Cars - history, now and future

[Robert Roaldi]

It was a software instruction issue, the programmers thought it makes more sense to just slow down when the sensors don't detect any moving objects in the intersection. ...

You know, allowing programmers to write code that violates traffic laws because they thought it was a good idea does not fill me with confidence. This sounds like another instance of the silicone valley mentality of forcing users to be unpaid beta-testers. And I say that as someone who earned his living as a software engineer for 25 years.

Has it been resolved who gets the traffic ticket when the AI mis-behaves? Who can you sue when someone gets killed? Is this kind of thing covered by regular insurance policies?

[LesPalenik]

I thought roundabouts were mainly to eliminate traffic lights.  In residential areas, there's not enough room to make roundabouts and the four-way stops are there because it's cheaper than traffic lights. You could use yields on one street and stops on the other.  But I think there are more accidents that way.  A four-way stop forces everyone to stop and look before proceeding.  People don't pay attention to yields.  Apparently, neither do Teslas.

Roundabouts facilitate faster traffic flow through intersections. Saving the traffic light installation, exhaust and acoustic pollutions as well as energy consumption is just a bonus.

[Alan Klein]

It was a software instruction issue, the programmers thought it makes more sense to just slow down when the sensors don't detect any moving objects in the intersection. Common sense, but paradoxically a clear violation of US and Canadian traffic rules. They already made the software change which will be downloaded over the air to Tesla cars with FSD.

IMO, the 4-stop rule on an intersection is one of the stupidest rules. Made by idiots for idiots. In Europe, you see very few full stop signs, most intersections have Yield signs or roundabouts.

Maybe once most cars use some kind of Autosteer and motion sensors, they could do away with the full stop signs.

Les, How does the new program work?  Attached are the rules for crosswalks at least in Texas.  Of course, if your Tesla hits a Texan in the crosswalk, you'd better be ready for a gunfight.
https://www.thehartlawfirm.com/faqs/do-you-know-when-pedestrians-do-not-have-the-right-of-way-.cfm

[LesPalenik]

You know, allowing programmers to write code that violates traffic laws because they thought it was a good idea does not fill me with confidence. This sounds like another instance of the silicone valley mentality of forcing users to be unpaid beta-testers. And I say that as someone who earned his living as a software engineer for 25 years.

Has it been resolved who gets the traffic ticket when the AI mis-behaves? Who can you sue when someone gets killed? Is this kind of thing covered by regular insurance policies?

From what I heard about Tesla, the employees are allowed to make mistakes and if anything goes wrong, they are quick to fix the mistakes. In one specific week, they made 27 changes to the software and hardware, while GM made 2 changes in the entire year. Well, if you make only 26 cars per quarter, that's not so bad.

[LesPalenik]

Les, How does the new program work?  Attached are the rules for crosswalks at least in Texas.  Of course, if your Tesla hits a Texan in the crosswalk, you'd better be ready for a gunfight.
https://www.thehartlawfirm.com/faqs/do-you-know-when-pedestrians-do-not-have-the-right-of-way-.cfm

I don't know the current status. I guess, the car will come to a full stop, look around and take off like a rocket. Model S should depart from the scene quicker than Model 3.

[Alan Klein]

Roundabouts facilitate faster traffic flow through intersections. Saving the traffic light installation, exhaust and acoustic pollutions as well as energy consumption is just a bonus.

New Jersey where I now live is known for jughandles.  To make a left from most main roads, you turn off to the right and then cross back over the main road to make the left.  It looks like a jughandle from a drone. But not always.  Sometimes, there's a left-turn lane in the main road you're turning off of.  It's all very confusing to this New Yorker to know what's coming up and which lane you should be in.

 

[Alan Klein]

Of course, in New Jersey, the gas stations have to fill up your gas tanks, by law.  You're not allowed to do it. So they do something that's right. Especially when it's raining or snowing and 5 degrees. 

[Alan Klein]

From what I heard about Tesla, the employees are allowed to make mistakes and if anything goes wrong, they are quick to fix the mistakes. In one specific week, they made 27 changes to the software and hardware, while GM made 2 changes in the entire year. Well, if you make only 26 cars per quarter, that's not so bad.

I understand CEO Bara made the changes herself.

[LesPalenik]

I understand CEO Bara made the changes herself.

Poor lady. She's been roasted million times in the last few months. Now, we know why nobody wants her job.

[Robert Roaldi]

From what I heard about Tesla, the employees are allowed to make mistakes and if anything goes wrong, they are quick to fix the mistakes. In one specific week, they made 27 changes to the software and hardware, ...

This is no way fills me with confidence. Moving fast and breaking things might be ok in cute little "friendy" social media toy apps, but real world things need more thought and testing than that.

