The point there is that a human should not be evaluated by its attributes. No matter the circumstance. Why? Because we humans try to do the same.
The AI should not be permitted to do that. To it, it should be just two humans.
“All Humans are equal” is like one of the first things in most constitutions and international human right threaties. That’s probably why it’s the first thing Germany put in law … they also could make it so that AIs have to follow the constitution … although that could be a problem as it isn’t really designed with machines in mind.
I have to agree with playbenni. “All humans are equal” is a nice mantra, but I also think it doesn’t really reflect what people actually feel (even If they won’t admit it) and do.
It certainly should be the guideline, but I think in some situations going against that rule would be the more socially acceptable and moral option (grandpa vs child for example). Much worse decisions are made by humans all the time, for example the appalling abortions of fetuses with genetic abnormalities (down syndrome comes to mind). So I think there is definitely some wiggle room for exceptions.
The machine ultimately works for me and I payed for it, it should do what I want it to do, no matter how immoral that might be, so it should be user configurable.
There’s an amazing wrongheadedness in the idea that an automated car can enter a situation where it must result in someone’s death, yet still be able to moralise over who should be the one to die. If a car can be programmed to only kill particular people in the case of an accident, it should be programmed to avoid the damn accident in the first place.
There are solutions for this. Signed OTA updates for firmware and no remote access to vehicular automation functions.
Nope. I described a situation that nobody can handle. This “kill or be killed” scenario is an absurd extreme - unless the car is normally careening around roads, largely out of control.
Engineers are already designing automated cars so that they don’t enter situations they cannot control. That’s why they turn control over to the driver if they sense any departures from their defined envelope. But what we’re going to find is that engineers will be able to create cars that drive far better than people. When that happens, they will not turn over control to passengers. But when that one day comes when even the car’s capabilities are exceeded and the car senses that somebody is going to die, it will have to follow some builtin logic to decide what to do.
In that case, the lives of those in the car are lower on the priority list than the lives of those outside the car.
The reason that I describe highly-capable self-driving cars in this discussion is that we’ll have that high capability available by the time cars are capable of deciding who dies. As I said, the far more interesting discussion is how engineers are going to code relative values below the “kill or be killed” level. I assume that will be governed by a series of laws once a few key events take place. For example, the first car that runs over a beloved dog instead of driving into a tree and producing $5,000 in damage to the car’s exterior.
That will be part of the beauty of having all those sensors. Just as dashcams constantly record and discard, so too could the car’s sensors. Hit the “record event” button when something happens and all that information is now stored and available to understand what happened. The car’s sensory apparatus should be dramatically better at describing events than the people themselves.
Well we are talking about a hypothetical situation here. And I think there can and WILL be incidents in the future when an Ai must make a decision which human it has to save. Have you even seen iRobot? The whole premise of that movie starts with that decision.
I think we can create guidelines which life is more valuable. Furthermore they already exist in a way. When that fool crashed the germanwings flight into the alps, the insurance payments for each murdered passanger differed wildly. Relatives of small children only received a few thousand euros in payment, relatives of high payed adults received the most money. They based the payout on the monetary loss of each person. Sounds cruel, but that is how the law handled it.
It is a fact that all humans are not equal in the eyes of the law, this notion in the constitution is laughable.
Abortions of people with down syndrome are allowed in germany up to the moment of birth. When the child is already long viable. People with down syndrome are therefore categorized as subhumans and their murder is legal and commonly practiced in germany. How fucked up is that?
Assuming the scenario in which a kid suddenly (unpredictably) jumps inside of the car, giving control over to the driver isn’t really a good idea. The driver, however strongly the company is going to urge the them to pay attention, probably isn’t paying attention. When automation takes over, “driver” interest in what is happening on the road is going to drop to near zero. By the time the driver is alerted that the car is outside the envelope, the driver knows what’s going on, and reacts, it’s going to be far to late.
At some point, this scenario will happen, therefore the car companies must plan for it. IMO, the driver is not a good fallback plan. Ideally, with the driver having hands on wheel 100% of time, feet right above the brake/accelerator, the driver is indeed in the superior position to make a split-second moral decision. But what’s the point of automation if we need to baby sit it 100% of the time? The odds of something very bad happening varies from point to point, but there is always the exception, which is what we are currently debating.
There must be a contingency program built into the vehicle.
That said, the way we can tell that automation is working is simple: Is it performing better than human drivers? As long as this criterion is met, we’re in the clear. About 40,000 people die in the US every year. So if there are significantly less people dieing on the road, it’s working.
There’s a fairly simple solution to this. The car doesn’t move if the driver isn’t paying attention to the road. If the driver’s attention is getting distracted during driving, the car can find a safe place to stop.
Well, the removal of crossings, for a start. It’s also possible it’ll allow the driver to do things that don’t sufficiently distract them. The attention requirements will also naturally become less as the car becomes more capable, as it’ll be pretty evident a bad situation is unfolding before the car needs human intervention.
That would be ideal, and I can see that happening for the early years of the self driving vehicles entrance unto the mainstream auto market. I’m certainly not against paying attention while driving, indeed many ‘drivers’ I come across are actively not paying attention.
However, as the technology advances, and we humans are going to be more and more disconnected from the driving process. Eventually the self driving car will be in a situation where it would have to make what we would call a moral decision.
I still contest that until someone proves that is the case, the focus should be on preventing that happening over making a moral decision. Because otherwise you are intentionally building murder machines
I know this is sarcasm, but I agree with the point you’re making.
