions?" You're literally pointing at other unnamed solutions as better alternatives to road deaths, it is very reasonable to ask you to be more specific.

It is in no way up to driverless car advocates to complete an argument that you made.

It's fairly obvious that things like pedestrian detection systems with automatic braking, lane departure warnings, tiredness detection, etc. will make a significant difference as they become widespread. These exist in production today and similar technologies are in active development and deployment.

There is no reason to believe that human driven cars can't be made safe through technology without requiring self-driving. There may be other benefits to self-driving cars that are desirable, but as I said, the road deaths argument is clearly just fear mongering based on the false idea that other technological safety measures are not being developed.

As for costs and numbers - those certainly aren't provided by self driving car advocates, who can't even say when the technology will be available or how basic legal and economic problems it presents will be overcome.

We are much better off, of course, but that isn't enough to get the thing done. We also need to figure out how the "much better off" gets distributed, in a way that convinces everyone to actually do the thing.

If, for instance, Google is held liable for accidents while reducing the overall rate, self-driving cars aren't going to be a thing (unless you fix that particular problem). Or if the victims of self-driving car accidents have difficulty getting compensated.

Is my above scenario contrived, yes. Is it demonstrative of the kind of ethical questions that are going to have to be answered by developers, rather than individuals. It absolutely is. If you note I'm not even dismissive of "self driving cars", I only point out that its going to be INTERESTING if we ever see them.

The reality is that at some point, some where, some piece of software is going to have to make a very hard choice. The road will always have exposure to irrational actors (kids, drunks, a random person crashing their bike).

Are we at a point where we WANT software making that decision? If I have terminal cancer I might make a different choice than if I just had a baby and I was a human behind the wheel. Will we be able to inform the system that were willing to commit self sacrifice? Is there going to be weighing on the system based on ages of those involved? What happens when someone decides to kill themselves by jumping in front of a self driving car?

Lets forget about the legal battle that ensues the first time this happens. We will pretend that no lawyers will be allowed to be involved. The guy who wrote the code that said 4 kids > granny do you tell him? Maybe you have to go back and tell him his code killed 4 kids, and NOT granny and he has to fix it, cause that was wrong? I saved 32,999 people in THEORY but I know my code killed 4, they told me, it was a bug, I have to fix it now.

Lets look for a real world example where we see some of this today. Train drivers kill people on a fairly regular basis. Some times its because people choose to kill themselves, sometimes it really is an accident where there was no intent on the part of the person killed. Do you know what happens to those folks? Do you realize the impact these actions have on their lives? They certainly do, its part of the job, they get training before they go out and drive, the get counseling after. For some of them IT DOESNT HELP. Stats aren't people, they don't react the way we want or expect.

Like I said it is going to be INTERESTING.

Shouldn't you got the pick the morality of the car you're in? Which will probably ends up with everyone choosing to protect themselves. And if that causes too much externality to society (some kind of Prisoner Dilemma where everyone tries to protect themselves being the worst?), then we have regulation step in and regulate the morality of the systems.

It's not going to be easy, but I don't agree it's harder than the gazillions moral choice we already made as a society nowadays.

harder than the gazillions moral choice we already made.

Im not so sure on that. The two closest things I can find, that address something "remotely close" are the following, and they more or less drag the law into it (sadly) but it is a sort of moral arbiter:

This is an actual case: https://en.wikipedia.org/wiki/R_v_Dudley_and_Stephens

And the more philosophical version: https://en.wikipedia.org/wiki/The_Case_of_the_Speluncean_Exp...

There is a quote in the Dudley and Stephens one that is rather interesting:

"We are often compelled to set up standards we cannot reach ourselves, and to lay down rules which we could not ourselves satisfy. But a man has no right to declare temptation to be an excuse, though he might himself have yielded to it, nor allow compassion for the criminal to change or weaken in any manner the legal definition of the crime."

Where do you live that little old ladies are jumping out in-front of fast moving traffic with cliff edges that lack any safety barrier?!

The car is going to make the best effort to minimise impacts to its self, given there's a cliff for it to fall off it's not going to swerve over it. Also as said in a sister comment while you're car full of precious children go over the cliff you can find peace in the knowledge that on average less children have died because there's self driving cars.

I agree that my scenario is contrived. It was meant to be illustrative of the actual ethical dilemmas were going to face when this happens.

