Autonomous Cars

[AJI]

...I´m wondering about the reason I took the effort to reply. You will never decide to kill yourself or kill others, you will always try to avoid the accident and kill none...

It's a conundrum Andres, an interesting conundrum, that's why you replied... Yes it's morbid, but it's a conundrum that bears thinking about.., no?
I've implied in the past that there's an actuary out there who has accounted for the outcomes of this circumstance (and every conceivable circumstance for that matter...) The manufacturers have been running the numbers forever. Remember the Pinto?

[Greg Locock]

You are postulating an example which requires perfect knowledge. That is why the trolley problem is such an idiotic example. You deal with each emergency as it arises. Would you run 3 children over because you might drive into a ditch if you swerve? No. And then you might die? Right, so why expect an AV to be any different.

I wouldn't kill myself to save others that were clearly in the wrong.... Why would I allow a machine to force that decision into my life?

Seriously, you would rather run into 3 children who were in the road than risk driving into a ditch? Good grief. You cannot imagine how deep my contempt for you is at this point. And as for downvoting me, --- it's a badge of honor. ....... By the way, in many legislatures, cars do not have a right of way.

Personal comments removed

[Zynerji]

You are postulating an example which requires perfect knowledge. That is why the trolley problem is such an idiotic example. You deal with each emergency as it arises. Would you run 3 children over because you might drive into a ditch if you swerve? No. And then you might die? Right, so why expect an AV to be any different.

I wouldn't kill myself to save others that were clearly in the wrong.... Why would I allow a machine to force that decision into my life?

Seriously, you would rather run into 3 children who were in the road than risk driving into a ditch? Good grief. You cannot imagine how deep my contempt for you is at this point. And as for downvoting me, --- it's a badge of honor. ........ By the way, in many legislatures, cars do not have a right of way.

What about my children becoming fatherless? That rational self-interest bothers you so much?

[hollus]

Keep it civil, please. That goes both for trolling and for insults.

[Just_a_fan]

What about my children becoming fatherless? That rational self-interest bothers you so much?

You have had children, and they can go on to have children so you have secured your future. By running over the other children, you have prevented their futures.

Or one could argue that the children in the road are entries for the Darwin Awards...

[henry]

I wouldn't kill myself to save others that were clearly in the wrong.... Why would I allow a machine to force that decision into my life?

Seriously, you would rather run into 3 children who were in the road than risk driving into a ditch? Good grief. You cannot imagine how deep my contempt for you is at this point. And as for downvoting me, --- it's a badge of honor. You are an utterly miserable specimen. By the way, in many legislatures, cars do not have a right of way.

What about my children becoming fatherless? That rational self-interest bothers you so much?

It’s interesting you mention rational, the part of your brain that would make the decision under consideration is not rational. It’s autonomous, basing its decisions on prejudice.

I use prejudice in its pure form. It bases on prior knowledge and experience. Some of that is very prior, coded in DNA and it’s manifestation in the brains structures. Some of it is taught to it by the rational part of the brain by repeated stimulation. This is why some elements of driving are a “skill”. When we first start everything is rational, we have to teach the brain, slowly because rational thought is slow, the movements that result in the car behaviours we want. This is the one of the things the autonomous car is trying to simulate and it is taught to it in a combination of offline training and online refinement, potentially it is better than the average human at this.

Once that skill is in place, further skills can be taught around situational awareness. Only a proportion of these get to the autonomous level, dependant on the focus the rational mind gives to them. This is where AVs are at a disadvantage. Humans can use similar experiences to load up the autonomous part of the brain. So moving about is generic and one can learn from crawling, walking and running. As is understanding an classifying the visual world and the objects in them.

It’s not clear that your rational choice in this scenario could be transferred to your irrational autonomous brain. 2 or 3 instances a month of your muttering under your breath about ill-disciplined children may or may not do it. But of course you may have used other experiences to train your mind so the behaviour you want may come about based on prejudices you’ve inherited or learned about the people about you.

Whatever happens in such a situation, if you survive your rational brain will take plenty of time to fit the behaviour of your autonomous brain to your preferred model.

