Do you feel safer with self-driving cars on the road?

[Spinnor]

When self driving motorcycles (why would we need such things) can better John McGuinness's best time at the Isle of Man we will know humans are redundant?

 

[jack action]

I highly doubt technology reduces the work load of human beings:

Since the 1960s, the consensus among anthropologists, historians, and sociologists has been that early hunter-gatherer societies enjoyed more leisure time than is permitted by capitalist and agrarian societies; for instance, one camp of !Kung Bushmen was estimated to work two-and-a-half days per week, at around 6 hours a day. Aggregated comparisons show that on average the working day was less than five hours.

Subsequent studies in the 1970s examined the Machiguenga of the Upper Amazon and the Kayapo of northern Brazil. These studies expanded the definition of work beyond purely hunting-gathering activities, but the overall average across the hunter-gatherer societies he studied was still below 4.86 hours, while the maximum was below 8 hours. Popular perception is still aligned with the old academic consensus that hunter-gatherers worked far in excess of modern humans' forty-hour week.

When someone tries to convince me that we have more today because we are more productive, I tend to believe we have more because we work more.

We are far from «not having enough work to do», we are working much much more than we need to, to just stay alive.

Worst case scenario (All work necessary to maintain all humans alive done by machines), everyone can still study whatever they want to better understand our surroundings. Even if it just to read whatever «thinking machines» have already chew up for us. That will keep anyone busy for a lifetime.

 

[_PJ_]

I absolutely feel safer with self driving cars.
Humans are "flawed" for thes emain reasons:

1) Choosing to disregard convention, law or logic (There's a fair amount of circumstance and such here, and a wealth of specifics, but the key point is the "CHOICE" - As meant distinct form a computer's (reagrdles s how neural-net and facotrs may influce the alogiorthms)
Put a human in exactly the same situation, they will not necessarily perform the same, certainly different humans will perform differently and likely inconsistently. Given the exact same input parameters, different autonomous vehicles (at least, those using the same software base) can be expected to perform consistently

2) Emotion and distraction - being angry, happy,m upset, anxious, stressed, tired, kids yelling in the back, atttractive pedestrian, phone ringing, sudden UFO sighting can all distract humans whose brains have so much to process ever y instant of driving a car. Of course, modern cars take such a lot of the pressure off with the sensors and systems as well as being improved for safety all the time - but these are all aspects that autonomous vehicles benefit from too.

3) Unpredictability - (Top my knowledge) all the accidents involving self-driving cars have been a result of human introduced chaotic influence that simply would not exist were ALL the cars autonomous. Although I personally feel like some centralised hivemind (cybersecurity issues notwithstanding) that could help cooridnate EVERY autonomous vehicle would be a great advantage - the various different companies developing their own tech and increasing competition might limit the availability or perceived viability of such a thing. However, a scenario at least with ONLY autonomous vehicles and NO human drivers I am sure would always be safer than human drivers. (It also might help combat crime without getaway vehicles)

As such, autonomous cars are in my view, ALWAYS better.

The problem with establishing data, though is that statistics are horribly inequivalent.
Just as "it's safer to fly than drive" doesn't take into account the proportionality dfifferences in regards to passengers per cars, metres of car per square metre of road (airline flight paths), time spent in the vehicle,

Statistics relating to autonomous cars are similarly disproportionate

I am, however, convinced that they will lead to a much safer and more environmentally preferable road transport system - if autonomous vehicles are more used as public transport service (they woudn't even need a set, established repetitive route, but function more like generally shared uber), this increases the efficiencies all round.

___
I think I might have mentioned this before, but there are two difficult areaa.

1)
In order to ascertain the decision making process and maximise safety/prioritise human life, there are sadly always going to be situations where for whatever reason, the only options possible will result in some accident. Naturally, the focus is to opt for the best case scenario, the best case defined ... by public consensus.
A huge data collection endeavour tests volunteers in giving them choices, such as (the cases are all that there is no opportunity to brake in time etc. and binary options remain the only possibilities)
Hitting a pedestrian with a baby when the car is full - Or hitting 3 people one of whom is elderly, the other two, a young couple-
Hitting a brick wall that will harm the occupant of the car, or hitting a pedestrian.
etc.

