media bias and self driving cars

[TheDrake]

So, we wind up seeing stories broadcast far and wide whenever a self-driving car gets in an accident, despite the obvious disparity between the average driver and self-driving vehicles. Is the media plotting against the technology, or are they just trying to gain market share by writing stories they know will be spread?

I recently listened to a podcast of Malcolm Gladwell's Revisionist History, and the clear determination was that Toyota cars did not have a fatal flaw in software, but that driver error was largely responsible for several inadvertent acceleration accidents. It is a curious situation. It involves the media manipulating their own research to make it look like Toyota cars took off on their own.

What do you think?

For those that might be interested in hearing this full story, see the link below.

podcast link

[JoshCrow]

It's not some sinister plot. I read an article lately that made the case that the laziest story in journalism is "first X to do Y". Since everyone is waiting for self-driving cars to show some kind of flaw, such an article is immediately clickbait.

[DJQuag]

Self driving cars are the future. Most Americans will be against it, as they are so "indepedent." And don't want to give it up.

Yeah, clickbait will ensue.

In other news, water is wet.

[D.W.]

I can't wait for self driving cars.  That is unless the "ethics of AI" end up being capable of sacrificing me (the passenger/driver) to save other lives that could potentially be "at fault" by acting recklessly, not having the right of way, or just having made an error.

Then I'm not so sure... 

[TheDrake]

In general, I think this is largely a hypothetical rather than a real case scenario. I also don't think that it is much different from how a human behaves today. Most drivers will instinctually swerve away from a pedestrian or slam on the brakes rather than mow them down to maximize their own safety.

And naturally, the more widely self-driving cars are adopted, auto v. auto or auto v. truck can have the AIs coordinate a half mile away from an intersection to determine who will have the right of way, and slow down one by 2% to avoid even seeing each other.

[TheDeamon]

I can't wait for self driving cars.  That is unless the "ethics of AI" end up being capable of sacrificing me (the passenger/driver) to save other lives that could potentially be "at fault" by acting recklessly, not having the right of way, or just having made an error.

Then I'm not so sure...

The challenges will be how the human driven vehicles interact with the computer/sensor driven vehicles. I see the prospect of things being "platooned" initially, where there is a human driving the car/truck in front, and it shares telemetry with the vehicles behind it. Just hope like hell the person in front knows what they're doing.

Driving on sensors alone is something they already know is non-optimal solution, particularly in crowded urban areas with heavy traffic, because humans will drive much more closely to each other than a purely autonomous car would on its own(as it would be rigidly practicing the defensive driving techniques most of us were taught, but rarely apply fully).

The REALLY fun one is going to be self-driving vehicles and distinguishing between animals(/road kill) in the road and humans in the roadway. For a big rig, they'll normally go over a deer rather than swerve because that might roll the truck. While that might be fine when it's a deer carcas lying across the roadway, but if it's a human body...

Well, then a lot more subjective calculations start to come into play(including taking the risk of rolling the truck). And a human driver is more likely to correctly identify what's in the roadway than an AI at this point in time technology wise. Otherwise I'd advise bikers to not wear any animal hide colored jackets....

[JoshCrow]

Driving on sensors alone is something they already know is non-optimal solution, particularly in crowded urban areas with heavy traffic, because humans will drive much more closely to each other than a purely autonomous car would on its own(as it would be rigidly practicing the defensive driving techniques most of us were taught, but rarely apply fully).

I just wanted to chime in on this - this bit is true under mixed human-AI conditions but not under all-AI conditions. A fleet of strictly self-driving cars could (upon networking and confirming that they are not in the presence of a human driver) drive within an inch of each other. If it became the law, for example, to only have AI-drivers (for safety reasons), it would permit narrower lanes, vehicle-managed stoplights, and other advantages. However with a mix of human drivers in the picture you are correct that the programmers would make sure the cars treated human-driven vehicles as potentially unpredictable.

[Fenring]

It gets even better

If it became the law, for example, to only have AI-drivers (for safety reasons), it would permit narrower lanes, vehicle-managed stoplights, and other advantages.

