Hacker News Reader

I'm also of the opinion that automation is not compatible with safe driving. The issue is AI is able to handle the simple tasks first, trusting humans to takeover in outlier scenarios. But then, where does a new driver gain experience to handle the outliers?

We already went through this with airlines and decided to require pilots fly a certain number of hours without the automation, so when automation fails they are able to control the aircraft with some confidence.

As it is self driving cars are a luxury item for experienced drivers but I fear a future where a plurality of people panic when control is handed over to them once the computer determines the situation is too risky to continue on autopilot.

Some great backstory on automation in airlines is 99% Invisible's "Children of the Magenta":

https://99percentinvisible.org/episode/children-of-the-magen...

We don't let 80 year old pilots with bad eyesight fly planes. We don't let pilots fly planes 15/hours a day for multiple days in a row. Pilots must pass tests every year.

If we are going to hold car drivers to the same training and restrictions that pilots are held to we will have far fewer drivers and much safer drivers. But it seems very unrealistic to think that will happen in a democracy. Realistically it seems that technology is the most realistic way to reduce deaths.

The overwhelming majority of mayhem on the roads is caused by the long tail of bad decisions made by humans who would be able to pass whatever dumb requirements you concoct.

Kicking the elderly off the road, throwing the book at the one guy who gets caught doing NYC-DC and back in a day and onerous regulations on equipment is fundamentally a waste of resources chasing a long tail and is going to do nearly nothing to actually affect things relative to the expense. Most of the problems are caused by people doing things they knew were risky and stupid at the time and did them anyway, breaking rules in an irresponsible manner, distractions, intoxication, etc.

Being applicable to a broad cross section of the actual problem is why low hanging fruit type stuff like AEB has such a big impact.

Really, though, the most straightforward technology to implement to reduce automotive deaths isn't self-driving or driving-assisting vehicles; it's to reduce the number of cars on the road, and the number of miles driven. Given that people will still want to get places, it makes sense to replace cars with safer options such as trains (more buses would help, though not as much).

It would cost trillions of dollars to rebuild our cities in a way that car driving isn't necessary.

So? It cost trillions of dollars to rebuild them in such a way that car driving is necessary. Most cities had functional public transportation before cars reached the tipping point that pushed cities into car dependency[1]. In the mean time, a lot of slack could be taken up by buses, especially by converting existing lanes into dedicated bus lanes.

[1]: https://en.wikipedia.org/wiki/Car_dependency

We don't let them fly a commercial flight, but they absolutely could fly their own Cessna.

I’m not arguing we be equally strict, I’m arguing we don’t allow self driving to take over the majority of our hours at the wheel.

But yes, I also think we are much too liberal in handing out licenses and the roads could be twice as safe by removing the bottom 5% of drivers. At least require a portion of the driving test on expressways, as it is you do a written test that you know what the signs mean and you drive around the block proving you know the gas from the brake. Unfortunately because you can’t get around most places in America without a car, we consider it an inalienable right to drive.

We can also have people drive less and have more public transport and riding E-bikes etc.

Cars aren't the solution to all problems.

> The allure of having a car that drives itself is to relax the mind from paying attention to the road, handling 99% of the situations correctly through the software. But the 1% that require human input are made much more dangerous because the driver not only needs to handle the situation, they need to instantly catch up and familiarize themselves with the situation they've found themselves in.

There's also the separate recognition that no matter Musk's bombast on Twitter, Tesla's FSD stack is _nowhere near 99%_ correct handling. It's still absolutely regularly confused by bike lanes, only recently resolved roundabouts, and I could go on. But bottom line, not 99%. I'd be thinking closer to 60-75%, at best.

> The allure of having a car that drives itself is to relax the mind from paying attention to the road, handling 99% of the situations correctly through the software.

I have this with a Hyundai Palisade with L2.

I still have to watch the road. I still need to identify emergency situations and navigation. I still need to _assess_ many things - just like I would without any support. However, I don't need to take action on everything. This is mentally and physically freeing. On well marked highways, in particular, it means I can just cruise along without much worry.

The problem with Tesla is they set the expectation that you don't have to pay attention. They've set an expectation that the car is capable of _assessing_ every situation when it's clearly not. They further re-enforce this with the car's UI and behaviors such that drivers get lulled into not paying attention.

To me, it almost sounds like Tesla would have better UX and safety if they reframed the capabilities of the self-driving to make it clear that the driver still needs to be assessing.

> They've set an expectation that the car is capable of _assessing_ every situation when it's clearly not.

