I'd like to thank Tesla for making this happen. They mislabeled their own self driving product and pushed it to market too early. Now everyone is afraid of something that should be safer than the status quo.
I would like to point out that the criteria of "I don't want to see a single story in the news about self-driving accidents" is equivalent to an infinitely high standard that would immediately disqualify all humans from driving if applied evenly.
Unrelated to what he said. Tesla is at fault not because their system as flaws, they will always have, but because their marketing strongly implies it's something different than what it is, people believe the marketing (like in any market), and so for people it turns out the system does not work, since they compare it to expectations.
I’ve talked to some people this year who thought Musk in 2017 was talking about current capabilities when he was talking about summoning his car to LA from NYC.
It's unrelated to what he said now -- I should have quoted the post. In any case, yeah, the marketing is exaggerated, but it nags you if you stop paying attention. From a safety perspective, how does that not keep expectations in check?
Because it's not about MY expectations, it's about the expectations of the guy behind the wheel of the car crossing my path.
Therefore I don't need it to be good enough to convince ME, I need to go be good enough to convince HIM not to abuse it, and given the large variety of him out there the system needs to have a very high safety bar.
The guy behind the wheel gets beeped at if he stops paying attention and shut down if he keeps it up. Are you really arguing that this is insufficient to set the expectation that he should be paying attention? What, in your view, would be sufficient?
This is a silly game without data. I'm open to the idea of escalating consequences to effect deterrence, but we need data, not "+1 forever" to figure out the correct threshold. In any case, that's a completely separate argument. My claim is that the current system seems more than sufficient to reset any mistaken expectations due to marketing. You don't seem to disagree, so I rest my case.
> My claim is that the current system seems more than sufficient to reset any mistaken expectations due to marketing. You don't seem to disagree, so I rest my case.
I do disagree, I just don't know the solution, this not being my job nor my field.
As for the first part of your message, if you cannot see a gigantic differences between a matter of minutes, and just a few seconds, when talking about emergencies and the need to figure out what's happening to react in one or two seconds, then this is not a productive conversation.
You clearly don't know how the Tesla lockout works, can't be bothered to learn before making up your mind that it sucks, and can't even be bothered to learn after getting called out on it.
No, this will not be a productive conversation. Goodbye.
No criteria will ever be “good enough”. Humans don’t evaluate their own mistake/crash/kill rate for strangers with the same reasoning and logic and emotion that they evaluate an outsider’s. Automated cars are an outsider, and so they’re held to an impossible standard: Zero kills, zero crashes, zero mistakes.
Tesla absolutely hurt the chances for AI driving, but it’s not one hundred percent their fault that the US population is xeno-hostile to robots. They would have been better off adding a Clippy chatbot to their cars than self-driving, because then people would be excusing their car’s mistakes as they would another human’s.
My issue here is the labeling. Tesla sold their product under the name "Full Self Driving", which I think implies level 5 self driving. But what customers actually got is level 2 self driving.
"Now everyone is afraid of something that should be safer than the status quo."
I agree when talking about it at a population level. But that still doesn't mean much. The status quo is extraordinarily low, mostly due to people's lack of knowledge or bad decision making. These autonomous systems can address part of this on the reactive side (object in the way, apply brakes). It's still lacking in the proactive side (someone is weaving, maybe I should leave some extra following distance; I'm about to be rear ended at a light, let me pull ahead if its safe).
So yeah, the vehicles are better drivers than the worst people, probably better than average. But I'm not convinced they're safer than the best drivers, especially when looking at overall accident avoidance and not just accidents they're liable for.
School bus drivers are safer than airline pilots per exposure hour, in fact they are safer than walking.
The big problem with driver assistance products is the automation bias issue with human drivers. Mixed in with humans being really bad at maintaining attention while bored.
We see this even with highly trained airline pilots with backup crashing due to automation bias.
Despite the common myth, drunk drivers have higher fatality rates than those who are hit.
Selling driver assistance as self driving or allowing it to be used without serious attention detection safeguards won't make distracted or impaired drivers safer.