The real test of AI will come when it has to deal with the world's teenagers. Once all those kids figure out how much fun it is to step out onto the roadway and have all those cars come to a grinding halt, traffic will become a nightmare, especially near high schools. You thought that throwing eggs at buildings was a nuisance?

Might even make car jacking easier. 

[Robert Roaldi]

I saw a funny comment on Facebook yesterday where someone wondered why a company like Tesla that has over 100 lawyers on staff could allow software to be distributed that allowed the car to roll through a stop sign. I figure this is a just a case of letting customers do the beta testing, a long held silicone valley strategy, imo.

Just a follow-up question regarding legal questions re FSD or even partial AI control in case there are people reading here who might know about these things. Tesla (and other manufacturers) have control over the software that runs the cars. Users do not have access to the code and neither do insurance companies. So when something goes wrong, and something will go wrong, what is the legal framework? Who's to blame? Who gets sued?

I assume that a Tesla user (or other) either implicitly or explicitly has to agree to new software downloads in the same way we all agree to terms for any software we download/install. Are there clauses in those agreements (that no one reads) that absolves the manufacturer from culpability? Maybe I'm cynical but I can't see them doing otherwise. There's a long history of corporations downloading blame and culpability to users. At the very least, I can see it being the initial default position that might come in handy in later court trials. Have they been paving the way for this with other legal precedents?

Are there clauses in insurance company agreements that absolves them from payouts in case of AI failure? I can see where they might want that. And are there insurance companies that refuse to sell insurance to drivers of some models?

I may be wrong but the impression I have is that in the US these things don't tend to be designed comprehensively planned well from the outset. The impression I have is that the modus operandi is to let things happen and then let others sort it out in the courts later. This may be a completely incorrect impression on my part, it's only a feeling I have. Is the area of FSD proceeding in that manner?

I can imagine a situation where someone downloads an AI version with a bug that results in a bad outcome. The injured party sues, the insurance company refuses to pay on some pretext, and the owner of the car cannot produce the code to prove that it was a software error. Not many drivers will have the resources to sustain that 10 year court battle.

[Alan Klein]

I saw a funny comment on Facebook yesterday where someone wondered why a company like Tesla that has over 100 lawyers on staff could allow software to be distributed that allowed the car to roll through a stop sign. I figure this is a just a case of letting customers do the beta testing, a long held silicone valley strategy, imo.

Just a follow-up question regarding legal questions re FSD or even partial AI control in case there are people reading here who might know about these things. Tesla (and other manufacturers) have control over the software that runs the cars. Users do not have access to the code and neither do insurance companies. So when something goes wrong, and something will go wrong, what is the legal framework? Who's to blame? Who gets sued?

I assume that a Tesla user (or other) either implicitly or explicitly has to agree to new software downloads in the same way we all agree to terms for any software we download/install. Are there clauses in those agreements (that no one reads) that absolves the manufacturer from culpability? Maybe I'm cynical but I can't see them doing otherwise. There's a long history of corporations downloading blame and culpability to users. At the very least, I can see it being the initial default position that might come in handy in later court trials. Have they been paving the way for this with other legal precedents?

Are there clauses in insurance company agreements that absolves them from payouts in case of AI failure? I can see where they might want that. And are there insurance companies that refuse to sell insurance to drivers of some models?

I may be wrong but the impression I have is that in the US these things don't tend to be designed comprehensively planned well from the outset. The impression I have is that the modus operandi is to let things happen and then let others sort it out in the courts later. This may be a completely incorrect impression on my part, it's only a feeling I have. Is the area of FSD proceeding in that manner?

I can imagine a situation where someone downloads an AI version with a bug that results in a bad outcome. The injured party sues, the insurance company refuses to pay on some pretext, and the owner of the car cannot produce the code to prove that it was a software error. Not many drivers will have the resources to sustain that 10 year court battle.

It's more complicated than that.  You can't just drive a car automatically.  The law in most states requires the driver to have complete control of the car, both hands on the wheel.  I believe California (and maybe some other states) has granted Tesla a waiver to test their software and cars under certain circumstances.  There's also discussion to pull that right because of accidents.  So you'll have to check the rules that allow the right.  It may require insurance coverage beyond normal that Tesla is paying for. 

[LesPalenik]

I saw a funny comment on Facebook yesterday where someone wondered why a company like Tesla that has over 100 lawyers on staff could allow software to be distributed that allowed the car to roll through a stop sign. I figure this is a just a case of letting customers do the beta testing, a long held silicone valley strategy, imo.

Just a follow-up question regarding legal questions re FSD or even partial AI control in case there are people reading here who might know about these things. Tesla (and other manufacturers) have control over the software that runs the cars. Users do not have access to the code and neither do insurance companies. So when something goes wrong, and something will go wrong, what is the legal framework? Who's to blame? Who gets sued?

