If all cars on the road were driven by automated A.I would there be more or less crashes?

[Actual_Criminal]

I seen a video just now on YouTube of a guy caught sleeping whilst his Tesla was doing 80 MPH on the highway. In the video, the officer at the end says "Never let technology take over." - Source video title: Tesla Was Doing 80 MPH While ‘Driver’ Slept: Cops.

 

Anyway, most accidents I have seen whilst working in the car insurance industry for a year (a few years back) were down to human error.

 

Obviously we are very far from everyone owning a Tesla (or similar) and even if everyone had one, I bet a large majority would prefer to drive it themselves. However, if all cars were running off automated A.I controlled on an individual basis rather than a network, would there be more or less accidents? (Everything from minor bumps to fatal collisions.)

 

What can a human do that the A.I can't do or can't do well enough (yet)?

 

EDIT: I have intentionally left out other vehicles like lorries, buses, motorcycles etc. So bear that in mind.

[Nathanpete]

With every normal car being AI controlled, WAY less. So unless there is malware or an attack, a homogenous system where cars from different maker's can communicate will result in zero collisions between two working AI controlled vehicles under almost all situations. 

[SignatureSigner]

Just now, Actual_Criminal said:

I seen a video just now on YouTube of a guy caught sleeping whilst his Tesla was doing 80 MPH on the highway. In the video, the officer at the end says "Never let technology take over." - Source video title: Tesla Was Doing 80 MPH While ‘Driver’ Slept: Cops.

 

Anyway, most accidents I have seen whilst working in the car insurance industry for a year (a few years back) were down to human error.

 

Obviously we are very far from everyone owning a Tesla (or similar) and even if everyone had one, I bet a large majority would prefer to drive it themselves. However, if all cars were running off automated A.I controlled on an individual basis rather than a network, would there be more or less accidents? (Everything from minor bumps to fatal collisions.)

 

What can a human do that the A.I can't do or can't do well enough (yet)?

From what I have seen AI may be able to react quicker to certain situations and follow rules put in place better but humans have a better ability to improvise and solve issues where ai would have to experience it multiple times to figure it out. Hypothetical situation time...

Lets say for some reason you have to get across the countryside where there are no maps. Your 'ai' controlled car would be unable to navigate this and find ways through but a human has the ability to see and route through the countryside. I Believe someday "AI" powered cars may be able to do this but we are a while out.

 

Overall if all cars on the road where AI powered and the programming in them was flawless I think there would be less accidents. But in reality who knows. What happens to your AI car when there is a snowstorm and your sensors for your car get clogged and it can't drive?

[Nathanpete]

The thing from what I have seen that AI cars struggle the most with is bizarre or extreme weather situations, or when things of certain colors and shapes are blocking a road. Like this: 

 

But to be frank, humans are idiots too and some still go out driving in extreme weather so they tied for possible accidents in situations of extreme or bizarre weather. Stuff like flooding, a landslide, or black ice. 

[Chris Pratt]

I think we have to reach a point where the AI is universal and not dependent on the car. What I'm talking about there, is instead of the AI being built into the car, it's built into the actual infrastructure, and the car acts as a receiver for commands.

 

As it is, the car's AI, even if it's programmed flawlessly, is entirely dependent on numerous sensors and a camera array. Something as simple as a camera lens getting too dirty because you hit a mud puddle, can severely diminish the AIs ability to effectively control the car.

 

With an infrastructure driven system, all the positions of all the cars on the road can be a known quantity, along with even road hazards. Additionally, the processing power can be near limitless, whereas you're always going to hit some limit of what can be thrown into a car.

[aDoomGuy]

Well considering we don't have true AI yet (all we can do now is program it) we'd be pretty screwed if all the cars was driving around fully autonomous.

[komar]

17 minutes ago, Nathanpete said:

But to be frank, humans are idiots too and some still go out driving in extreme weather so they tied for possible accidents in situations of extreme or bizarre weather. Stuff like flooding, a landslide, or black ice. 

That's because we gotta raise the standards who gets to drive and who doesnt. Pretty much any retard can get a license yet I still feel safer with them around me instead of autonomously driven cars that can easily crash/glitch/loose data for the environment/easily destroyed with a EMP.

[CerealExperimentsLain]

38 minutes ago, Nathanpete said:

The thing from what I have seen that AI cars struggle the most with is bizarre or extreme weather situations, or when things of certain colors and shapes are blocking a road. Like this: 

 

But to be frank, humans are idiots too and some still go out driving in extreme weather so they tied for possible accidents in situations of extreme or bizarre weather. Stuff like flooding, a landslide, or black ice. 