I'm fact for first responders, highway workers and pedestrians, any form of driver assistance that allows drivers to checkout of the driving task, they will be far less safe.
The only way that these companies claim that there systems are safer today is the they disengage quick enough to claim they weren't at fault but without time for the distracted drivers to respond.
By ignoring automation bias, which is a well known human flaw, they pad their numbers and make them seem safer, while not having enough data to even make statistical claims that they do.
Outside of naive beliefs that autonomous cars are safe driving confirmation bias the common myth that self driving cars are safer is baseless.
I get that is the goal, but assuming it is true without evidence is part of the reason the public is starting to doubt the claims.
It’s not Tesla’s fault that CBS runs a story every time someone crashes their Model Y. If you’re reading about a Tesla crash odds are it wasn’t on autopilot. The news is happy to run the story either way because it generates clicks.
It is Tesla's fault that they 1) apply the label "Full Self-Driving" to a system that is not even close to full self-driving, 2) push extremely-untested beta software out to millions of cars on the road where every other company proceeds with extreme caution and 3) try for camera-only sensing even though literally nobody in the industry except Elon Musk thinks this will be good enough. Anyone could have predicted these three things would eventually shatter public self-confidence in self-driving cars, which is exactly what has happened. Stop making excuses for them.
It is full self driving, the car computer is controlling steering and acceleration. That doesn't mean it's autonomous, there must be a human driver monitoring and responsible for its operation at all times.
They absolutely test releases ahead of time. They test on more data & better data than anyone else and they dogfood every version before wide release, which they still call a beta. They used to gate it behind a safety score until it started to get good and they still have not-paying-attention detection, nags, and lockouts.
But yeah, if you don't consider any of their testing and precautions I guess they don't have any testing and precautions.
If the software is so dangerous where are all the crashes? FSD beta is running on 400k+ vehicles. I’ve got it running on my car and my biggest complaint is that the software is too hesitant.
You can complain about Tesla’s methodology all you want, but when the rubber hits the road they do an excellent job of maintaining driver attentiveness, which all but guarantees safety.
If public confidence is shattered it has very little to do with the quality of Teslas software and everything to do with the currently fashionable narrative in media.
Someone I know drives an M series BMW and was considering a Tesla switch because his close friend in LA told him the autopilot is so good, he can now routinely drive home drunk (this was a guy in his mid 30s with a kid). As a result, my friend now also had confidence autopilot was reliable, no matter what I told him.
So, relative to the confidence most non-technical people seem to have had, this percentage is still not high enough.
Like most tech adoption curves, the hype here is still too high relative to reality. However, if you are someone who actually understands L1 - L4 scale, I think you should strongly correct people who don't.
Believing in the hype in crypto might lose you some money. Believing in the hype in self driving might lose you a life.
DUI checkpoints are set up to only catch the stupidest of criminals. At least in my neck of the woods, they publish the locations of the checkpoints in advance, and are always on arterial roads, but never state highways or interstates. It's entirely possible to avoid the checkpoints altogether. If you're stupid enough to get wrapped up in one of those, you probably shouldn't be drinking or driving.
I once got pulled over because I was using autopilot. And I was stone cold sober. Damn thing just couldn't figure out how to do easy corners on I-5 without ping-ponging. Turns out there was a cop behind me.
Fortunately it was resolved quickly with chuckles all around.
They're probably pretty close to right. When I think back to impaired drivers I've noticed on the road, the first two things I've noticed were inadequate lane-keeping and wild/baseless speed changes. Autopilot is good enough to mask those two signals. (I'm not saying this is a good idea; I'm just saying they're probably correct in that it reduces their chance to be pulled over.)
In urban environments in high-density countries like South Korea it's fairly pervasive that you can use a phone app to call a driver. They will drive your car home for you.
That would be urban environments in high-density countries with low wages. That driver needs to get to the car and home and someone needs to pay for that.