It might be scary to the human passengers, but traffic lights could possibly be removed altogether in certain cases, where the cars would 'just know' when to go and when to stop. I suppose there would have to be a way to manage pedestrian traffic in such cases. However this reminds me of another problem, which is that in downtown areas where pedestrian traffic is heavy (or in places like Times Square in NYC) the pedestrians rarely if ever care to let cars turn the corner and will even run onto the street to 'make the light' rather than wait 40 seconds and let the cars that have been waiting go through. Many drivers know that the only way to progress the traffic in such cases is to bully your way into the pedestrian traffic and sort of oblige them to create a path for a few cars to go through; else the traffic would literally never progress. With self-driving cars I'm sure they will be programmed not to do that, in which case something would have to be done about the pedestrians to ensure they don't monopolize an intersection at the expense of the timidly-programmed AI cars. At such a time as cars are entirely AI driven I could see a case being made to begin enforcing jaywalking laws and levying fines for obstructing the intersection.

That being said, I predict that all traffic problems in almost every city would vanish if human drivers were removed from the equation. Maybe someplace like L.A. would still be bad, but whenever I have seen heavy traffic in major cities I can usually identify the primary or secondary causes of it (structural, human error, etc.), and most of the time it's a mixture of incompetence and occasionally the city doing something stupid.

[TheDeamon]

That being said, I predict that all traffic problems in almost every city would vanish if human drivers were removed from the equation. Maybe someplace like L.A. would still be bad, but whenever I have seen heavy traffic in major cities I can usually identify the primary or secondary causes of it (structural, human error, etc.), and most of the time it's a mixture of incompetence and occasionally the city doing something stupid.

Yeah, most of the traffic backups I've seen(and I drive Over the Road) are to poor traffic engineering which creates "weaving traffic" scenarios, or bad driver habits like people failing to "properly honor" a traffic merge(and further conflicting this is the Obstructing Traffic laws/further bad road engineering, and the tendency of some LEO's to ticket the people who try to "force" people to honor the merge by pacing the lane beside them).

The prime example of "a valid obstructing traffic" ticket due to poor road engineering is a road lane that ends in an off ramp. Well, traffic backs up in all lanes except the "exit only" ramp. Most people in that lane are rightfully trying to exit, but others are not "honoring the merge" and instead jump in the open lane, drive as close to the exit as they can, then merge back into traffic, further exacerbating the traffic backup. Someone could pull into that exit lane to block off the people who are skipping ahead, but then they're obstructing the exit, and will often get ticketed accordingly.

The "grey" ones are the miles long traffic backups where truckers/other traffic will do a rolling block on a non-exit lane, but as a consequence, what would be 1.5 miles worth of 2 lanes of backed up traffic suddenly becomes a 2.5 mile traffic backup because of all the dead space in the lane that is subject to the rolling blocks.

The "total BS" ones are where the merge point is in sight, and less then 1/2 mile away, and someone playing the rolling road block gets an obstruction ticket for their efforts to help speed the traffic along.

But yeah, just getting AI involved in handling traffic merge situations would be a huge deal.

The traffic weaving thing under AI control would probably be just as, if not even more so, terrifying than the intersection with no stoplight. Could you imagine being in that car the first time you have it come hurtling onto the freeway at 65MPH, as it then promptly cuts across 5 lanes of heavy traffic to hit an exit on the other side of the highway a half mile down the road?

[NobleHunter]

I'm so glad I don't commute.

[Fenring]

Poor road engineering creates the conditions where, if the drivers are incompetent, traffic will inevitably become congested. It doesn't necessitate it in principle, but in practice it will happen. However there are elements to the road engineering which are 'functionally' poor even though in theory they shouldn't be. For instance if a highway is build around a natural obstruction, such as a hill or parts of the city that were there previously, the curve of the road will cause people to slow down. It shouldn't! But it will. There is literally no reason in such cases for drivers to apply the brakes, but they will do so anyhow because doing anything other than driving in a straight line scares them, I guess. Far worse than a curve in the road is an incline of any kind. If the road inclines more than a trivial amount it will automatically create a traffic jam due to people failing to maintain speed, or even braking as a result of losing sight of the road in the distance. When I see a seemingly inexplicable traffic jam my typical assumption is "it's a hill." Usually that's the case. Between curves in the road and inclined sections of highway I think we can probably account for a vast proportion of highway traffic. The issue of lanes mysteriously ending (such as in the case of an exit) is an issue too, but even so that is part of the greater problem of drivers on the aggregate being extremely poor at changing lanes. This is both a judgement deficiency and a deficiency in mechanical skill, where the mere act of one person changing a lane poorly can create a cascade effect and bring all traffic to a standstill. My current theory is that it only takes a minute quantity of actual blunders to create a permanent traffic jam (until such a time as the number of cars on the road reduces). The shockwave's ability to self-correct will be reset back to 'gridlock' as long as a certain small number of people continue to feed into it periodically.