None of these are easy decisions of course, and there's a huge amount of factors (a young couple can go on to have many children - the senior citizens might be likely to die form injury than someone more youthful, a disabled person might be less able to move out of the way themselves) so on, so forth. But these situations (although one truly hopes will be rare if they happen at all) are necessary in forming a "moral code" to the decision making process for the AI.
I honestly, don't think there is a truly ideal solution to approaching this, and of course, it's necessarily a good thing that there IS some moral (by which I suppose I really mean, a consideration of the value of life and wellbeing) aspect to that process - but the robotic binary application "Do you kill the mother or the pensioner" kind of presentation is kinda unnerving.

2)
in the event of an accident, in which (for the sake of argument/example/case) the accident is caused by such a choice as described above in which, although unavoidable that "someone" would be harmed, the individual thus harmed was done so as direct result of the circumstances under which the decision making algorithm selected a decision to harm that individual.

Of course, the algorithm doesn't know the individual personally. But where does the accoutnabiltiy lie?
It was just an accident, sure - but what if this happens a lot? Surely there eneds to be some level of responsibility.
The algorithm itself is just that, you can't trial or jail a software. Extremely unlikely the exact specifics could be attributed to any individual programmer. More likely,it was a group effort simply converting the data from (1) into digital form where perhaps a 1 instead of a zero (simplifying the option) would have meant more people died - who could ever know (and maybe not knowing the precise natur e of the individual numbers is a good thing) - I doubt such data would be manually entered at all, rather the algorithm designed to calculate and select the best result from the database, given certain parameters for which parameter settings such as "human life= maximum" could hardly be argued with
So what about the designers of the systems? Are they to blame? Again, I would find it hard to point any finger for any individual incident - but after repeated situations, maybe I'd expect someone to step down - but this raises more problems: Assault and such negligence can be criminal. Can a company exec be jailed because of a decision made by an algorithm? Should they be?
Further, and finally, at what point is it considered that "too many" accidents highlight failures or problems?

 

[jack action]

@_PJ_ :

Very long and complete thought. But I must disagree, as this kind of thinking have all the typical signs of irrational fears. The strongest indicator is beginning this thought process with «Humans are flawed». There are not. They are the results of millions of years of evolution. Whatever characteristic passed the test of time, it's there for a reason. This is true for emotions, curiosity (which you refer to as being distracted) or choosing to disregard convention, law or logic (it is needed to adapt oneself to the unpredictability of life).

Speaking of predictability, it is not always the best option. Predictability can often direct you to a certain death when something else changes.

When I read your post, I felt like listening to a preacher selling his religion. First, somehow, people are unfit to take decisions. But, fortunately, there is a god called AI that possesses all the answers, at least answers better than the ones people can come up with. Finally, there are those priests, that you refer to as designers of the systems, that will be the link between the people and this new god, our savior. Because, you know, the common man won't be able to understand this complex AI they so desperately need.

As a person with a scientific mind, I see someone building a driverless car and think to myself: «Cool! That looks like a fun toy!» (That's an emotion). Then curiosity kicks in: «I wonder how it works?» Maybe I'll follow thru, maybe I won't. But if someone starts selling (forcing?) this to me like it is the solution to all my problems (Do I really have problems?) and I should be obliged to have one, well, I don't find this so cool and fun anymore and that makes me feel sad and frustrated. That's another emotion, one that exists specifically to tell me something is wrong. Maybe there's nothing wrong because just like AI would be, this system of emotions developed for the last millions of years is not perfect in every situation. But, along with other tools I possess, I still trust it more than any new, unproven, tool invented by men.

Maybe I'm putting more intentions in what you are saying than you cared for. But to me, this is where this kind of speech is heading and I prefer to tame the wording at its infancy before I wake up with an authority forcing me to buy what, without any surprise, it sells.

