[katieHaylor (OP)]

This feedback on discussing the ethics of driverless cars has come in from Paul.

I enjoy your podcast, but I almost hit delete when you started talking about ethics for driverless cars.

I keep seeing this drivel.  Could you not have taken a step back and asked someone in the industry (not the philosophy industry, but the autonomous car industry) before wasting your airtime and advertising another of these absurd papers that keep coming out on this topic?

1. There is not a question of responsibility.  The existing individual-driver insurance industry would like to think there was, and the car companies who would of course like to push away responsibility, but the reality is obvious: The car companies will be entirely responsible for everything which the car does.  This will be the case because they will be the manufacturer of both the hardware and the decisions the car makes.  They are the driver and the car.  They may sue suppliers of faulty parts, but only after settling lawsuits themselves.  They bought the parts.  They QA'd the systems.  They monitored or negligently did not monitor the systems in use.  (When your left-hand secondary camera gets 20% dim, your car will pull over and stop, waiting for a tow truck, because the car company will not let it move forward because that would be negligent behaviour.)

2. You won't get to drive your car because your insurance won't let you.  It's not necessary for you to drive, and therefore anything you hit while driving is due not to an error but to negligence.  This would incur multiplied damages.  The whole insurance model breaks down in the face of this.  It will be too expensive for you to touch a steering wheel.  Again, all responsibility will be held by the car manufacturer because they will do all the driving.

3. When conditions make it unsafe to drive, the car won't drive.  See #1.  The car is driving, and it (as an extension of its manufacturer) won't risk the negligence, as driving when you cannot be sure what is ahead IS negligence.  The example given of a fog clearing is absurd.  The car knows how far it can see.  It knows how quickly it can brake.  It will only drive as fast as is safe.  The decision on who to hit will never occur.  When it does, the manufacturer will have made a mistake and will be sued.  They will then fix the software which caused that situation.  (Sudden jump outs of humans the car did not see do not qualify either.  The car will slow when passing hidden areas -- See again #1.  You would need to fire people at the area in front of the car with a trebuchet to construct the situation these philosophers propose.)

4. The whole idea assumes AI is magic.  "Do you hit the woman or the child or the dog" really translates to "do you hit the moving blob/box about 5' tall, or the one 3' tall, or the 1' tall one which happens to be longer than it is tall."  WOMAN?  A car won't be able to tell the sex of a human for 20 years or more.  It won't be able to tell the difference between a human and a manikin for almost as many.  This topic comes from people who simply don't understand how these systems currently work, or how computers currently work.

5. What these absurd morality systems ask for is massive in complexity -- far more complex than driving a car.  So we would have to power these systems, tote the weight of their hardware needs, spend time writing these systems, and worst: live with the bugs all this complex code adds to the system.  If you could write such systems, they would diminish the safety of cars -- bugs kill people in this context.  They would also add expense and thus reduce the uptake of autonomous cars, which would make us all less safe since autonomous cars will reduce traffic deaths to a vanishing number and reduce emissions which would improve respiratory and other body-system health for the whole planet.

6. Let's say you COULD put work into the handling of results of a collision.  Mitigation of damage at this point reduces the urgency of work to keep those situations from happening in the first place.  These absurd, imaginary systems would, again, make us less safe.

The affect of all the above together: These people who write on this subject are parasites.  They want to get published and use this topic to get attention.  They either know that their work is destructive or haven't done the simplest research necessary to see that it is.  They do not have the good of the world at heart.

When you give them attention without doing the minimal research yourself, you are part of the problem.

[katieHaylor (OP)]

Jeff sent in some further feedback on this:


The interview on the ethics of driverless cars was full of errors, supported statement and fanciful prediction—none of which are supported by scientific evidence.
 
Peter stated that 90 percent of all accidents are human caused.  First, humans are at the controls 100 percent of the time and therefore are always partially responsible unless 10 percent of the cars are already driverless.
 
Second, Peter is assuming, with no evidence, that humans are 100 percent responsible, that is there are no objective external reasons for the accident, in all of the 90 percent of the accidents attributed to human causes.  That is flat out wrong.  It is impossible to know what the contributing factors are because the data are not collected and if they are, are not consistent across agencies.
 