Does my car have ethics? If I'm terminal can I tell my car that in a me or someone else situation choose them? What other factors come into play if Im not saying I'm willing? Is it age? Does my car have some sense of WHO is involved and makes a decision based on some arbitrary stat? does it flip a coin?

In theory 32k people live with self driving cars. In theory 32k people live if we all stop driving. If you are give you the choice of writing two programs, one that prevents cars from working at all or one that is going to kill four kids what program are you going to write? There are a lot of things about this that aren't cut and dry at all. Like I said its going to be interesting if we ever get them.

How do you cope with situations where the person driving might not be aligned with your decisions? (taxis, Ubers, buses, etc.)

It doesn't bother me actually. The point of the clearly contrived situation is to make one think about the ethics involved.

If I'm terminal with cancer, can I tell my car that when it gets to a me vs them, that I'm willing to sacrifice myself? What happens when the me vs them scenario is someone attempting to kill themselves by jumping in front of a car? How does the car resolve the conflict in a case where it has NO other parameters? Are we going to say "auto sacrifice car occupant"? Do we have the ability to make a guess based on age of the involved? How about the "value" of the people involved? Would you trust a self driving car built by bill gates in a scenario between you and bill gates to choose you over bill? Like I said, its going to be interesting.

I'm having trouble imagining a scenario where I have enough time to consider who should live and who should die, make a decision, and act on it... but not enough time to just brake and avoid the accident.

Maybe autodrive cars, but I certainly do not want to see driverless cars. I want meat onboard. Licensed meat. Someone to get the tickets. Someone to sue when the robot does something stupid. Someone with a foot on the hydraulic braking system and mechanical steering for when the electrics fail.

Where is the meat operating Predator/Reaper/Global Hawk drones? Thousands of miles away in an air-conditioned cargo container in the outskirts of Las Vegas.

The chances of those drones flattening me and my motorcycle are extremely slim.

Motorcycles are already 45 times more dangerous than a car. I have a hard time (even as an ex-motorcycle owner/rider) feeling bad for motorcycle riders who are worried about self-driving cars (which are already better drivers than humans).

Except that in many ways a motocycle is no different than a bicycle, wheelchair, pedestrian, moped, horse+rider, or anyone else not wrapped in steal. Cars must share the road with non-cars.

Right. Google and several other tech companies are going to keep pouring vast amounts of money into self-driving tech (to ensure its ability to "share the road with non-cars"), and naysayers are going to yell angrily at clouds.

In this case, I appreciate the momentum behind self-driving vehicles. The sooner people are replaced, the better.

I see the momentum but question its direction. Why not start with driverless freight trains? That would seem a far more controlled environment and some cities (Vancouver, DisneyWorld) have been doing it with passengers for years. Then self-driving planes. They already have autopilots. When I see people lining up to board a pilotless 747 then I'll start trusting driverless cars to not flatten me from behind as I sit on my bike helpless at a red light.

Do you trust human drivers not to do that? Statistically, they're more likely to drive poorly in that situation.

Ok is there a difference between "writing software for global hawk" and "writing software for self driving cars"

As humans killing others is something we do, and based on history were pretty good at. We have the ability to justify our actions in some fairly ugly ways. There is a great quote from fog of war (check it out if you haven't seen it) on the firebombing of japan in WWII

"LeMay said, "If we'd lost the war, we'd all have been prosecuted as war criminals." And I think he's right. He, and I'd say I, were behaving as war criminals. LeMay recognized that what he was doing would be thought immoral if his side had lost. But what makes it immoral if you lose and not immoral if you win?"

So thats war, thats folks who build the engines of war, and send folks out to die and kill. I would say that in writing a missile guidance system, your writing it as "fit for purpose" and should KNOW and COMPREHEND what is going to happen when it is used.

Is that going to be the same mind set for someone building a self driving car? Would I, or you, feel any better knowing that I had saved lives in 'theory' or 'statistically' only to know that I had caused actual death with my programing? I don't know how I would feel, I'm somewhat thankful that I'm not facing that dilemma in my current job.

I think you missed my point. Ignore for the moment what drones are used for (a pilot still makes the final decision to deploy ordinance).