So, once more, I don't feel safer with self-driving cars on the road (nor do I feel safer NOT having self-driving cars on the road). Although, I don't mind that some people choose to work on such project. I actually find that stimulating, the only downside being this talk about AI being the solution to all our problems.

 

[Dr. Courtney]

My feeling is that it's hard for any technology to compensate for the loss of safety introduced by inattentive drivers glued to their electronic devices. The inattentive drivers are unlikely to adopt the technology in sufficient numbers to improve my safety or that of my family in the areas we commonly drive.

Further, I think we're a few generations away from self-driving cars which are able to look into the other cars around them and assess the level of attention and possible impairment of the other drivers on the road. Odds are most of the focus in the first few generations will be on what the other vehicles are doing. I'm more worried about the other drivers than the other vehicles. When I see an inattentive or likely impaired driver, I give them lots of room, often even pulling over to let them pass and get well ahead of me before I continue. How long until self-driving cars are capable of this?

 

[Noisy Rhysling]

Watch "Car Crashes Time" on Youtube. A robot couldn't do worse.

 

[CWatters]

I voted no (not convinced yet.)

Most testing seems to be done on American roads or motorways. These are vastly different to the single track narrow roads of Devon and Cornwall in the UK. One car wide, no kerbs, no white lines, sharp bends, hedges right up to the edge of the road that meet above your head, people on horse back, live stock.

Where can I find the International driving test track for self driving cars?

 

[ibkev]

I vote yes, partly because of this quote:
"If I asked my customers what they wanted, they would have said a faster horse." - Henry Ford
A lot of the discussion in this thread has focussed around the safety consequences of autonomous vehicles being faster horses but I think the other possibilities are more interesting.

My crystal ball has never been that great but, I do imagine there will be a time when at least parts of large cities will become automated-vehicle-only zones. The whole traffic lights system is designed for use by people but what if that could be replaced by a networked traffic control system? I suspect cost and convenience will play a significant role here. What if choices for reducing commute times include massive expenditure widening roadways, with corresponding additional taxes/tolls/etc, or reserving a lane for automated cars who will be allowed to drive significantly faster? We saw with the advent of cellphone-based cameras and mp3 players that people will choose convenience over other factors much more often than anyone expected. There was large-scale adoption of these technologies long before quality caught up to the minimum standard of the devices they were replacing.

Then there's the idea that fleets of roving autonomous taxis/vans could dramatically reduce the need for personal ownership of vehicles. No body likes, getting oil changes or the feeling that they might be getting ripped off when they take their car in for repair, etc, etc.

Interestingly, security will be a big issue for these vehicles. It wouldn't be surprising to see kidnappings carried out by remotely commandeering a self driving car, or assassinations by remotely driving a car into on coming traffic, etc.

 

[Dr. Courtney]

Another consideration that comes to mind is that once one is exercising the highest level of care humanly possible, one's greatest odds of a collision then becomes fairly unusual circumstances like colliding with a deer. I've lived in a few places with lots and lots of deer and often regard deer as my greatest risk of a collision when driving in specific areas.

My hundreds of hours of deer hunting have given me an uncanny ability to recognize deer (and other wildlife) on the side of the road, as well as an uncanny ability to know when deer are more likely to be moving and attempting to cross the road. I'll often drive slower at these time and ask my passengers to help me keep a lookout for deer in or near the roadway. I doubt automated driving systems are likely to approach my human ability to spot and respond to the threats of deer near the roadways any time soon.

Other factors that add danger to driving like black ice, snow, leaves, and other conditions relating to weather and road conditions are going to be much harder for automated systems to recognize and adjust to compared with attentive and experienced drivers. The designers of these systems are most likely comparing them with 50th percentile American drivers. That's a pretty low bar. Based on miles driven since our last accident, my wife and I are likely well above the 95th percentile. So that is the level we have in mind when considering automated systems.

 

[cosmik debris]

Based on miles driven since our last accident, my wife and I are likely well above the 95th percentile. So that is the level we have in mind when considering automated systems.