I don't in any way support drunk driving, but I do think it's unfair to frame it as "a few bucks." That would only be true if you were within walking distance and sharing a cab with others. For most of the situations I've been in, it's more like $60 plus a big gamble that there will be any rides available. In many areas Uber/Lyft are the only options and during low hours (particularly early morning 12 to 2) there aren't any drivers.
I would agree if the buying the car was only for driving home from the bar. That seems unlikely. I would guess that in nearly all cases, the person already owned the car before they even decided to go to the bar that night.
> So, relative to the confidence most non-technical people seem to have had, this percentage is still not high enough.
My confidence in self driving cars is boosted by my extremely low confidence in people. I know too many drivers who have no business being on the road sober, and other people who routinely drive drunk. I'm confident self driving isn't there yet, i'm also pretty sure that it won't be too long before it gets better than the average driver.
Self-driving is not going to be better than the average driver for a long time.
But 2 things.
1) average doesn't matter.
There's unlikely to be extremely bad self-driving cars.
2) the road & nature of transportation will adapt to self-driving cars so that they don't need to be as good as they need to be on current roads, and eventually it won't matter (but that's like 20 years in the future).
If you would've looked at cities when the car was invented, you'd have thought that cars would never catch on. Cities adapted to the car. They'll adapt to self-driving. But it's going to take a long time.
> 2) the road & nature of transportation will adapt to self-driving cars
Are you kidding? Most of this country can barely keep potholes patched for the benefit of regular cars. Proposing major infrastructure spending to accommodate luxury car features is surely a joke.
And then there is the average driver being distracted -- by cell phone, by alcohol, by passengers in the car, from road rage.
Frankly, the average IQ is 100; self driving should be able to surpass that in reaction time, physics (following distance), and legality; so I think it's likely 75% better than than the average driver. But 100% better than the 50% that are distracted.
You're throwing out a lot of numbers, can you back them up?
There's an old joke that something like 75% of drivers think 75% of everyone else is a bad driver. This is just a demonstration of perception. It only takes one bad driver per lane to really screw up traffic. The correct answer is that the vast majority of drivers are quite competent, but you don't notice them.
It's still not clear to me why there hasn't been more discussion about dedicated lanes and roadside sensors to make self-driving safer & more reliable.
Even if the tech grows significantly in capability, the unpredictability of other drivers and the ways snow affects roads seem to me to make these kinds of assists necessary.
The average driver is pull down by a lot of people who: 1) are drunk 2) speed unnecessarily 3) drive shitty cars 4) drive distracted by tech 5) drive tired
If you eliminate those factors, then major car crashes get reduced to almost nothing.
That is what the autopilot is competing against, not the "average" driver. Until it gets better than a human that is sober, rested and drives defensively, autopilot won't take off. We are a long way from that.
Exactly right. I don't want a self-driving car that is "average safe" because I am never "average drunk" when I drive myself. I refuse to drive after even one beer earlier that day, it's a decision I make for myself; I choose not to be an average driver. Why should I pretend that I am not myself, but rather am an embodiment of a statistic with no free will of my own?
inb4 "everybody thinks that", such inane arguments don't make me drunk.
Also, 6) drive at night, and 7) drive in terrible weather.
Even with all the people who drive dangerously, driving is remarkably safe given how many miles we drive. As you say, take away the outliers and it's very safe. And an autonomous system won't gain acceptance among the majority of drivers who don't take these risks until it's conclusively better than them.
The saying is "it takes two to have a car accident", and I found out over the years that it's mostly true, I've forgiven a lot of mistakes by other drivers, and viceversa.
It's also the cultural dimension to this: we as humans are very good as establishing a culture or fitting into a new culture when new to it. We are very good at predicting what the other drivers are doing, and we can avoid a lot of incidents that way. When could the AI have a perception of culture change, and say an AI trained in SF would have a hard time adapting in DC or New York - all with different styles of driving. Not to mention moving to other countries...
But we can't eliminate those factors; Indeed, those factors apply to 95% of American drivers.
(you left out Angry, Old)
- 43% of Americans — and 56% of Men — Admit to Drinking and Driving
- 60% of American drivers admit to texting while driving, and 70 percent admit to using a handheld cell phone while driving – 43 percent say they use a mobile device while driving as a habit.