[Andres125sx]

You are postulating an example which requires perfect knowledge. That is why the trolley problem is such an idiotic example. You deal with each emergency as it arises. Would you run 3 children over because you might drive into a ditch if you swerve? No. And then you might die? Right, so why expect an AV to be any different.

I wouldn't kill myself to save others that were clearly in the wrong.... Why would I allow a machine to force that decision into my life?

Seriously, you would rather run into 3 children who were in the road than risk driving into a ditch? Good grief. You cannot imagine how deep my contempt for you is at this point. And as for downvoting me, --- it's a badge of honor. ....... By the way, in many legislatures, cars do not have a right of way.

Personal comments removed

He downvotes anyone disagreing with him

Or maybe he´s killed so many children to save his own life from crashing at a residential street (40-50kmh... ) he don´t want to talk about this, and that´s the reason he´s afraid an AV will make a different decision and will kill him

[Andres125sx]

...I´m wondering about the reason I took the effort to reply. You will never decide to kill yourself or kill others, you will always try to avoid the accident and kill none...

It's a conundrum Andres, an interesting conundrum, that's why you replied... Yes it's morbid, but it's a conundrum that bears thinking about.., no?
I've implied in the past that there's an actuary out there who has accounted for the outcomes of this circumstance (and every conceivable circumstance for that matter...) The manufacturers have been running the numbers forever. Remember the Pinto?

The Pinto? No, no idea what´s that

But I disagree, it´s not a conundrum, it´s demagogy. As I´ve explained when we face a critical situation like running over 3 children, nobody is telling us "you will die if you do anything but running over them". That situation is pure demagogy, in that situation you only know for sure if you run over 3 children you will probably kill some, but you can´t know what will happen when making an evasive maneouver, so by instinct we will always try to avoid the accident. Even if there´s a wall or a tree at both sides

If anyone is prepared to run over 3 children on purpose just to avoid crashing his car with whatever, IMHO that person should be retired from the roads. Specially because if he present this scenario like someone will die and you must choose that´s probably because he was driving way way too fast at a residential zone. If not, then you can always brake at least enough to prevent the accident from being fatal for anyone

[AJI]

...I´m wondering about the reason I took the effort to reply. You will never decide to kill yourself or kill others, you will always try to avoid the accident and kill none...

It's a conundrum Andres, an interesting conundrum, that's why you replied... Yes it's morbid, but it's a conundrum that bears thinking about.., no?
I've implied in the past that there's an actuary out there who has accounted for the outcomes of this circumstance (and every conceivable circumstance for that matter...) The manufacturers have been running the numbers forever. Remember the Pinto?

The Pinto? No, no idea what´s that

But I disagree, it´s not a conundrum, it´s demagogy. As I´ve explained when we face a critical situation like running over 3 children, nobody is telling us "you will die if you do anything but running over them". That situation is pure demagogy, in that situation you only know for sure if you run over 3 children you will probably kill some, but you can´t know what will happen when making an evasive maneouver, so by instinct we will always try to avoid the accident. Even if there´s a wall or a tree at both sides

If anyone is prepared to run over 3 children on purpose just to avoid crashing his car with whatever, IMHO that person should be retired from the roads. Specially because if he present this scenario like someone will die and you must choose that´s probably because he was driving way way too fast at a residential zone. If not, then you can always brake at least enough to prevent the accident from being fatal for anyone

This conversation has run a little off the rails... I apologise to all if it was in any way my fault.

I've been following this thread since strad started it. I haven't commented much of late, but I still follow it quite closely.
The proposition of AV's, in broad brush strokes, has always been a question of:

a) whether we can actually achieve it (from technological and integration POV's)

and

b) who is responsible if a human is removed from the decision making process of driving

It has been suggested many times in this thread that AV's are technologically unachievable (to a proper safety standard for a reasonable cost), and the question of who is responsible for the decisions of the vehicle has been debated both from a legal POV and also simply by suggesting that an AV can never drive as well as a human.

While I disagree with the suggestions that we are technically incapable of producing an AV and that humans can always do something better than a machine, the legal responsibility argument is a tricky one. A conundrum... There’s no space for emotion in the legal responsibility debate, it will be decided on numbers.