This is part of the problem, any computer system can be 'tricked' or 'trip up' on things a human never would.  You see that in any automated system right now too.  'Ah this machine vision system identifies all the objects in our store!  ...Why did it just think that box of Lucky Charms was a DOG???'

But humans do plenty of stupid things.  Do you know how many times humans crash by swerving to avoid squirrels and stuff?  Their brain goes 'DON'T HIT THE THING THAT'D BE BAD' and does that in such short order it doesn't look ahead to see how much worse that hard left will be before they go upside down in the ditch.

 

The thing is, people forgive stupid humans more than stupid AI.

[Rayven]

It honestly depends on how advanced the ais controlling the vehicles are. If its like teslas autopilot where it requires a driver to be at the wheel to take over, it would probably cause more accidents. But if it was, humor me here, like self driving cars in movies or futuristic novels where the cars are without fail, then obviously accidents would go down.

[Jtalk4456]

well let's put it this way... In Pittsburgh, PA, the roads are expected to be covered in snow and ice often during the winter, and usually pretty heavily. It's normal for them, happens every time, just part of the climate there. I live in NC, in the piedmont where snow is rare even in the middle of winter. We might get black ice a few times, but basically we have no experience driving in heavy snow around here. Yet when I lived in PA for a year, I was having no problem traversing the snow in my low weight toyota and the natives in their 4 wheel drive jeeps and snow tires and chains were stuck on the side of the road...
Why can I drive in conditions I'm not used to when they couldn't? Simply put, I knew the physics. Other people just bought the jeep or the AWD or 4wd thinking that would fix it, but didn't know how to use those to their advantage. 
For AI driving this points out a balancing act. Do we just trust the tech to do it right? No, the "tech" isn't just some all knowing entity, it's something WE made. That AI will be taught by the same species that spends their whole life shoveling snow half the year while still not being able to drive safely in it. The tech is only a tool, and the effectiveness of the tool can be affected by the designers of the tool and the users of the tool. Unfortunately the users have this tendency to not care about things they should care about, like physics and safety. So making it automatic would definitely reduce crashes. the tool is far more effective when taught what to do by smart people versus being used by those with little common sense or care. 

5 hours ago, Chris Pratt said:

I think we have to reach a point where the AI is universal and not dependent on the car. What I'm talking about there, is instead of the AI being built into the car, it's built into the actual infrastructure, and the car acts as a receiver for commands.

 

As it is, the car's AI, even if it's programmed flawlessly, is entirely dependent on numerous sensors and a camera array. Something as simple as a camera lens getting too dirty because you hit a mud puddle, can severely diminish the AIs ability to effectively control the car.

 

With an infrastructure driven system, all the positions of all the cars on the road can be a known quantity, along with even road hazards. Additionally, the processing power can be near limitless, whereas you're always going to hit some limit of what can be thrown into a car.

Amen. 100% agree

5 hours ago, komar said:

That's because we gotta raise the standards who gets to drive and who doesnt. Pretty much any retard can get a license yet I still feel safer with them around me instead of autonomously driven cars that can easily crash/glitch/loose data for the environment/easily destroyed with a EMP.

I get where you're coming from, but statistically, humans "glitch" far easier and more often. we can be "destroyed" with far less expensive/sophisticated a tool than an emp. We are simply less reliable overall. And there are definitely good drivers, but the percentages are not in our favor by any stretch of the imagination

[Beerzerker]

5 hours ago, CerealExperimentsLain said:

This is part of the problem, any computer system can be 'tricked' or 'trip up' on things a human never would.  You see that in any automated system right now too.  'Ah this machine vision system identifies all the objects in our store!  ...Why did it just think that box of Lucky Charms was a DOG???'

But humans do plenty of stupid things.  Do you know how many times humans crash by swerving to avoid squirrels and stuff?  Their brain goes 'DON'T HIT THE THING THAT'D BE BAD' and does that in such short order it doesn't look ahead to see how much worse that hard left will be before they go upside down in the ditch.

 

The thing is, people forgive stupid humans more than stupid AI.

There's also the factor of unexpected external interference, that too can affect a system and cause unpredictable behaviour.
 Take for example a TV, radio or telephone and suddenly you have either lines on the screen, snow, static coming over the radio/telephone or even hear something else obviously not what you were listening to.
In this case a high powered CB/Ham radio system can and will cause these effects.
I know that's now an older "Thing" but bear in mind all this was new once and there is no telling what may be the next new thing about such external influence/interference taking place affecting who knows what today.