So yeah, all of this would vanish with self-driving AI, and I can't wait until it happens.

[TheDrake]

A simple example are "throttling" lights for on-ramps. It is shown that this highly increases the efficiency and fairness of freeways. Tons of ramps don't have them, I suspect due to expense. These are a clumsy solution compared to AI, who could avoid the unnecessary "full stops"and use a smoother algorithm to achieve the same cars/minute at better fuel efficiency and lower wear.

Eventually, I see large urban areas going to all-AI. London is a very likely candidate.

But you won't find the lamestream media touting all the potential benefits of AI autos. Just challenges, risks, and preposterous hypotheticals.

[Pete at Home]

Hacker terrorism.

[Pete at Home]

For example, imagine this happened to a network of self-driving cars:

http://www.bbc.com/news/technology-38127096

[TheDrake]

Well, I'd prefer that the self-driving cars not be operated by the government. The lack of security and technological savvy exhibited here, the launch of healthcare.gov, and officials who don't know how to use a PC would give me great pause.

[Seriati]

I just wanted to chime in on this - this bit is true under mixed human-AI conditions but not under all-AI conditions. A fleet of strictly self-driving cars could (upon networking and confirming that they are not in the presence of a human driver) drive within an inch of each other.

Except that only works if you run them through tunnels with limited access points.  In a large country, there is no way to be absolutely certain about all traffic conditions and/or obstructions.  In your one inch model, one deer causes how many cars to crash?  Dozens, more?  Maybe better on average, maybe not.  I've seen mattresses and furniture fall off a car, tires explode (and even once a RIM come apart), trees fall on a road, plenty of animals, children, sporting equipment enter a road way. 

Safety margins will never be able to reduce as far as you say, even if you could perfectly avoid any risk of surprise obstruction, even trains aren't perfect on that front, because you always have the possibility of equipment failure.

If it became the law, for example, to only have AI-drivers (for safety reasons), it would permit narrower lanes, vehicle-managed stoplights, and other advantages.

Narrow lanes, maybe, but again that increases the risks if oncoming traffic deals with an unanticipated event.  Could already have GPS triggered stop lights if they really wanted to add/seriously upgrade computer technology at every stop light (and the time for repair and "connection" issues).

However there are elements to the road engineering which are 'functionally' poor even though in theory they shouldn't be. For instance if a highway is build around a natural obstruction, such as a hill or parts of the city that were there previously, the curve of the road will cause people to slow down. It shouldn't! But it will.

Why shouldn't it?  Basic math says that an object going around a curve will have a tendency to go off the curve.  On high speed roads we can often fight that with banking (but not always).  We're pretty much limited to making sure our velocity stays below what the friction of our tires can handle.  If anything self driving cars would have a much higher gradient as there would be no reason to slow them from the safe maximum, which will be substantially higher to a straight away.

There is literally no reason in such cases for drivers to apply the brakes, but they will do so anyhow because doing anything other than driving in a straight line scares them, I guess.

Or because if they go as fast around the curve as they do on the straight, they literally fishtail/spin out, or push into the wall/off the road? 

Far worse than a curve in the road is an incline of any kind. If the road inclines more than a trivial amount it will automatically create a traffic jam due to people failing to maintain speed, or even braking as a result of losing sight of the road in the distance.

Losing sight over the top of a hill should cause a slow down.  I had a guy that tried to illegally pass me on an uphill almost plow headfirst into a car that came over the other side that he couldn't see (then he did fish tail and almost hit my back end when he desperately braked and cut hard to avoid the contact).  Even a self drive car wouldn't be expected to see the other side of a hill with an obstructed LOS (and no manufacturer should ever be allowed to let a car rely on external sensors as the sole source of info with obstructed on board sensors).

No one's mentioned some "reasonable" concerns.  What happens when a car gets hacked?  Stolen car drives itself away, kidnapped by getting in your own vehicle, murdered when you car takes a turn off a cliff.  Heck what if it just malfunctions, I've never had a single piece of computing technology that has worked without flaw every time, let alone over the lifespan of your average car.  Now add in defects and faults in every piece of interacting technology.  What happens when both sides of the green light turn on?  What happens when a car's network connection is glitch?  They can't even get satellite radio to work in some areas of the country (urban and remote), location services fall off the grid all the time, mapping apps still run you miles from the destination on occasion.  Given that any criminal would be an idiot to get in a vehicle that the police can reroute to jail, you can expect "broken" cars to show up as well.