My point was, software has advanced far enough for unmanned vehicles can fly on their own and navigate without human intervention. The X47-B (the Navy's "Salty Dog" UAV) can take off from an aircraft carrier, navigate to waypoints, refuel on-air from a tanker, and land back on an aircraft carrier deck in the ocean without any human intervention.

Its not a stretch to think that self-driving cars are of similar complexity, and the problems around them will be solved in due time as well. There is no reason to have "meat" in the driver's seat anymore.

https://en.wikipedia.org/wiki/Northrop_Grumman_X-47B

I don't think the pilot makes that decision. They push the button according to essentially a decision flowchart. Their decision is based on evidence gathered by others, wanted lists drawn up by politicians, standards written by committees, and orders from commanders. They do not see the guy's face in the truck and match it to a wanted poster. (Even in than scenario the decision is partially made by the person who created the poster.) The pilot/operator may decide the exact moment the ordinance is released, but he is not the person making the decision to kill. That is a very shared responsibility.

A drone isn't the same thing as a car with a person in it.

Look at the problem tree for the drone, if it comes to a decision between crashing into the big expensive carrier or dumping it self in the sea, whats it going to choose? You can base this decision on dollars only and still make the right call.

The self driving car is going to face a genuine ethical dilemma at some point (or its coding will have to deal with one, as it isn't "self aware"). For the sake of simplicity lets make that "occupant or pedestrian". Are the lives of equal value? How does it choose? Am I allowed to influence that choice as the "passenger" in the car? By getting in the car do I have to agree to some TOS that says the car will sacrifice me to save someone else when it is unavoidable?

> Someone with a foot on the hydraulic breaking system and mechanical steering for when the electrics fail

A huge number of accidents are because of inattention or distraction today. How are you going to the the bag of meat to pay attention in case of an anomalous situation if 99+% of the task is being taken care of for them?

The meat is the last hope for when the machines fail. And they do fail.

For below: I totally agree that autopilots can be distracting. The same was said of cruise control and even automatic transmissions. We adapt.

Not sure if you missed the point of my question. Unexpected situations happen very fast while driving. If the human driver isn't sitting there ready to take over, they might as well not be there for that type of situation, and if they haven't had to do anything for the last half hour while driving, it's highly unlikely their eyes will consistently be on the road with hands ready to take over the controls.

Getting the human to take over from a failing autopilot is a fool's gambit. It takes the human thirty seconds to become as competent as a drunk driver, about a minute to get to normal competency levels.

The time it takes a human to regain competence suggests that the only way to safely hand over is for the robot to come to a complete halt before ceding control.

I think you strongly overestimate how helpful having a human being in the car would be in any of those scenarios. If my mother is the driver and the power steering goes out, she might as well not be there. And we already have well established systems for giving out tickets to cars whose owners aren't physically in the car (license plates, registrations tied to VINs, etc.).

Really? How can a cop issue a google-powered car with a ticket for reckless driving? Are cops now to submit bug reports when they see cars making mistakes or misreading signs? They can churn out the tickets, but the likelihood of a ticket altering the machines behavior is very slim. Now if a person is in the car, they can learn where the software falls down and adapt accordingly. If the robot doesn't spot the new speed limit sign, the human can.

I'm still waiting to see how these google-cars manage with hand gestures, people directing traffic. Just today I was waved through a red light by a volunteer firefighter not in any sort of uniform.

The actual problem will be that cops will have nothing to do and no ticket revenue to pay them with. We'd better get on what to do with a 70% decline in cop workload before they come up with some bright ideas.

> Someone with a foot on the hydraulic breaking system and mechanical steering for when the electrics fail.

That already doesn't exist in a lot of cars (drive-by-wire).

I have yet to see a production car that doesn't have manual brakes, either hydraulic in the case of master-slave cylinders or mechanical in the form of parking/emergency brakes, that can be engaged without electricity. There might actually be a rule about that, else parked cars could start drifting away if their batteries run down.

Many cars have electronic parking breaks.

But is there a model that, if you disconnect the battery, has no brakes? The few electronic parking brakes I've seen fail into the on position, much like truck trailers and or traincars. They lock if they loose voltage/pressure from the controlling system. And even those cars had master cylinders under the pedal.

There are no break by wire production cars in the USA. Infinity sells the only steer by wire car. It uses clutches that engage a direct mechanical link if the electronics fail.

Reply on news.ycombinator.com