When it comes to driving everyone is above the 95th percentile. :-)

Cheers

 

[Dr. Courtney]

When it comes to driving everyone is above the 95th percentile. :-)

Cheers

Yes, if you allow them to self-rate. But there are several possible objective measures: miles between accidents under various driving conditions, occasions of distracted driving, occasions of driving under the influence, etc. Vehicle data recorders can do even better: sudden stops and starts, occasions of excess speed, etc. The technology will soon be there to also keep track of occasions of following too closely. High risk driving behaviors and outcomes are no mystery. The data exists in most cases for objective assessments of driving risks.

 

[Spinnor]

I doubt automated driving systems are likely to approach my human ability to spot and respond to the threats of deer near the roadways any time soon.

I don't know how your vision is at night, but mine stinks. How is your infrared vision? Don't underestimate machines of the near future.

 

[Greg Bernhardt]

Other factors that add danger to driving like black ice, snow, leaves, and other conditions relating to weather and road conditions are going to be much harder for automated systems to recognize and adjust to compared with attentive and experienced drivers.

I think something to consider is just how many drivers are attentive and "experienced"? I live in a mid major city and every day I shake my head at the irresponsible, dangerous and idiotic driving I see.

 

[Dr. Courtney]

I don't know how your vision is at night, but mine stinks. How is your infrared vision? Don't underestimate machines of the near future.

View attachment 215077

Thermal imaging provides much better contrast detecting deer when the background is significantly colder than the deer. It is a common method of doing aerial deer population estimates in the winter in the mid-west. It is less reliable in the summer. Judging from the contrast between the deer and the road sign in the picture, the attached picture is a thermal image certainly taken on a cold night. It will not work reliably most of the year in places like my home state of Louisiana. If the proposed mechanism uses a light source emitted from the car, then the deer will likely also be illuminated by the headlights.

The photo is also misleading in that deer are easy for drivers to spot once they are in the roadway illuminated by headlights. Most car deer collisions occur from deer jump out from the side of the road at the last instant, not when they are standing in the road for some time. To be comparable with the detection skills of a hunter with my experience, an automated deer detection technology needs to work for deer within 50 feet or so of the edge of the road and needs to also be effective when deer are mostly obscured by cover. One study found that infrared deer detection drops to an accuracy of less than 50% when 40% or more of the deer's body is blocked by an occluding object (See p. 32 of https://arc-solutions.org/wp-conten...u-Thermal-Image-Based-Deer-Detection-2013.pdf )

Also notable is that even a $1500 thermal camera has a low effective shutter speed and is not expected to provide clear images if the camera or the target are moving. The proposed infrared deer detection system uses fixed roadside cameras, not vehicle mounted infrared camera.

Deer are actually much easier to spot at night for drivers who know what they are doing. The headlights illuminate their eyes which shine right back toward the car like bright reflectors. All that needs to be unoccluded is the head. They are hardest to see in dim daylight: raining, dawn, and dusk.

I don't think I'd mind a deer detection system in an automobile, but I'd prefer an approach that alerted the driver rather than one that automatically implemented evasive action.

 

[Spinnor]

I don't think I'd mind a deer detection system in an automobile, but I'd prefer an approach that alerted the driver rather than one that automatically implemented evasive action.

Maybe there are other less costly sensors to detect deer? The price of technology almost always goes down.

Where I drive at night there is almost always oncoming traffic and I am nearly blinded by the oncoming lights. I drive under the speed limit. My greatest fear while driving at night is not seeing someone in dark clothing walking right next to the side of the road while blinded by oncoming traffic. I think about it all the time. I should probably slow down even more. At night self driving cars are probably much safer then me already and they will just keep on getting better as software and sensors improve. Soon I think they will better most drivers and in time every driver.

There are some intersections where it is more dangerous to proceed, say at a stop sign where it is hard to see traffic coming around a bend or where vegetation makes it harder to see. I can imagine simple sensors that alert your car to oncoming and partially hidden traffic, kind of like the transponders aircraft use.