[TheDrake]

What happens when anything gets "hacked"? We don't seem to have people raising and lowering bridges, making phones explode remotely by overriding the battery, or randomly turning off people's nintendos.

Risk models for security evaluate likelihood, sophistication, etc. We don't fly planes manually because they might get hacked. We don't use gold bullion and barter because banks might get hacked. We defend as best we can based on the threats.

Self driving cars don't add significantly to the over risk of being "hacked" it just sounds like a super scary bogeyman.

Meanwhile weighed against 35,000 traffic fatalities every year in the US alone, you'd have to have a lot of hacker induced deaths and kidnappings to avoid the break-even point.

[Seriati]

What happens when anything gets "hacked"? We don't seem to have people raising and lowering bridges, making phones explode remotely by overriding the battery, or randomly turning off people's nintendos.

We do have people remotely overriding the cameras on their video game consoles, and even Swatting them (unless that's a urban myth?).  Stealing people's identities.  Raising and lowering bridges is rather pointless, and certain to bring down a literal federal case, than minor crimes.  We have endless cases of people modifying cars, car bombs, brake lines, not to mention car jackings.  Why do you think it would be different with computers?

Risk models for security evaluate likelihood, sophistication, etc. We don't fly planes manually because they might get hacked. We don't use gold bullion and barter because banks might get hacked. We defend as best we can based on the threats.

Self driving cars don't add significantly to the over risk of being "hacked" it just sounds like a super scary bogeyman.

Why do you think planes are a greater risk of being hacked (when few people understand them well enough to do it meaningfully) than a vehicle that there are over 250 million of in the US, and that virtually every person in the country has some contact with on a daily basis?  People hack phones within days of release, why wouldn't they hack cars?

Meanwhile weighed against 35,000 traffic fatalities every year in the US alone, you'd have to have a lot of hacker induced deaths and kidnappings to avoid the break-even point.

That's a fairer response.  I do have a visceral reaction to the innate lose of control and trust in the great traffic overlords in the sky.  I find it creepy.  I do  agree though, you'd likely have a significant drop in traffic deaths.

[Fenring]

Seriati, are you just being contrary or are you attempting to make a real case against AI-driven cars? I think you will be hard-pressed to argue that AI would be less safe or less efficient than human drivers. Your comments above seem to suggest that the advantages of AI may not be as great as we're suggesting, which is not the same as suggesting AI would be worse than what we have now. What's your actual position?

[TheDrake]

We do have people remotely overriding the cameras on their video game consoles, and even Swatting them (unless that's a urban myth?).  Stealing people's identities.  Raising and lowering bridges is rather pointless, and certain to bring down a literal federal case, than minor crimes.  We have endless cases of people modifying cars, car bombs, brake lines, not to mention car jackings.  Why do you think it would be different with computers?

My point wasn't that cars cant be hacked but that they don't pose much greater a safety threat than 1,000 other things that can be hacked. Stabbing and shooting will remain the method of choice when people want to physically harm each other. If someone wants to steal a car, carjacking and breaking in require less effort. Like I said:

Self driving cars don't add significantly to the [overall] risk of being "hacked" it just sounds like a super scary bogeyman.

Other than that, we only have the possibility of this being an avenue of mass hacking for nefarious purposes by a state or non-state actor, which is why I brought up bridges and planes.

[Seriati]

Seriati, are you just being contrary or are you attempting to make a real case against AI-driven cars?

Just a case for toning down the optimism.  We don't have an infrastructure that comes close to what we'd need to make self driving cars a reality.  Mixing human and non-human drivers may go more swimmingly than I expect (Y2K didn't destroy the world after-all), but we are no where close to getting into self driving cars that run at 120 mph with no safety margins.  Honestly, I'd expect that in a mixed environment the self driving cars will underperform the humans, because the safety margins will have to be large and the AI will be inflexible in following them, which means they will defer to human controlled vehicles.

How would an AI handle protestors obstructing an interstate?  Keep you parked?  Even if they were being hostile?

I think you will be hard-pressed to argue that AI would be less safe or less efficient than human drivers. Your comments above seem to suggest that the advantages of AI may not be as great as we're suggesting, which is not the same as suggesting AI would be worse than what we have now. What's your actual position?

I think they will be safer, it's hard to imagine they'd be less safe that some of the local drivers.  I think, you absolutely need an override, would be okay if it didn't work once the car was going over 50mph, but particularly in low speed areas and poorer roads, you'd need and want the ability to be hand's on.