The Amish have been reminding Americans what they miss out on by ignoring their past. In fact anyone with a horse and likes a drink at the pub, knows its best to the take horse and not the car!
California's Vehicle Code (VEH) would still allow for a DUI charge when riding a horse. According to urban legend (and a lot of dubious websites), there are cases where Californians have been convicted of a DUI on a horse, though I haven't been able to find reputable sources for those claims.
VEH 1.670 [0]
> A “vehicle” is a device by which any person or property may be propelled, moved, or drawn upon a highway, excepting a device moved exclusively by human power or used exclusively upon stationary rails or tracks.
VEH 11.12.23152(a) [1]
> It is unlawful for a person who is under the influence of any alcoholic beverage to drive a vehicle.
IIRC many states other than just CA will give you a DUI for driving a lawn tractor drunk. On the road, at least. And in plenty of places just being in the car drunk is enough, even if you're in the back seat trying to sleep it off.
If you're capable of operating the vehicle, it can still count. A driving instructor once told me that if you have to sleep off drunkenness in the car, find a place to hide the key (ideally outside the car) so you are not considered capable of operating it.
Really, these laws are pretty absurd, but drunk driving kills a lot of people so here we are.
hah - but having grown up in rural Wisconsin - the rural highways there are full of drunk drivers because there are no ubers or taxies out there, and full of amish buggies that are nearly invisible at night (they just have one reflector on the back). It is a terrifying combination. Just have your drink at home.
This is more of an indictment of alcoholism than self driving cars. Anyone who’s routinely getting drunk has a big problem that will cloud their judgement.
I'm curious what the split is on comfort with an autonomous car on city vs highway driving.
I know I'm personally uncomfortable with most stuff on the road right now in dense urban environments (Waymo, Cruise), yet a little more comfortable with highway autonomy (BlueCruise, Autopilot, Comma.ai).
You don't really need any of this advanced autonomous tech for highway driving. Most standard passenger cars today come with adaptive cruise control and lane keep/change assist. I have a Hyundai that can safely drive itself on the highway indefinitely with only occasional touches needed from me.
Also a full time pedestrian. While I have near-zero trust for drivers, autonomous cars are not the savior.
At an intersection you can usually see where the driver is looking to help inform your frogger-like quest to not die. There's nothing to look at in an autonomous vehicle. You are operating on hope that the OS was designed to see you and stop accordingly.
I wonder if self-driving cars are going to be the second technology (after nuclear power) de-facto killed or at least stymied by demanding unrealistic/uneconomic safety standards.
Our media overhype fear and the civilizational consequences may be dire.
While possible, the reports I've seen in the media suggest that Google/Waymo's are actually safe, Tesla's is only at the level of the average American driver on all roads despite being mostly limited to the safest roads, and Uber's is unsafe at any speed.
(The media I follow doesn't spend much time on any other car AI).
So for now, even though that could happen, that's like worrying about overpopulation on Mars.
However, way less than I'm afraid of any random idiot human who may or may not have passed a driver's test on his 5th attempt in some random state, driving at whatever diminished capacity (tired, distracted, drunk, etc.) in the worst weather conditions, who is angry at the car in front of me for some reason, driving on the highway near me.
I mean, those two things aren't mutually exclusive.
You can easily have the worst of both worlds, where somebody disables the autonomy to engage in some road rage, then re-enables it, but without driver awareness features (lots of videos of people doing this) so they can have a nap and not take over when the autonomy disengages.
So you can easily have all the disadvantages of autonomous cars and idiot reckless driving.
Only time we'll be free is if there's ever L4/L5 autonomous driving, which seems a long way off.
edit: even "full self-driving" in Tesla's are L3, since you need to be available to take over, driver awareness is enforced, etc.
And we can thank Tesla for that, not even because of the issues of their system there will always be some, but because of the issues with their PR and marketing.