Lives will be saved, jobs will be lost, we live in interesting times.

 

[Dr. Courtney]

Maybe there are other less costly sensors to detect deer? The price of technology almost always goes down.

I was one of a very small group who had permission to shoot deer at night in the late 1990s for crop damage control. Consequently, I have kept close watch on infrared detection technologies for the past 20 years. There has been remarkable improvement in technologies for deer that are moving slow - but just like all low light imaging methods, detection is much easier at slow shutter times (> 1/30 second) that do OK in static situations than for fast shutter times (< 1/100 second) needed when the camera and/or the subject is moving. Given the relative motion of the vehicle, shutter times are going to need to be between 1/100 sec and 1/4000 sec to reliably detect deer.

Nothing close to this exists currently. Infrared advancements in the next 20 years will need to outpace advancements of the past 20 years by about a factor of 10 for deer detection from a moving car to approach what an attentive and experienced driver can do. So not only does the price need to come down, the technology needs to get a lot better.

The deer recognition problem is really only an illustrative subset of the issues. There are analogous challenges recognizing other threats. Real time video recognition and response systems are a challenge with threats moving toward the roadway. How will a system distinguish between a motor cycle or an adult cyclist heading toward an intersection and likely to stop from a child on a bike heading down a driveway and less likely to stop? A human can make this determination and apply the brakes if needed in a fraction of a second, even if most of the possible threat is occluded by intervening vehicles and only a head is visible above car roofs or through car windows. (One of my biggest concerns as a driver is a child on a bike, skateboard, or big wheel popping out from behind a line of cars in the driveway into the roadway. My habit is to watch carefully through the parked car windows, over their roofs, and between the parked cars for this possibility. My wife and children immediately point out young children playing in and near driveways even 50-100 feet from the road, so that I can slow my speed and exercise all due care.)

Finally, there is no need to jump to self-driving cars to take advantage of these purported automated detection technologies and see how well they really work. Technologies to alert the driver to threats are already becoming available. They are fairly good at recognizing unoccluded threats reasonably within a few feet of the roadway/vehicle path, relatively close to the car, and unoccluded. They are nowhere near as good as an experienced attentive human at recognizing mostly occluded threats further from the roadway, more than 100 feet in front of the car at the roadside, and more than 50% occluded by intervening objects. The redundant system of attentive and experienced human vision and recognition PLUS the best in automated detection will likely remain superior to only automated detection for a few more decades.

 

[StatGuy2000]

I voted "No Opinion", not because I do not have an opinion on this issue, but I feel that self-driving vehicles are still in its infancy, so it is difficult for me to assess whether, at this preliminary stage, the vehicles would in fact be safer. I should also add that it isn't entirely clear to me how the technology would be adopted in the broader public beyond the current prototypes.

My own speculation is that the first places we'll see self-driving vehicles would be along routes that are relatively fixed (e.g. train stops, subway cars, streetcars, buses on fixed routes), for the simple fact that the machine learning algorithms will have a fairly steady stream of relatively reliable "predictable" data. I think it would be quite a while before we'll see a more widespread adoption to other passenger vehicles.

 

[Jetflyer0]

I won't really feel safe with self-driving cars until they are all ai controlled and can communicate all their data.

 

[Greg Bernhardt]

I won't really feel safe with self-driving cars until they are all ai controlled and can communicate all their data.

Which self driving cars are not ai controlled and don’t communicate data? Do you mean until all cars are self driving?

 

[Jetflyer0]

Which self driving cars are not ai controlled and don’t communicate data? Do you mean until all cars are self driving?

Yes, I meant once they are all self driving and can tell the routes of other cars that may affect their own route then adjust accordingly. Search self-driving cars intersection gif for an idea of what I mean

 

[Greg Bernhardt]

Yes, I meant once they are all self driving and can tell the routes of other cars that may affect their own route then adjust accordingly. Search self-driving cars intersection gif for an idea of what I mean

Totality won't happen for a very long time. But can you not agree as more diver-less cars take to the road the safer it gets?