How did they do that ? The materials that new Tesla owners receive are specifying that the driver needs to pay attention in all circumstances.

From the https://www.tesla.com/support/autopilot page:

> Autopilot and Full Self-Driving Capability are intended for use with a fully attentive driver, who has their hands on the wheel and is prepared to take over at any moment. While these features are designed to become more capable over time, the currently enabled features do not make the vehicle autonomous.

>The materials that new Tesla owners receive are specifying that the driver needs to pay attention in all circumstances.

Can you please provide an example of material that owners actually receive that states as much? It's safe to assume that we can't expect everyone to read a support page on a website, just like how it's safe to assume that most people don't read the owner's manual when they buy a new ICE vehicle.

>How did they do that?

It's in how they promote it publicly. It's in how Musk constantly says it's "a year away". It's in the fucking name - "autopilot" has certain connotations within society writ large that Tesla can't just handwave away with legalese.

You are right, I can't give you an example as I don't own a Tesla. Yet I find it hard to believe that when someone is purchasing 45K+ of hardware and 10K+ of beta software they can't be arsed to read the support page for any of it.

I hope that people that pay these sums are responsible adults that don't go out there putting everyone in danger because "autopilot" is right there in the name, and of course it's a totally different thing than when the word is used for airplanes.

The irony is that to anyone who takes 2 minutes to read a wikipedia article, Autopilot is like cruise control or cruise control + lane assist. It is not "full self-flying" at all. However, too many people believe that a Tesla is years ahead of the rest of the pack when it comes to "full self-driving."

That's marketing for ya.

I found the example in the end, but I bet you'll think it's also part of those documents that you can't expect anyone to read. The subscription agreement[1] when buying a subscription to Tesla's FSM contains the following in the "Responsibility" section (sixth paragraph):

> Active Supervision; Responsibility. Full Self-Driving capability features require active driver supervision and do not make the vehicle autonomous.[...]

[1] https://www.tesla.com/legal/additional-resources#full-self-d...

> How did they do that ?

By going out of their way to call something FULL self driving that is, by their own admission in your quoted text, not full self driving.

You are right of course, but a lot of things in this world are named for the things that they will become, not the things that they are. If I was a Tesla client I would make sure what they mean by it before I took a nap in the driver's seat.

You can't call a skateboard a car because it may become one.

Can we agree that the name of FSD was selected before it actually was capable of anything? And if that's the case how should they have named it: "someday it might be full self driving"?

I totally understand your point, and I also agree with it, but I also can understand the point of view of someone as ambitious and driven as Musk that saw an idea, and then committed to it by choosing a name that reflects what they intend it to become, not what it's capable of doing at the moment.

You can see this very pattern if you watch Shawn Murray's first interviews about No Man's Sky. He shows the game as being capable of doing all these wonderful things that were proven to be bullshit when it was first released, yet over the years the game was built almost into the state that he first wanted it as. Was that an intentional way of cheating people, or was it just an ambitious person telling you about what they see at the end of road for their project? In my opinion it's the later, but people that bought into the hype at launch day, I bet they felt pretty angry.

So my advice (and apologies if it sounds patronizing), if anyone is in the market for buying a product which sounds too good to be true, maybe research the fine print and the owner's manual before putting your life and others' in danger by trusting the marketing team's selection of a name.

For many things, Owners manuals tell users not do to certain things while clearly designing a product to be capable of doing such things. Actions speak louder than words.

* Tesla has had multiple events highlight the "full self-driving capabilities".

* Their marketing indicates "full" self-driving while tucking the nuance away (for the owners manual).

* Their UI is specifically designed to highlight all of the things a Tesla is identifying. From what I've seen, it really doesn't indicate what a Tesla is confused by.

* Their driver engagement check systems clearly allow drives to be distracted and not paying attention.

* They've had multiple situations where accidents with stationary objects have occurred. This indicates a pattern of drivers believing these vehicles are more capable than they are.

Most likely scenarios:

-Driver stopped himself due to medical reason (driver was 74).

-Car's power-train issue caused it to stop. Less likely in dual-motor configuration. Even less likely to happen suddenly.

-Autopilot was on and driver fell asleep - Autopilot is programmed to come to a stop if driver input is not detected.

Because it's Tesla there's great deal of attention given to Autopilot but in this circumstance, where car stopped completely, driver most likely have been unable to make the car go (for medical or technical reasons).

> driver fell asleep - Autopilot is programmed to come to a stop if driver input is not detected

Hopefully it slows gently, signals a lane shift, and parks on the shoulder with hazards on, rather than slamming on brakes in the middle of the highway with cars close behind.