If you promote something as "autopilot" and "full self driving" and "safe" and yes sure the fine line says you need to stay alert and ready to act in under two seconds but the marketing really implies it's just a regulatory thing, and then you have said system kill people because it's not self driving at all it's drive assists, then people lose confidence.
Hell, even people who are fully aware of that are getting worried now, not because they don't understand what it is or isn't, but because they worry the person behind the wheel might be misunderstanding.
You don't think the news media deserves any thanks? It's well known that the media locks on to anything that causes fear as it's the number one driver of clicks/attention, and it's been a deluge all over the incidents the last couple of years.
I'm not making a value judgment on whether this is good or bad (I truly don't know), but it is important to know.
Not really, nobody was scared of line keeping cars despite them having essentially the same risks, because nobody ever thought they could close their eyes or play on their phones or watch a movie while line keeping is enabled. Some people did, and crash happened, and people died, but when it did everyone saw it as "another idiot in a car endangering people", not a failure of the car's system.
It's not like they were hammering on some obscure startup. The world's richest man promoted FSD at every opportunity for years and sold it to customers.
The car is self driving, while the driver monitors and is responsible for its operation at all times. How is that a difficult concept? The car tells the driver this every single time autopilot is engaged.
Why do people work so hard to intentionally misunderstand what self driving and autopilot mean? (they don't mean autonomous)
> Why do people work so hard to intentionally misunderstand what self driving and autopilot mean? (they don't mean autonomous)
Because of statements like this: "The person in the driver's seat is only there for legal reasons. He is not doing anything. The car is driving itself."
Because the ways that real humans use the system. People defeating the wheel and texting is commonplace.
It is just reliable enough to disable human vigilance and then be really unsafe.
It seems to me you're the one working hard giving humans far too much credit for being as aware and cognizant of things as you are. People buy Teslas because they are fashion statements, not because they are anal super detail-oriented Type A personalities.
Like I said, there is a difference between the facts, that you state, and how the PR and marketing are presenting them.
You don't even need to go far at all, here are the very first words, right now, on tesla "autopilot" page [1]:
> Tesla cars come standard with advanced hardware capable of providing Autopilot features, and full self-driving capabilities
You cannot in good conscience tell me that this doesn't implies something different than reality, and please don't be one of those that try to claim "no, the word autopilot doesn't ring any bell in people's head they don't have any preconception".
You need to go all the way down half the page, a full 4 scrolls down on my 1600px tall screen, to see "Current Autopilot features require active driver supervision and do not make the vehicle autonomous."
So yes, like I said, the fine line always give the trick. But the marketing pushed forward implies something different, and if marketing didn't work we wouldn't be doing it.
The same way that cruise control = self accelerating car, autopilot = self driving car. In both cases the human is using a tool to operate the vehicle.
Are you asserting that other systems like drivepilot, pro pilot assist, or blue cruise are safer and promote a better understanding of their capability because a few letters are different in the marketing name?
I am asserting people have a set idea of what the word "autopilot" means, and it doesn't mean "unable to pilot on its own without me being fully attentive to it at all times". And I am also asserting that this name was chosen specifically for this implication.
> Why do people work so hard to intentionally misunderstand what self driving and autopilot mean? (they don't mean autonomous)
Self driving is not autonomous? I think maybe your definition is quite different from Average Driver, and so the intentionality may be more yours than theirs.
Auto nomous:
Not controlled by others or by outside forces; independent.
Independent in mind or judgment; self-directed.
Independent of the laws of another state or government; self-governing.
It's more than that. Elon Musk has made himself a lightning rod with all his crazy person antics. People want to pile on him and if a video of a tesla going haywire helps them do that then it gets spread around everywhere far beyond normal public interest in such a thing.
I disagree, I think this is a very tech community centric point of view. You will find most people outside of this circle cares very little about Elon Musk when thinking "do I worry or not about autonomous cars when going to buy groceries or crossing the street with my daughter".
Why is this downvoted? This is an important point. The sample is highly likely to be non representative, yet many people are drawing conclusions from it.
The relative increase mentioned in the article is still interesting news in itself, but many conclusions being drawn are far more than just the increase in the same sample.
