There's an old saying around gun owners: "There's no such thing as a firearms accident. Only negligence."

I think the same thing applies to autonomous things.

In NYC the urbanists have a trend of saying "crash" instead of "accident" when referring to traffic collisions. They feel that "accident" gives the driver too much credit in situations where they were completely at fault for not looking, going too fast, running a light, etc.

> In NYC the urbanists have a trend of saying "crash" instead of "accident" when referring to traffic collisions.

Everyone I've known in law enforcement, especially CHP officers, does this by default (though they tend to prefer “collision”) when referring to an event where a car hit something. As one of them explained, “collision” or “crash” is a readily discernable physical fact; “accident” is a conclusion about the mental state of the participants.

Which by the way is idiotic. Last century, people switched from "crash" to "accident" for exactly the same reason -- to shift liability from the machine to the operator. IT's a nonsensical euphemism treadmill.

The idea that an "accident" isn't the drivers false is an absurd premise. Drivers are obligated to be diligent. "Crash" is a neutral term that doesn't assign liability to anyone.

I feel that this sentiment could be applied to almost anything and isn't really meaningful. Almost every accident ever could be attributed to negligence somewhere.

I think the point here is a firearm in particular is designed to project deadly force and should be treated as such. If someone "accidentally" shoots someone, it means they were negligent in their treatment of the instrument in their hand.

Autonomous vehicles are expected to be able to safely navigate a large variety of streets and highways, and possibly trails and countryside in the future, without killing anybody else around or killing the people inside, nor destroying the fixtures and buildings that make up your typical city street. The only problem is cars are inherently deadly, go look up the number of people that have died in automobile "accidents" since the automobile was invented, total it up for just one country. They need to be treated with and afforded the same respect as any other instrument intended to project deadly force, because they do, as do whatever algorithms are controlling the instrument.

I try pretty hard to refer to "vehicle collisions" and things like that instead of "car accidents".

So yeah.

Agreed; more generally, there's a movement to avoid calling these incidents "crashes" instead of "accidents", since it's arguable that crashes are avoidable: https://www.crashnotaccident.com/

Accidents are avoidable. Legislating euphemisms is feelgood nonsense.

This is a dumb saying that ignores the fact that humans are fallible beings who make mistakes.

Sure, you might shoot yourself because you're just careless with your gun. Or you might be also thinking of an argument you had with your spouse, or you might be tired when you're handling your everyday carry on your way to work, or you might get startled, or a million and one other things that might prevent you from singly focusing your attention the gun you're handling.

Machines exist to serve us, not the other way around. If a product doesn't take human habits and flaws into account, it's a bad product, because over a large enough sample size someone will make a mistake.

> This is a dumb saying that ignores the fact that humans are fallible beings who make mistakes.

It's not, because the basic rules of gun handling are designed in such a way that failing to obey one of them will not result in injury. They are:

* Never point a gun at anything you don't wish to destroy * Keep your finger off the trigger until you're ready to fire * Always assume the gun is loaded * Know your target and what's beyond it.

> Machines exist to serve us, not the other way around. If a product doesn't take human habits and flaws into account, it's a bad product, because over a large enough sample size someone will make a mistake.

Modern firearms are absolutely designed to this criteria - it takes a willful act to discharge them. If someone "mistakenly" fires a gun and hits someone, it requires multiple failures and clearly rises to the point of negligence.

As long as you follow the basic rules of car safety, you won't crash. Stay within the speed limit, be aware of your surroundings, keep a safe distance from the car in front of you, signal 100 feet before a turn. Unless someone else doesn't follow the basic rules of car safety!

You know what system we've come up with to improve car safety that has only one rule to remember? Always wear your seatbelt.

You know what we've come up with that requires zero effort from the driver? Laminated glass. Crumple zones. Airbags. Automatic braking.

This isn't about whether people use a product negligently — put it in enough hands and it will happen. The point is to design products such that the consequence of negligence is not injury or death.

I agree completely. Here, there's perhaps negligence of the system, and the driver that was supposed to be overriding it in situations like this.

Yeah, it will be interesting to see if this was an intentional (algorithmic) response or a failure of sensors somewhere.

I go back and forth with thinking the algorithms in things like self-driving cars should be public. Anything which must make a choice such as smash into bus or smash into bicycle — the public should be aware of the choices the company chooses to implement. Obviously this example is simplistic but it illustrates my point.