Of course, Autopilot gives driver visual and audio warnings before stopping, and it takes a while to activate. It does so because it's a Level 2 system and requires driver's supervision. In absence of supervision it has to stop as it's not autonomous.

GM Supercruise pulls over as I described: https://gmauthority.com/blog/2016/11/gm-super-cruise-technol...

Tesla brakes gradually and turns on hazard lights: https://www.youtube.com/watch?v=oBIKikBmdN8

As common sense suggests, a sleeping driver does not cause a Tesla to slam on its brakes in the middle of high-speed traffic. That is not the cause of this accident.

What makes you rule out an issue with the AI?

The system is not autonomous so it requires driver's supervision. Autopilot slowing down can be quickly overridden with a gas pedal, yet the car came to a complete stop.

What leads you to think that the AI could not bring the car to a complete & abrupt stop?

Because the car is not autonomous. Driver has full control of the car and can override Autopilot at any time, especially such obvious maneuver as slowing down. In this case, there's no evidence to even suggest it was an abrupt stop.

> In this case, there's no evidence to even suggest it was an abrupt stop.

That's exactly what happened based on this article, assuming we agree that "slam on the break" means an abrupt stop.

"At first, the Independence Police Department speculated that a mechanical issue caused the Tesla to lose power, but officials are now saying they are unsure what made the vehicle stop unexpectedly [...]

Last month, the National Highway Traffic Safety Administration launched a preliminary investigation into Tesla after hundreds of owners reported that their vehicles would sometimes unexpectedly slam on the brakes for no apparent reason. According to the agency, these incidents, known as "phantom braking," sometimes happened at highway speeds.

Sudden braking occurred when owners had switched on cruise control or Autopilot, a system that automatically brakes, steers, and accelerates to aid with highway driving. NHTSA is investigating the phenomenon in Model Y and Model 3 vehicles from the 2021 and 2022 model years. Owners of 2019 and 2020 vehicles have also complained to NHTSA about phantom braking."

'phantom braking' slows down the vehicle, it does not stop it and can be overridden, at any time, by a driver by pressing accelerator. None of the incidents reported to NHTSA resulted in an accident, but braking is unexpected and unwelcome - likely reason why drivers reported it.

Given that earlier reports suspected loss of power - stopping could not have been abrupt. We don't even know if the car was on Autopilot, but we do know that driver was 74 years old, with higher chance for medical emergency.

The standard argument in favor of FSD vehicles is that they perform better than the average human driver. Except it doesn't take into account the amount of adrenaline involved.

I do not see the point of conversing about full self driving when it is nowhere close to existing yet.

Tesla use this names unforunetly "FSD" and "Autopilot", instead of lane assist or driver assist... we need everyone including fanboys to demand Tesla stop push fake capabilities, maybe name it MFFSD = Maybe in future Full Self Driving

Yes, that was the point of my comment. Tesla does not unfortunately use the words “full self driving”, they intentionally use the words “full self driving” for marketing purposes to mislead people. How it is not a clear case of false advertising for regulators to go after, I do not know.

I could even envision the plausible deniability/marketing bullshit behind “self driving”, but to go out of your way and declare FULL self driving, and not get smacked for false advertising is beyond me.

It also doesn't take account how much the average is skewed by a minority of really terrible drivers.

And the issue is the numbers are fake: different population, different car types are compared. Dis-engagements are not noted as "20K dis-engagements = 20K times the human saved himself from the bad AI.

Would people get operated by an AI doctor that is better then the medicine students and the drunk doctors but only when a good doctor is paying attention and intervenes in case the AI shits itself?

While AP is sometimes touted as "FSD safer than a human driver" in reality it's just a set of driver assists that can help a driver drive safer. Since AP doesn't drive itself except for very limited definitions of driving, it can't be itself safer than a human driver. Like when a pair of sneakers help a runner bit can't logically be called "better than a human runner". Let's not forget that the human does most of the work in either example.

As long as any "self-driving" car operates only under human supervision it's not relevant to compare that system taken independently to the human. Compare the FSD alone with the human alone and then we have stats. Everything else is candy for people with more money than sense.

Do we know this was the result of FSD and not another malfunction, such as a seized component, power failure etc? Seems we should not speculate if we don't know the RCA?

No, the article says police speculate it was mechanical failure and loss of power, but I'm sure we'll see plenty of rigorous, unbiased followup in the following weeks.

It's funny to see the pro and anti Tesla crowd arrive at truth through verbal combat, at least.

Let's make sure we wait for the investigation to be done before accusing the AI so quickly.