After seeing someone just blow through a stop sign in their normal non-autonomous car on my walk from the store this morning, I don't see how autonomous cars could be any worse.
I would agree and have always felt like it would ultimately be the insurance companies that end up "mandating" autonomous driving, but... I have a Tesla and after a year with the FSD "beta" I finally tried it out. OMG, that was scary. It works fine until you come to a place with the slightest bit of driving ambiguity (and those abound in the metro Boston area) at which point it becomes a vacillating bundle of indecision not knowing whether to go left, right, forward or slam on the brakes, so it basically tries them all in short order. I don't drive drunk, but I can tell you that if the choice was me driving drunk and letting it take over anywhere other than a relatively clear interstate road, I'd drive every time.
Yep... I never use mine anymore. I rarely even use "autopilot" anymore after it almost killed me due to a rapid change in lighting conditions the cameras couldn't compensate for (but didn't let me know it was effectively blind lol).
The worst part is that they claim it crashes less than humans... neglecting to account for the fact that you obviously disable it in difficult road conditions / areas / situations (or it disables itself, eg in foggy conditions).
So the samples are obviously not remotely comparable. Bugs me to no end that they grossly misrepresent the stats. There are other tidbits like FSD returning control to the driver if it detects an imminent, unavoidable collision, to reduce legal liability. Again, this makes the whole "FSD is safer!" spiel so slimy.
I'm afraid of all cars! The environmental impact, the personal economic impact, the amount of space they take up that could be parks or farms or whatever.
dont forget the raw physical risk of hurtling around in a thin can at speeds well in excess of what could kill you, just one mistake away from smearing yourself, your tin can, and those of the 10 other people doing the exact same thing mere feet away.
or how with a mere moment of inattention could you seriously disable or kill some random person who isn't even playing your high risk game.
not only did we normalize this - we mock anyone who takes exception to it
I used to regularly cross the old Goethals Bridge between NJ and Staten Island early in the morning, maybe 3-4am. At that time there was volume (trucks and cars) AND speed. The lanes were ten feet wide, and tractor trailers are 8.5 feet wide. No shoulder. I look back and have no idea how I did it without freaking out.
I'm more worried about the idiots driving 80mph and texting. Police don't seem to care or they are just so overwhelmed they have given up. I would welcome autonomous cars.
I agree, and sooner or later almost everyone will have one - even the people who don't want one - because, imo, the insurance cost on a regular car is going to skyrocket, and fsd will be the only affordable option for many people.
With all the trips I did back and forth to my kids schools for many years (60 minutes round trip), I would have loved the ability to just send the car to pick them up after school, or get them there when they missed the bus.
Too late for my kids, but my guess is by time I am too old to drive myself I will have FSD option available instead.
You're comparing the risks quantitatively when you should be comparing them qualitatively.
We have a lot of decades of understanding of the extent and surface area of the risk around human drivers and non-autonomous car technology. The technology and the norms around safety and regulations came about in a different time, when it was less perverted by profit and the chaos of software bugs.
We have no data about the safety of autonomous vehicles at scale, in the full spectrum of driving conditions and scenarios.
And there's the extra nuance that any change to a model or algorithm could have regressions for an entire fleet at once that at best invalidate any safety metrics we learned from the previous version, and at worst make the whole fleet unsafe overnight.
We have metrics that tell us roughly how many drivers will be distracted or drunk on a given day. It's not perfect, but we evolved an equilibrium that makes us feel safe enough about getting on the road. And we have accountability for the people who are unsafe.
You might not even distrust autonomous car technology, but it's logical to distrust the new forces of unaccountability for the technologists who inevitably introduce regressions that kill people at scale, the corners that will be cut in today's version of capitalism, laws being disregarded by technology giants because they can get away with things that automakers in a more innocent time couldn't.
I am eagerly looking forward to a day when all vehicles (or the vast majority) are autonomous. Not only would that allow drivers to get back all the time they would otherwise devote to operating their vehicles, but also it opens up a world of accessibility to blind people, the elderly, and others who (for whatever reason) can't drive themselves right now.