Today I’m definitely leaning towards it should be public and scrutinized. Next month I may be worried this would hamper progress.

I'm pretty sure everyone involved and observing also vacillate between those two mindsets. Regardless, events like this will likely hamper progress.

Forgive me, I don't understand what you mean. Please elaborate.

I suspect they mean the trolley car problem: "you are either going to crash into a group of school kids, or into a group of nuns. Who do you kill?"

I.e. did the uber car kill this 1 pedestrian, because it avoided killing 2 other people? Did the uber car opt for the "least killing" in this siutation?

I think this whole trolley car question - while a nice philosophical question - is silly though. I dont think humans would do any better in an emergency situation. In that split second when you suddenly find yourself in the "oh sh-<IMPACT>" situation, do you have time to a) make a deliberate, reasoned & fully-informed decision, and then b) control the vehicle effectively to follow through on that decision? I doubt it. If you had a second or two to think it through and then deliberately make a choice and steer towards that crowd of nuns, you'd probably be able to entirely avoid the accident anyway.

More likely is you'd probably do what I am sure most people do which is gasp, stamp on the brakes, shut their eyes and hope for the best ... assuming you even had time to realise there was about to be a crash before it happened. Some people might swerve, but I suspect they do that instinctively to avoid something in their way, not as a decision to hit something else.

I just dont think self-driving cars will get themselves into situations where they have to chose who to kill in the first place. And even if they did get into those situations, I really, really doubt it would be due their actions, and I'd certainly trust it to avoid the crash in the first place a whole lot more than the average human driver in the same scenario.

Some could argue "Ah but yes computers are so fast that they CAN make that informed decision about who to kill in milliseconds! So they have to make a choice! The question stands!". I'd argue back that in those milliseconds before the crash was inevitable, they'd act to prevent the accident before a human would even know what the hell was going on anyway. After that it just starts getting into a game of who can conjure up the most ludicrous hypothetical situation that rarely - if ever - happens in real life.

The most obvious example I've seen is when the car (or indeed driver) has the choice to endanger the passenger or a pedestrian.

For example, if a kid runs out into the road in front of the car, it can drive into the kid, likely with no injuries to the passengers. Or it can swerve and drive off of the road, or into another lane, likely saving the kid but putting the passengers at much greater risk.

As you say, in this or any similar situation the human reaction is probably going to be to instinctively brake or swerve without a chance to consider options or consequences. I feel like a computer will have a few cycles to spare to make decisions like that.

One of the most oft-discussed questions around self-driving cars is what should happen if your car can choose between two options: 1) An action that will have a better chance of protecting the people inside the car, 2) An action that will protect people outside of the car, with a whole bunch of variations.

For example - What if someone steps out into the road and you can either hit them, or run into a wall? The former makes it far more likely the occupants of the car will be safe, the latter saves the life of a pedestrian at the expense of possible injury to people in the car.

It is kind of an interesting question, because people can't generally react quickly and rationally enough in a situation like that for it to matter, but in theory, a computer can. But it also dominates the discussion in a way that likely far overstates how frequently this will actually be a decision a computer has to make.

I think that is what the person you're replying to is asking. Given that this occurred at night and on (very likely) uncrowded streets, I feel like its far more likely the car (and backup driver) just somehow missed the pedestrian.

They're basically asking if the car avoided doing something that would have been more destructive (e.g not hitting 2 people but only hitting one person instead). Or if there was no conflict decision making that happened and the car accidentally ran into the person and it resulted in a fatality.

- AI decides between one life lost as a better potential outcome

- AI made no decision and just happened to hurt someone

I think it's a reference to the imaginary trolley problem, which is often raised as an objection to self-driving capability.

The idea is that making a decision between killing one person vs killing something more valuable than one person (eg. multiple people) is something that drivers do on a regular basis. Computers may not be capable of correctly evaluating the ethics of that decision.

I believe they are asking whether that the vehicle struck pedestrian A in the course of avoiding something else like a pedestrian B.

To maybe make it maybe better: Pedestrian B and C.

Please don't, he's going to launch into some philosophical dilemma that's not relevant because these cars aren't making decisions.

No, I'm not going to do that. I just didnt know what they meant.