There are also largely unknown cases of it too, like for example once I found out a friend's car stereo system had to be worked on to stop it from affecting the automatic doors of the grocery store they worked for.
Seems the doors were "Jamming" to the beat as in activating (trying to open and close) due to some RF leakage from the system affecting the controller on one particular door. They had already noted the weird behavior always happened right when the stock crew was leaving out to go home, it only lasted for like a minute or two but it did.
They investigated further and once they figured out these activations were in sync to music coming from a certain vehicle that nailed it. 

They let the emplyee know what was going on, the employee didn't have it cranked or anything like that but it was enough to cause it. He'd start his car and let it warm up for a minute, listening to some music while waiting for it to be ready to go and once he had pulled out of the parking lot it stopped.
He had it checked and found out there was a bit of exposed wire to the speakers, the installer had apparently "Nicked" a speaker wire leaving a tiny spot of copper exposed. They found it and replaced the wire - That stopped it.
Imagine that.

[Quackers101]

Someone made a thread, although the focus being a little bit different.

as in "Opinions on Safety with AI cars on the road, future of Tech and Safety".

 

Driving visuals (FSD) make me trust the AI a lot more, than without, or I would not know when to act or why it behaves a certain way.

Tesla driver on some of this

Spoiler

 

drive "blind", testing the different sensors and how they act when "blinded".

 

 

On the topic of AI vs Non AI, saved compared to human response vs failed

 

More videos around this above^

 

[komar]

13 hours ago, Jtalk4456 said:

I get where you're coming from, but statistically, humans "glitch" far easier and more often. we can be "destroyed" with far less expensive/sophisticated a tool than an emp. We are simply less reliable overall. And there are definitely good drivers, but the percentages are not in our favor by any stretch of the imagination

True, but people are predictable.

Also I gave the EMP as an example - here are two others - paintball gun at the sensors/cameras or even big puddle splashed from others. What happens when the computer suddenly can't see?

Also with the current style of manufacturing stuff to brake soon after warranty is over, so you have to buy new, do you feel safe letting something with so many points of failure take control over the vehicle you travel in? I dont....

[Spindel]

On highways, with no wild animals, no snow, no ice and no fog or heavy rain yes
 

In the city, with pedestrians and bikes no. 

[Quackers101]

What I find weird with tesla, will they support cross vehicle communication that doesn't interfer with other signals? (maybe a light/laser sensor?) starlink

Communicate in the network that has been talked about for a while in future cars, knowing speed, status of the vehicle (out of control or has control or similar to airplane TCAS). Airplane TCAS, A traffic collision avoidance system or traffic alert and collision avoidance system. Something that can give directions on were to go, if collision is going to happen between vehicles that can communicate with each other and give a path that might be less deadly or avoid crashes.

[atxcyclist]

Probably much less, but everyone being in an AI-run vehicle is just a pipe dream of people that hate cars and driving, so it will probably never be realized. With that in mind:

 

Humans can look through the windows of the vehicle in front of them and see a pedestrian illegally entering a roadway, long before an AI vehicles sensors can. Humans have flaws, but we are not burdened with a confining set of parameters like software is. I've driven and rode motorcycles for a half million miles at least, you learn a lot from experience; Anticipating long-observed behavior is one of those gray areas in AI that probably will never be good enough to match a a human. 

[Spindel]

I've discussed this in some other thread and I say this. 

 

We can continue this discussion when AI starts to preemptively lower the speed (bellow set speed limit) for potentially hazardous objects out of sensor range (like a kid running from their parents along a sidewalk lined with parked cars). Or changes driving pattering during winter conditions (anyone living where there are winters knows what I mean). 

[Guest willies leg]

Depends on who builds the AI and how it works.

Some people build great code, some build crap code. Same with training the AI.

[Mihle]

With how the AI is right now, no, but in a few years, yes.

[Jtalk4456]

On 5/27/2021 at 8:40 AM, komar said:

True, but people are predictable.

Also I gave the EMP as an example - here are two others - paintball gun at the sensors/cameras or even big puddle splashed from others. What happens when the computer suddenly can't see?