And there are network effects! If practically all cars are autonomous, they can talk to each other in sophisticated ways that are not feasible for manual operators, who largely rely on very rudimentary signaling (e.g. turn signals, flashing your brights, one-finger salute, etc.). This is fantastic for both safety and efficiency.
That said, what I think we should fear right now is not autonomous cars themselves, but misplaced expectations by drivers about how reliable the technology is right now. As long as everyone accurately understands their limitations, we can both limit our own risks and prevent the sort of backlash we're seeing now.
We're going to have to isolate those cars to separate roads that don't have pedestrians on them. Of course, those roads will need entry and exit points for people, but it's a lot safer if they are mostly isolated.
And maybe we can have those cars join big convoys that are traveling to the same place so they save energy by speeding up and slowing down as a group.
And then eventually we will start calling those autonomous cars "trains."
but the infrastructure for trains is too costly. what if we could reuse the existing infrastructure and everything else you said. we could call these autonomous vehicles 'busses' or something like that.
It's not good for efficiency because you still have to keep those rudimentary signals for non-digital users of the streets. Like, you know, humans. While ths might work on closed tracks, it also means that "all vehicles" will never be autonomous, and unless you ban vehicles from cities, you'll not get close to "vast majority" either, so you'll never reap the complete benefits of this efficiency.
I tried to be open minded, but after my friend demoed his FSD "beta" to me I want these things banned. It drove worse than a drunk student driver and I legitimately feared for my life and made him stop after only a few minutes.
"Beta testing" such things on the general public is morally abhorrent. Let Musk by some land in a desert and build his own mock town to test his toys, but get it off the public roads.
Enabling FSD Beta is categorically irresponsible. To hear him praise such an abysmal product definitely altered the way I consider him. He's still my friend but I'm now aware of a new dimension to his personality; he has poor judgement when it comes to fancy gadgets. I now see that he is prone to getting swept up in futurism and gadget consumerism. I told him this and we agreed to disagree, so long as he never enables that shit with me in the car again.
I don't need to square anything, my experience is what it is. It nearly drove us into other cars several times in the span of only a few minutes. It's total trash.
You'll get no argument from me. Mine is in the FSD beta, and I never use it after similar experiences.
The worst part is that for some people, FSD does seem to work exceptionally well.
But I am in rural Canada, and the thing is a death trap here. I imagine in wealthy pockets with an abundance of Teslas (and subsequent data/attention, well maintained roads, mapping, etc.), it probably does handle really well.
You can just look at what the imaging is picking up and displaying on the console to see this stuff is just not ready. Mine rapidly flickers between trucks, cones, pedestrians, etc. The traffic lights do this too. I imagine it is the same on all Teslas, so even at the image detection level something is really underbaked.
As they should be. I'm afraid of what Tesla's marketed as 'autonomous' and the precedent their ability to get away with it sets for the day we hand over a greater deal of trust to these systems.
Who needs regulation when you've got litigation? Surely that's the foundation of a reasonable government - let the bodies pile up and sort it all out in court.
When the consumer is the unwitting beta-tester and the corporate policy is "ship fast and break things" well the consumer starts to grasp they are the thing that is going to break.
123 comments
[ 4.7 ms ] story [ 127 ms ] threadI would like to point out that the criteria of "I don't want to see a single story in the news about self-driving accidents" is equivalent to an infinitely high standard that would immediately disqualify all humans from driving if applied evenly.
Therefore I don't need it to be good enough to convince ME, I need to go be good enough to convince HIM not to abuse it, and given the large variety of him out there the system needs to have a very high safety bar.
Mercedes's system, ensuring that the cars does not disconnect for a minute (instead of seconds) feels safer to me, for exemple.
"I think two minutes isn't safe enough. They should lockout for three minutes!"
"I think three minutes isn't safe enough. They should lockout for four minutes!"
...
https://www.youtube.com/watch?v=xJxwvZ5SE_c
This is a silly game without data. I'm open to the idea of escalating consequences to effect deterrence, but we need data, not "+1 forever" to figure out the correct threshold. In any case, that's a completely separate argument. My claim is that the current system seems more than sufficient to reset any mistaken expectations due to marketing. You don't seem to disagree, so I rest my case.