The Ford Pinto disgrace was my example. The numbers were calculated and Ford chose death and payouts to those affected over recalling the car and modifying the design. It was simply the cheaper option...

[Zynerji]

Seriously, you would rather run into 3 children who were in the road than risk driving into a ditch? Good grief. You cannot imagine how deep my contempt for you is at this point. And as for downvoting me, --- it's a badge of honor. You are an utterly miserable specimen. By the way, in many legislatures, cars do not have a right of way.

What about my children becoming fatherless? That rational self-interest bothers you so much?

It’s interesting you mention rational, the part of your brain that would make the decision under consideration is not rational. It’s autonomous, basing its decisions on prejudice.

I use prejudice in its pure form. It bases on prior knowledge and experience. Some of that is very prior, coded in DNA and it’s manifestation in the brains structures. Some of it is taught to it by the rational part of the brain by repeated stimulation. This is why some elements of driving are a “skill”. When we first start everything is rational, we have to teach the brain, slowly because rational thought is slow, the movements that result in the car behaviours we want. This is the one of the things the autonomous car is trying to simulate and it is taught to it in a combination of offline training and online refinement, potentially it is better than the average human at this.

Once that skill is in place, further skills can be taught around situational awareness. Only a proportion of these get to the autonomous level, dependant on the focus the rational mind gives to them. This is where AVs are at a disadvantage. Humans can use similar experiences to load up the autonomous part of the brain. So moving about is generic and one can learn from crawling, walking and running. As is understanding an classifying the visual world and the objects in them.

It’s not clear that your rational choice in this scenario could be transferred to your irrational autonomous brain. 2 or 3 instances a month of your muttering under your breath about ill-disciplined children may or may not do it. But of course you may have used other experiences to train your mind so the behaviour you want may come about based on prejudices you’ve inherited or learned about the people about you.

Whatever happens in such a situation, if you survive your rational brain will take plenty of time to fit the behaviour of your autonomous brain to your preferred model.

I don't really see anyone's issue here. I have a responsibility to myself to preserve my life, marriage, and bloodline. I have no concern over other's liberal view that I "should choose" to die to save another's bloodline when they are grossly negligent, because that is idiocy.

I would swerve to avoid catastrophe, but I would not drive off of a cliff to prevent it (an AV might), and I will not apologize for the fact that self preservation is my "prejudice".

Judge me how you will, my responsibilities in life will still be upheld. I already sleep well with 33 confirmed kills from my time serving as a US Marine. My rational brain processes these after effects very simply and efficiently.

[Greg Locock]

You are changing the argument. You said you'd rather run into 3 children than risk driving into a ditch.

[roon]

Can we throw in a ramp and a fire hoop? Might happen.

One side effect of AV use may be the degradation of one's driving ability if one becomes too reliant upon automation. A person may go years from their driving test until finally faced with a situation the computer cannot resolve.

Tesla, best AV attempt in the near term, admit that lack of road paint lines is a challenge. Snow, fog, heavy rain, country roads without clear edges. Situations may present where deference to a human driver will be needed if the vehicle is to be moved.

[Andres125sx]

What about my children becoming fatherless? That rational self-interest bothers you so much?

It’s interesting you mention rational, the part of your brain that would make the decision under consideration is not rational. It’s autonomous, basing its decisions on prejudice.

I use prejudice in its pure form. It bases on prior knowledge and experience. Some of that is very prior, coded in DNA and it’s manifestation in the brains structures. Some of it is taught to it by the rational part of the brain by repeated stimulation. This is why some elements of driving are a “skill”. When we first start everything is rational, we have to teach the brain, slowly because rational thought is slow, the movements that result in the car behaviours we want. This is the one of the things the autonomous car is trying to simulate and it is taught to it in a combination of offline training and online refinement, potentially it is better than the average human at this.

Once that skill is in place, further skills can be taught around situational awareness. Only a proportion of these get to the autonomous level, dependant on the focus the rational mind gives to them. This is where AVs are at a disadvantage. Humans can use similar experiences to load up the autonomous part of the brain. So moving about is generic and one can learn from crawling, walking and running. As is understanding an classifying the visual world and the objects in them.