The part that galled and should have been challenged was Peter's statement that driverless cars will reduce accidents by 90 percent, based on an unnamed study.  Peter is saying the driverless cars will eliminate 100 percent of so-called human caused accidents without causing any new forms of accidents, such as the computer confusing the break with accelerator or the driver ignoring the white truck because he assumes the computer can see it.  I did find a study that makes this assertion but it is a whitepaper from a US consulting firm not a rigorous scientific analysis (https://www.sciencealert.com/driverless-cars-could-reduce-traffic-fatalities-by-up-to-90-says-report).
 
There are too few data to even begin to estimate the impact driverless cars will have on accident rates. 

[chris]

This topic was also raised by a listener previously, in response to an earlier item about the subject of driverless cars:
https://www.thenakedscientists.com/forum/index.php?topic=69317.0

[alancalverd]

Child/dog/moose runs into road, or cyclist falls over. However good the driver or computer, the laws of physics do not permit instant deceleration. Swerve or skid? It doesn't matter which, someone will get hurt. But who is to blame? 

Cow in road. Blow horn. Maybe dead. Wait for it to revive or drive around it?  Sorry, it isn't a dead cow but a live one in a trailer, waiting for the school crossing to clear. Or maybe a dead one in a trailer. Bugger, wrong choice. 

Black ice can form on the top of a hump-back bridge, but the conditions are fairly critical. Experienced driver recognises the possibility from a knowledge of recent and present weather, and anticipates the need to brake because there may be stationary traffic on the other side (it's rush hour/school run/dog show), so slows down before mounting the bridge. That's a lot for a computer to digest, let alone acquire. 40 mph speed limit, clear cold morning, no visible traffic ahead, so off we go.....

Wet road. No problem. But if it's wet and the drains are clogged, the puddles will be very deep and we anticipate heavy drag. Is a puddle a problem (probably not if there are lots of puddles) or a symptom of a new pothole.(if all the others have dried out)? Difficult sort of decisions to program. 

Fact is that 100% of completely uneventful journeys are caused by human drivers.

[chris]

Fact is that 100% of completely uneventful journeys are caused by human drivers.

I like that turn of phrase!

[evan_au]

Could you not have taken a step back and asked someone in the industry

I would love to hear from someone working in the industry.

But I suspect that these people cannot speak "on the record" because they have non-disclosure agreements, and the commercial and legal issues are critical and could be harmed by an incidental comment.

So you are left with talking to academics or people who are a bit removed from the "hands-on" work, as they are not so constrained by employment contracts or commercial risks.

we would have to power these systems, tote the weight of their hardware needs

I heard some rough estimates that the human brain consumes about 25% of the human basal metabolic rate - let's say 40W.
- Humans can hold a conversation while driving, so let's say that we use 50% of their brain power consumption for driving, or 20W.

Apparently todays typical self-driving car draws about 1kW to power the analysis of image, LIDAR and radar returns. And I suspect that it takes up more than half of the 1.5 liters volume of a human brain...

[alancalverd]

Human - nonhuman collision. Human claims "the machine acted deliberately to cause the collision". Lawyers ask whether an adaptive program can be considered "deliberate" (dangerous driving - serious offence - custodial sentence) or merely "defectively reactive" (driving without due care and attention - lesser offence - fine and temporary loss of licence). People die, lawyers win, truckers strike because "if you can't hold a machine liable, then we are uninsurable", everyone starves .   

[alancalverd]

BBC2 today showed a neat clip of a potentially driverless truck on a US highway. Good lanekeeping and speed control - almost as clever as Trevithick's railway - but as with the Audi car shown last week on an autobahn, it is completely useless in a goods yard or building site with no lane markings. So the US team have added a trick: when baffled by urban traffic or laneless spaces, the truck calls up an expert remote-control driver who takes over by radio control and video camera, just like a drone. Which is fine until the goods marshaller starts shouting commands at an empty cab ("Go to gate 13" but my video camera doesn't know where the gate numbers are shown....) or two trucks run into trouble at the same time whilst one of the expert controllers is having a toilet break. And if my goods yard has poor radio reception, I'll end up with a dozen dead trucks..... Still, it will save on truckers' hotel bills and broken marriages, even if every other industry grinds to a halt..