I don't like the loss of control inherent in this product.  I really don't trust authorities not to abuse it.  Like the example of the police rerouting you to jail, but think bigger scale, permanent records of every location you go to?  It's bad enough with your phone doing it, do you need a car that interacts with traffic control where your travels can be accessed remotely and eventually without your knowledge?  Cars are a necessity for freedom of movement, and that freedom becomes strained if its constantly monitored.  How will you like it when your health insurance goes up because you went through McDonald's drive thru once too often?   Or when driving records are subpeoned every time someone is sued?  Like going to protests, better get a car pool if you don't want to be identified.

My actual position, is that they will be safer on the roads, less safe in other ways.  That it's inevitable that they come onto the roads, maybe that they become mandatory.  That we need a massive infrastructure upgrade to make them fully efficient, which may or may not occur and that honestly is money that'd be better spent on far more efficient mass transit, than making cars travel faster more safely (which makes the cars even less energy efficient compared to the alternatives because of the increase in speed).  I guess, I'd say, it's complicated.

[Seriati]

My point wasn't that cars cant be hacked but that they don't pose much greater a safety threat than 1,000 other things that can be hacked. Stabbing and shooting will remain the method of choice when people want to physically harm each other. If someone wants to steal a car, carjacking and breaking in require less effort.

You missed my point.  A hacked car puts you at personal risk, none of those other products do that in the same way.  Sure if you are important enough someone could hack a plane, with amazingly lucky circumstances just to get you, but mostly a plane is a generalized risk.  Cars are personal - you own yours, and can be identified by it - you physically get in it, and trust it to take you places.  If it's self driving, people will get used to them and stop focusing on where they're going and what they are doing, even while a passenger.  Don't really see how you can't see the ways a mobile prison that people voluntarily get into couldn't be nefarious purposes, do you get how a chaffered car with a bribed driver could be?

Or maybe you didn't miss it, but are just refusing to acknowledge that cars present a unique risk in the context of hacking. 

[Fenring]

Seriati, I wouldn't at all take issue with your reservations about being monitored while driving, except that at present people are already monitored in ways that are far more pervasive and intrusive. Granted, none of them involve your car spiriting you away from where you want to go, but on the other hand the use of monitoring Facebook, for instance, does result in the odd person finding the FBI at their door for posting controversial material. In terms of optimism I would hope that government surveillance would be stopped wholesale, including in future AI-cars. I would hope that the AI system would either be a closed network, or else not even a network at all but rather a lone AI in each car doing its own thing, unconnected to any other system. Later on, when we're talking about integrated traffic lights and so forth that connection would need to be there, but with a mixed human/AI road system I see no reason why the AI should have to be connected to any external network. I guess a kill switch would be in order so that the human could take over, but actually that worries me a little because if the person barely ever drives I'm not enthused that they should randomly start whenever they feel like it.

 

[TheDrake]

You missed my point.  A hacked car puts you at personal risk, none of those other products do that in the same way.  Sure if you are important enough someone could hack a plane, with amazingly lucky circumstances just to get you, but mostly a plane is a generalized risk.  Cars are personal - you own yours, and can be identified by it - you physically get in it, and trust it to take you places.  If it's self driving, people will get used to them and stop focusing on where they're going and what they are doing, even while a passenger.  Don't really see how you can't see the ways a mobile prison that people voluntarily get into couldn't be nefarious purposes, do you get how a chaffered car with a bribed driver could be?

Or maybe you didn't miss it, but are just refusing to acknowledge that cars present a unique risk in the context of hacking.

Perhaps you just have more supervillians out to get you.

If someone is targeting an individual, it would be far simpler to cut a brake cable, wouldn't it? If you want to kidnap someone, it seems there are far easier ways. Especially since, unless there were jamming available, you're going to give the police a play by play on where you're being taken. People have planted bombs on cars in the past. It is only a big problem if it is far easier to gain electronic access to the car than physical access.

If you are a head of state, an Iranian nuclear scientist, celebrity, or anybody receiving death threats, then yes this might up your risk. For the average suburban housewife, not so much.

As for chauffers, people regularly take taxis, limos, ubers, and other conveyances with less assurance than a self-driving car. In films, this is a real common way to capture your enemies. In real life, I think, less so.

[Pete at Home]

I tend towards Seriat's position here, but reserve the right to change my mind if other countries bigger than Iceland implement it without cataclysm.  I defy Italy don't want to see America as the first adopter.

Let's adopt the Metric System and mass transport and otherwise catch up to where the rest of the world was in the 19th century