Also with the current style of manufacturing stuff to brake soon after warranty is over, so you have to buy new, do you feel safe letting something with so many points of failure take control over the vehicle you travel in? I dont....

people are anything but predictable... Computers are by definition far more predictable as they run off a set of instructions. We do whatever the crap we want whenever we want, even if it is not logical. 
As for your example... 
Why are people doing paintball on the interstate or highway? Why are the cameras in range of getting splashed? those scenarios are not realistic for an AI designed transportation system. As for accidents or EMP's/acts of terrorism/violence, those are issues now. An EMP would hurt current cars too. Current cars use computers to manage all sorts of processes for the driving process. Also lets say paintball example, that would obstruct current driving too. If your windshield is suddenly not transparent, you can wreck just the same as an AI. BUT.... An AI would have a better reaction time, and at loss of input could be set to immediately begin slowing down and having flashers on or send a signal to all other cars of it's position while pulling over based on simple math calculations. 
And the biggest thing being forgotten here is that we already have a very successful AI based transportation system running. Autonomous trains are already running in places. With tracks or preset paths available and a larger ai just routing who goes when, most of the concerns disappear. In the far future, most if not all transportation can become fixed by rail or guide of some type, then an AI just manages routing. A better AI than in Cities Skylines, of course, but you can get a gist of the idea where that video game is mostly managing hundreds or thousands of cars following set paths defined by the software. Oversimplified, sure, but it's a video game, so....
Point being with a proper infrastructure, an AI based transportation system would be far more predictable, have LESS points of fault, and be far more efficient than we can ever hope to be as humans that can suddenly have a health issue and be unable to drive, even with proper visual input and lack of highway paintballs

[Kisai]

On 5/26/2021 at 10:17 AM, Actual_Criminal said:

 

 

Anyway, most accidents I have seen whilst working in the car insurance industry for a year (a few years back) were down to human error.

 

Two things:

a) We have had automated technology going back to the 70's, Unattended Automatic Rail operation is nothing new, and is the ONLY way to guarantee safety of passengers.

b) The reason why everything is new again with cars, is due to the switch to electric vehicles. Prior to relatively recently (eg Prius) all other vehicles were gas powered and didn't even generate more than 95w (barely enough to run a current 12" laptop) No AI can run on that, but in the 2000's we weren't there.

 

The Vancouver Skytrain, which is early 80's technology, has large computer systems both at central control and on the trains themselves for the signalling.

 

So, most automated driving has been focused on trying to replace a human driver with an AI that makes decisions the way a human would, which is doomed to fail because roads and signage are not consistent from city to city, let alone state to state, and a computer isn't going to know about temporary changes in speed or traffic directions, or what new roads have been built since the developer of the software last LIDAR'd the landscape.

 

There needs to be a central control aspect to tell the cars about real time traffic changes, which means that there needs to be a "free to access" vehicle network, as well as car-to-car communications. Unattended Automated trains already do this (central control and onboard control.) Even when an accident happens with unattended trains, it's always due to a human-issue, like a human driver or a maintenance issue ( a recent accident in Malaysia  https://www.straitstimes.com/asia/se-asia/six-critical-in-the-aftermath-of-malaysias-rail-crash-as-human-error-identified-as was due to a human driver driving a train in the wrong direction, and the automated train already stopped, but had no where to go to avoid the manual driven train) 

 

The lack of car-to-car communication is presently the obstacle that needs to be overcome, otherwise automated cars are essentially blind to traffic accidents, construction and detours. 

 

Then there is also having cars moving together (eg escort cars, motorcades, etc) which need that communication, even if the lead car is manually driven, you want all the following cars to keep pace.

 

[Blademaster91]

51 minutes ago, Jtalk4456 said:

Point being with a proper infrastructure, an AI based transportation system would be far more predictable, have LESS points of fault, and be far more efficient than we can ever hope to be as humans that can suddenly have a health issue and be unable to drive, even with proper visual input and lack of highway paintballs

I think in order to have a proper infrastructure, self driving cars would need to have separate driving lanes from regular cars. Also self driving systems don't work well in inclement weather like rain or fog, so it would be years before AI based systems could work in snow.

[Jtalk4456]

13 minutes ago, Blademaster91 said:

I think in order to have a proper infrastructure, self driving cars would need to have separate driving lanes from regular cars. Also self driving systems don't work well in inclement weather like rain or fog, so it would be years before AI based systems could work in snow.

Oh I agree totally. The proper infrastructure I'm talking about doesn't even have lanes for non automated vehicles. Same as not having an old bearded crazy man on a pump cart allowed on the modern rail lines. This would be a whole new setup. If we move to a fixed rail or guided system not needing to make minute direction changes, just routing, then even inclement weather would be far less of a problem, if an issue at all