It’s not clear that your rational choice in this scenario could be transferred to your irrational autonomous brain. 2 or 3 instances a month of your muttering under your breath about ill-disciplined children may or may not do it. But of course you may have used other experiences to train your mind so the behaviour you want may come about based on prejudices you’ve inherited or learned about the people about you.

Whatever happens in such a situation, if you survive your rational brain will take plenty of time to fit the behaviour of your autonomous brain to your preferred model.

I don't really see anyone's issue here. I have a responsibility to myself to preserve my life, marriage, and bloodline. I have no concern over other's liberal view that I "should choose" to die to save another's bloodline when they are grossly negligent, because that is idiocy.

I would swerve to avoid catastrophe, but I would not drive off of a cliff to prevent it (an AV might), and I will not apologize for the fact that self preservation is my "prejudice".

Judge me how you will, my responsibilities in life will still be upheld. I already sleep well with 33 confirmed kills from my time serving as a US Marine. My rational brain processes these after effects very simply and efficiently.

What´s the reason you assume an AV might drive you off of a cliff? Solidifying prejudices maybe?

Driving you off of a cliff is killing you, running over someone might be, or might be not, so the scenario is completely different to the previous discusssion. You´re moving the goal posts


Anycase, people do not fall from the sky, you keep assuming scenarios wich are not real (running over someone or drivingg off of a cliff). Speed limits have a reason, and it is to have time to react in case of an emergency. That scenario will only happen if you´re driving way too fast, and AVs will never drive too fast, so your hypotetic scenario is unreal. In that scenario the AVs will make an emergency stop, braking before running over anyone, or in worst case scenario hitting them when the car is slow enough to not kill anyone, and to not drive you off of a cliff

[NathanE]

One observation on learning applied to humans vs AV. In an AV environment learning can be pooled in a much more effective way than in a Human one. This should accelerate safety development significantly. I have probably driven around 1m miles in 30 years to accumulate my experience. This target will be reached in days with an autonomous fleet.

In addition, how many drivers have any real understanding of limit handling or other "exceptional" situations? Having raced for a number of years and spent significant time developing skills on skid pans and kick plates I know that road driving never have me any real sense of driving at or close to the limits of adhesion. In a pooled knowledge AV environment all vehicles will have a better starting point capability.

[henry]

One observation on learning applied to humans vs AV. In an AV environment learning can be pooled in a much more effective way than in a Human one. This should accelerate safety development significantly. I have probably driven around 1m miles in 30 years to accumulate my experience. This target will be reached in days with an autonomous fleet.

In addition, how many drivers have any real understanding of limit handling or other "exceptional" situations? Having raced for a number of years and spent significant time developing skills on skid pans and kick plates I know that road driving never have me any real sense of driving at or close to the limits of adhesion. In a pooled knowledge AV environment all vehicles will have a better starting point capability.

That’s a good point, particular for the car control part of the problem. I can understand why Tesla crow about the miles of experience they are building up. However this is only part of the equation. I can’t quite match your mileage but I don’t think it matters for car control, I believe that I reached the peak of my car control abilities, admittedly not high, after a very small subset of those miles.

The more difficult part has been learning situational awareness. I think this is a much more difficult problem for AVs. When a human encounters a situation they can use the rational, rather than autonomous, part of their mind to analyse the situation ,sometimes there and then but more often after the situation has passed. They can then, if they are interested, figure out a revision to the rules they want the autonomous part of the brain to use to deal with such situations “better”.

So AVs, Tesla for instance, gather data on situations they encounter but where is the rational mind that allows them to classify situations and then propose improvements to behaviour? As I understand it the neural networks are trained by being presented with a classified set of data of characteristics that might be important and outcomes that are acceptable. For identifying objects in a scene the base level comes from group of humans marking objects and classifying them. The NN can then learn characteristics of the objects, such as their likely trajectories.

What happens if a Tesla sees a two wheeled vehicle and misclassifies it as a motorcycle? Will that misclassification ever be corrected?

As time goes by and the libraries of situational data increase I can see how fleets of AVs might help reinforce learning making decisions more and more reliable, but I think the difficult thing is deciding on the structure of the libraries and I think that’s still going to be the preserve of human minds.