I assume that a Tesla user (or other) either implicitly or explicitly has to agree to new software downloads in the same way we all agree to terms for any software we download/install. Are there clauses in those agreements (that no one reads) that absolves the manufacturer from culpability? Maybe I'm cynical but I can't see them doing otherwise. There's a long history of corporations downloading blame and culpability to users. At the very least, I can see it being the initial default position that might come in handy in later court trials. Have they been paving the way for this with other legal precedents?

Are there clauses in insurance company agreements that absolves them from payouts in case of AI failure? I can see where they might want that. And are there insurance companies that refuse to sell insurance to drivers of some models?

I may be wrong but the impression I have is that in the US these things don't tend to be designed comprehensively planned well from the outset. The impression I have is that the modus operandi is to let things happen and then let others sort it out in the courts later. This may be a completely incorrect impression on my part, it's only a feeling I have. Is the area of FSD proceeding in that manner?

I can imagine a situation where someone downloads an AI version with a bug that results in a bad outcome. The injured party sues, the insurance company refuses to pay on some pretext, and the owner of the car cannot produce the code to prove that it was a software error. Not many drivers will have the resources to sustain that 10 year court battle.

As to the rolling stop, the way Tesla implemented it, it was quite safe. Definitely safer than most humans drive through the stop sign.
Their "rolling stop" feature allowed vehicles to travel through all-way stop intersections at up to 5.6 mph under certain conditions. Drivers had to set their vehicle to a "full self-driving" style that included the feature. The vehicle's "chill" setting, the setting a user can select that drives the most conservatively of the options Tesla gives drivers, is not described as including the feature. The other settings, "average" and "assertive," include the feature.
The "rolling stop" function also required that no relevant moving cars or pedestrian or bicyclists were detected near the intersection. The function wouldn't activate without sufficient visibility, and all roads entering the intersection need to have a speed limit of no more than 30 mph. Tesla complied with the National Highway Traffic Safety Administration request and disabled the rolling stop option.

At this time, FSD has not been officially released, it is still in Beta stage, and the drivers who use it, are fully responsible for driving, steering and stopping.

Most likely, at this time, all insurance companies stipulate as well that the driver is responsible for any accidents. Once FSD is officially released, drivers with Tesla insurance will be not be liable for the accidents, the insurance will cover it. I'm not sure how quickly will the other insurance companies adjust to the FSD or other self-driving versions.

As to releases of new FSD versions (after it has been officially released) I wouldn't worry about new bugs appearing in them. Tesla will test regressively all new FSD versions on their existing and simulated data which should catch any bugs.

[Robert Roaldi]

As to the rolling stop, the way Tesla implemented it, it was quite safe. Definitely safer than most humans drive through the stop sign.
Their "rolling stop" feature allowed vehicles to travel through all-way stop intersections at up to 5.6 mph under certain conditions. Drivers had to set their vehicle to a "full self-driving" style that included the feature. The vehicle's "chill" setting, the setting a user can select that drives the most conservatively of the options Tesla gives drivers, is not described as including the feature. The other settings, "average" and "assertive," include the feature.
The "rolling stop" function also required that no relevant moving cars or pedestrian or bicyclists were detected near the intersection. The function wouldn't activate without sufficient visibility, and all roads entering the intersection need to have a speed limit of no more than 30 mph. Tesla complied with the National Highway Traffic Safety Administration request and disabled the rolling stop option.

At this time, FSD has not been officially released, it is still in Beta stage, and the drivers who use it, are fully responsible for driving, steering and stopping.

Most likely, at this time, all insurance companies stipulate as well that the driver is responsible for any accidents. Once FSD is officially released, drivers with Tesla insurance will be not be liable for the accidents, the insurance will cover it. I'm not sure how quickly will the other insurance companies adjust to the FSD or other self-driving versions.

As to releases of new FSD versions (after it has been officially released) I wouldn't worry about new bugs appearing in them. Tesla will test regressively all new FSD versions on their existing and simulated data which should catch any bugs.

I don't think it is the proper place of Tesla engineers to redefine what a safe stop procedure is. Silicone valley does not decide public policy.

Your last statement is very optimistic.

[LesPalenik]

I don't think it is the proper place of Tesla engineers to redefine what a safe stop procedure is. Silicone valley does not decide public policy.

Your last statement is very optimistic.

I don't justify Tesla's interpretation of rolling stop. Although I think, the present full stop rule is idiotic and most probably designed to prevent accidents caused by idiots.

I spent some time developing testing software. You design a multitude of the tests based with specific inputs and expected results. Typically, every new software version will require additional test cases. You can then run the test program automatically every time a new software is released.

[Alan Klein]

I don't think it is the proper place of Tesla engineers to redefine what a safe stop procedure is. Silicone valley does not decide public policy.