This is how I see it: although we can put off having vehicles take over completely, unless there’s a concerted effort to “keep humans in the drivers seat”, and having the car forcibly pay attention, having a self driving vehicle is just asking the occupants to text, stream, or do their nails (I saw this) while in the car.
When, in the situation arises where in our hypothetical situation a kid jumps in front of the car, and the kid or the occupants die or are injured, it’s basically a lose-lose situation for the maker of the car. If the car was supposed to hand control over to the driver, even if the driver was ‘paying attention’, the people will demand why there wasn’t a program to avoid the accident. If the car had a sub-routine that kicked in, then people will demand why the program was so bad, or why there was even one in the first place. The reason why we or they would bother with self-driving cars is the fact that overall, they are simply better than us at driving, which would be reflected with a statistically significant decrease in overall accidents.
IMO, if self driving cars take over to compensate for our own inability to pay attention, we as consumers are going to rely on these self driving cars more and more to get us from a to b. Thus, the demand for less human input is going to be present by the consumers to the car manufacturers until we can do our preferred car activity without worrying where we are going. This is probably many years off, but it’s what I believe is going to happen.
But for the moment, yes, minimizing the risk is the best choice.
Topper, what makes you think that the car entered a situation it couldn’t handle? If a bicyclist suddenly emerges from a blind corner onto a street illegally in too short a distance for the car to stop, the car did not enter a situation it cannot handle, it was forced into the situation. Maybe a tire blows out, causing the vehicle to swerve unpredictably until the systems can correct. There are PLENTY of situations that are impossible to predict, and are therefore unable to be programmed for.
The only reasonable situation this can be asked is the following:
-The car is behaving 100% as intended by the manufacturer.
–This means it is also following all laws applicable to the situation, which makes the car NOT at fault for the situation.
–This further means that the non-occupants ARE at fault for the situation.
The car is capable of differentiation between pedestrians, cyclists, animals, and inanimate objects.
–Asking a car traveling at 35mph (a fairly normal speed limit in occupied areas) to identify race, gender, or age in a fraction of a second and make decisions on that knowledge is absurd. A person is a person, regardless of these attributes.
-The decision will either result in the death or serious bodily injury of the non-occupant(s), or the likely lesser odds of death or serious bodily injury of the occupants.
–There is also material damage to consider, but I believe that human life should be the highest priority over any amount of monetary damages.
Given the above assumptions, here is my belief on a variety of situations:
The car has a clear and open view of the surrounding area, and sees a path that will result in no immediate danger to persons outside of the initial situation: the car should choose the path with fewer POSSIBLE casualties. If it is one occupant, two non, then the occupant is fewest.
The car only sees paths of avoidance that include persons not initially involved in the situation: the car should never make an effort to involve previously uninvolved individuals, regardless of possible numbers of casualties.
The car only sees paths involving an unknown or unpredictable follow on pathing: the car should make every effort to reduce the accident’s effects while staying on the road. Roads are intended for vehicular travel: sidewalks, ditches, and other implemented are not.
My twopenny’s worth on automated vehicles and collisions:
My experience is that car accidents are mostly caused by someone doing something unpredictable (whether that’s the driver, or someone else).
In that situation, a human will likely react in an adrenaline-fuelled, emotional manner. Hopefully, if they are a competent driver, the instincts will be correct. However, the best driver can’t always take into account all the variables of the situation, nor do they always react fast enough.
Electric systems are faster. Through a range of sensors, they’re probably also aware of more detail in the immediate environment. The question is whether they are better at recognising when situations are dangerous - can we reduce a driver’s intuition to computer code?
My point is that extreme situations like the ones described in this thread are rarely, if ever, a decision. With a human, it’s probably brake, swerve and hope. A computer will attempt to brake and avoid with greater speed than a human, but the crucial difference would be that it might spot another child the driver didn’t, making it go the other way. It won’t follow instinct, it will follow the data it has at the time.
Right, but this thread is discussing what it will do with the data it has. Like you said, it will be reacting as programmed, not instinctively. As such, it must have deliberate actions programmed, and those actions are currently up for debate as to what is the most “correct.”
Topper, you’re asking questions which I’ve already answered. First principles: the operator brought the car into the situation and is responsible for its actions. So if there is a situation where the car must make the decision, the engineers must program it to favor the life of others. When a human driver makes that decision, we declare them a hero. When a human driver decides to save their own life and kill the other person, we declare them a villain. In the very least lacking in character.
I’d prefer to have altruistic values programmed into our machines rather than darwinist ones. Each can lead to dystopia, but the latter gets us there rather more predictably. And violently.
In the near term, autonomous cars are going to turn over control to the driver whenever the car can’f figure out if it’s operating within its performance envelope. Figuring out who must die is far beyond that level of autonomy.
In the long term, the performance envelope of an autonomous car is going to grow far beyond that of a human driver. They will never turn over control to the driver because driving will be a thing of the past. Human competency in driving will drop like a paralyzed falcon. Sixteen year olds will no longer take driver’s education or get a learner’s permit. They’ll just get into a car and say where they want to go.
That advanced car is going to be essentially the God Mode car. If something is thrown at it that is so extreme that it knows that somebody is going to die (and that’s after crumple zones, air bags, ejection seats, teleportation and anything else we come up with), it must be the car’s occupants. Extreme events are things like a bridge collapse. It’s outside the limits of what a car can do anything about.
And let’s remember that the God Mode car isn’t going to leave the other people happy and smiling, completely unaffected by the event. They will probably be injured in the car’s attempt to keep the death count as low as possible.