 

[Jetflyer0]

Totality won't happen for a very long time. But can you not agree as more diver-less cars take to the road the safer it gets?

Yeah, the less unpredictable driving patterns there are, the safer the roads will become

 

[Stavros Kiri]

I won't really feel safe with self-driving cars until they are all ai controlled and can communicate all their data.

Which self driving cars are not ai controlled and don’t communicate data? Do you mean until all cars are self driving?

Addressed before: (...)

Yes I do feel kind of safe, and I do want to kind of want to trust them, but I will feel a lot safer and almost trust them completely when/if, perhaps in the near future (by 2020 or so), as part of the internet of things, all(?) cars will be self-driving and controlled/coordinated via a superfast 5G mobile net, to avoid all accidents (in highways, city and rural roads etc.) ...

Maybe it will take longer though, to fully practically apply ... (2025 or so ...)

 

[Spinnor]

Night Vision for Self-Driving Cars,

"Elon Musk famously thinks that cars can be made to drive themselves without relying on expensive laser-ranging lidars. But while Tesla is moving ahead with one fewer sensor than most self-driving car companies, a new startup wants them to add yet another—an infrared camera."

AdaSky is developing a far infrared thermal camera called Viper that it says can expand the conditions that automated cars will be able to operate in, and improve safety.

“Today’s sensors are not good enough for fully self-driving cars and that’s where we come in,” says Dror Meiri, vice president of business development at AdaSky. “We think infrared (IR) technology can bridge the gap from Level 3 all the way to Levels 4 and 5.”

...AdaSky says its system is currently being evaluated by several car companies and suppliers, and the startup hopes to start mass production in 2020 or 2021.

From, https://spectrum.ieee.org/cars-that...driving/do-selfdriving-cars-need-night-vision

From, https://www.google.com/search?q=how...ome..69i57.10583j0j8&sourceid=chrome&ie=UTF-8

Self driving cars don't have to be better then the best driver to save lives, they just need to be better then the average distracted, sleepy, texting, slightly drunk Joe or Jill.

Watch Yamaha's Humanoid Robot Ride a Motorcycle Around a Racetrack

From,https://spectrum.ieee.org/cars-that...id-robot-ride-a-motorcycle-around-a-racetrack

What will they think of next?

 

[OCR]

...while Tesla is moving ahead...

Just that one, so far... right ??

 

[Spinnor]

Just that one, so far... right ??

Just one, what?

 

[fluidistic]

I've almost been hit intentionally by a car (I was on a bike in the bicyle path, impossible to bother any car) 2 days ago. I lost an aunt about 6 years ago because an adolescent decided to suicide himself on the highway with a front to front collision. Overall cars are less secure than train, planes, etc.
A car is a weapon and from what I read and see, humans are pretty bad at using it. Many do not care about other's lives.
I'd prefer all cars driven by AI. I'd feel more secure, yes.

 

[Spinnor]

On a good day. Notice the little "foot" shuffle at 30 seconds and the blooper at 50 seconds..



Something a bit scary about the above.

On other days,



From, https://www.newscientist.com/articl...n-dynamics-robot-do-box-jumps-and-back-flips/

Next they will be ridding bicycles.

 

[jack action]

A car is a weapon and from what I read and see, humans are pretty bad at using it. Many do not care about other's lives.
I'd prefer all cars driven by AI. I'd feel more secure, yes.

I understand your fear about a human driving a car. But what makes you think it is better with AI?

If you would have almost been hit intentionally by a car driven by AI 2 days ago, or that your aunt would have been in an accident involving AI malfunction instead of an adolescent "malfunction", would it make you feel better?

NO matter what, a car will always be a "weapon" and the fact that "humans are pretty bad at using it" (IMHO, with the many millions trips done each day around the world, I think they have a pretty good record) will never mean AI is pretty good at using it.

One thing's for sure: AI will NEVER care about other's lives.

 

[OCR]

Just one, what?

a-head

A joke, that I knew wasn't... .