Your last statement is very optimistic.

Agreed on both points.  I was at a four-way stop a couple of days ago.  I waived the other guy through first.  He had the right away since he was on my right. (We both hit the intersection at the same time.)  But he was just sitting there waiting for me. How does Tesla do that?  Does everyone just sit there like dummies waiting for the other guy to go first?   Maybe Teslas can be furnished with a fake hand that pops up to waive the other guy through.

[Alan Klein]

As to the rolling stop, the way Tesla implemented it, it was quite safe. Definitely safer than most humans drive through the stop sign.
Their "rolling stop" feature allowed vehicles to travel through all-way stop intersections at up to 5.6 mph under certain conditions. Drivers had to set their vehicle to a "full self-driving" style that included the feature. The vehicle's "chill" setting, the setting a user can select that drives the most conservatively of the options Tesla gives drivers, is not described as including the feature. The other settings, "average" and "assertive," include the feature.
The "rolling stop" function also required that no relevant moving cars or pedestrian or bicyclists were detected near the intersection. The function wouldn't activate without sufficient visibility, and all roads entering the intersection need to have a speed limit of no more than 30 mph. Tesla complied with the National Highway Traffic Safety Administration request and disabled the rolling stop option.

At this time, FSD has not been officially released, it is still in Beta stage, and the drivers who use it, are fully responsible for driving, steering and stopping.

Most likely, at this time, all insurance companies stipulate as well that the driver is responsible for any accidents. Once FSD is officially released, drivers with Tesla insurance will be not be liable for the accidents, the insurance will cover it. I'm not sure how quickly will the other insurance companies adjust to the FSD or other self-driving versions.

As to releases of new FSD versions (after it has been officially released) I wouldn't worry about new bugs appearing in them. Tesla will test regressively all new FSD versions on their existing and simulated data which should catch any bugs.

The driver is always responsible for the car by government statute not because the insurance companies say it.  Tesla or Tesla insurance can't waive it.

Additionally, the whole thing about who's responsible has not been tested in court as far as I know.  If a Tesla car runs me over because it didn't stop automatically at the stop sign, I sue the driver, the owner of the car, and possibly Tesla, the company. You don't sue the car.   A court will determine who's responsible and how much is their fault. 

The question is will Tesla honor the failure of their software and defend the lawsuit and pay the damages or is the driver on the hook?  What will other insurance companies do in these situations?  They're not going to cover errors in software by Tesla.  What will the insurance premiums be if a person is going to use auto FSD driving?  A couple of accidents and insurance companies will drop coverage across the country or charge premium rates for Teslas. 

That's why Tesla needs a hundred lawyers. 

[LesPalenik]

Agreed on both points.  I was at a four-way stop a couple of days ago.  I waived the other guy through first.  He had the right away since he was on my right. (We both hit the intersection at the same time.)  But he was just sitting there waiting for me. How does Tesla do that?  Does everyone just sit there like dummies waiting for the other guy to go first?   Maybe Teslas can be furnished with a fake hand that pops up to waive the other guy through.

That's a good example. Tesla FSD will slowly proceed if it doesn't detect any movement at the intersection (moot point now since FSD has disabled the rolling stop feature). I'm not sure if its decision will be different if it detects stationary objects there. OTOH, if one of the objects starts moving, FSD would be much quicker to detect that movement than a human.

As to the testing software, how much do you know about it?

[Alan Klein]

That's a good example. Tesla FSD will slowly proceed if it doesn't detect any movement at the intersection (moot point now since FSD has disabled the rolling stop feature). I'm not sure if its decision will be different if it detects stationary objects there. OTOH, if one of the objects starts moving, FSD would be much quicker to detect that movement than a human.

As to the testing software, how much do you know about it?

That removes defensive driving from many of the things we do.  When I get to an intersection with a four-way stop, I check the eyes of the other driver so I know he sees me before I pull out in front of him.  With FSD, that won't happen. 

I don't have to know how the software is written.  The question is how will Tesla handle this situation?  Will the software drive safely, legally, and effectively?  The responsibility is on them.   

Could you describe what the Tesla will do at a four-way stop?

[LesPalenik]

That removes defensive driving from many of the things we do.  When I get to an intersection with a four-way stop, I check the eyes of the other driver so I know he sees me before I pull out in front of him.  With FSD, that won't happen. 

I don't have to know how the software is written.  The question is how will Tesla handle this situation?  Will the software drive safely, legally, and effectively?  The responsibility is on them.   

Could you describe what the Tesla will do at a four-way stop?

As mentioned above, Tesla already disabled the rolling stop and complied with NHTSA request to perform full stop. The FSD monitors who stopped first and proceeds when is its turn. If the other dummy starts before it's his turn, FSD will put up with such a foolish action, stop and wait until the dummy leaves the intersection.