I do disagree, I just don't know the solution, this not being my job nor my field.
As for the first part of your message, if you cannot see a gigantic differences between a matter of minutes, and just a few seconds, when talking about emergencies and the need to figure out what's happening to react in one or two seconds, then this is not a productive conversation.
No, this will not be a productive conversation. Goodbye.
Tesla absolutely hurt the chances for AI driving, but it’s not one hundred percent their fault that the US population is xeno-hostile to robots. They would have been better off adding a Clippy chatbot to their cars than self-driving, because then people would be excusing their car’s mistakes as they would another human’s.
I agree when talking about it at a population level. But that still doesn't mean much. The status quo is extraordinarily low, mostly due to people's lack of knowledge or bad decision making. These autonomous systems can address part of this on the reactive side (object in the way, apply brakes). It's still lacking in the proactive side (someone is weaving, maybe I should leave some extra following distance; I'm about to be rear ended at a light, let me pull ahead if its safe).
So yeah, the vehicles are better drivers than the worst people, probably better than average. But I'm not convinced they're safer than the best drivers, especially when looking at overall accident avoidance and not just accidents they're liable for.
The big problem with driver assistance products is the automation bias issue with human drivers. Mixed in with humans being really bad at maintaining attention while bored.
We see this even with highly trained airline pilots with backup crashing due to automation bias.
Despite the common myth, drunk drivers have higher fatality rates than those who are hit.
Selling driver assistance as self driving or allowing it to be used without serious attention detection safeguards won't make distracted or impaired drivers safer.
I'm fact for first responders, highway workers and pedestrians, any form of driver assistance that allows drivers to checkout of the driving task, they will be far less safe.
The only way that these companies claim that there systems are safer today is the they disengage quick enough to claim they weren't at fault but without time for the distracted drivers to respond.
By ignoring automation bias, which is a well known human flaw, they pad their numbers and make them seem safer, while not having enough data to even make statistical claims that they do.
Outside of naive beliefs that autonomous cars are safe driving confirmation bias the common myth that self driving cars are safer is baseless.
I get that is the goal, but assuming it is true without evidence is part of the reason the public is starting to doubt the claims.
But yeah, if you don't consider any of their testing and precautions I guess they don't have any testing and precautions.
You can complain about Tesla’s methodology all you want, but when the rubber hits the road they do an excellent job of maintaining driver attentiveness, which all but guarantees safety.
If public confidence is shattered it has very little to do with the quality of Teslas software and everything to do with the currently fashionable narrative in media.
So, relative to the confidence most non-technical people seem to have had, this percentage is still not high enough.
Like most tech adoption curves, the hype here is still too high relative to reality. However, if you are someone who actually understands L1 - L4 scale, I think you should strongly correct people who don't.
Believing in the hype in crypto might lose you some money. Believing in the hype in self driving might lose you a life.
It's amazing to me what people will do to save a few bucks on taxi fare.
The autopilot doing its habitual random shenanigans that makes it look exactly like a drunk driver is at the wheel
You being asleep at the wheel
&c.
One of the nice things about living in the PNW is that both Oregon and Washington prohibit such unconstitutional crap.
Alcohol is responsible for 30% of road deaths here, 23% due to other drugs. That's a lot of death you can avoid with a 30 seconds spit test
Fortunately it was resolved quickly with chuckles all around.
That would be urban environments in high-density countries with low wages. That driver needs to get to the car and home and someone needs to pay for that.
> So, relative to the confidence most non-technical people seem to have had, this percentage is still not high enough.
My confidence in self driving cars is boosted by my extremely low confidence in people. I know too many drivers who have no business being on the road sober, and other people who routinely drive drunk. I'm confident self driving isn't there yet, i'm also pretty sure that it won't be too long before it gets better than the average driver.
But 2 things.
1) average doesn't matter.
There's unlikely to be extremely bad self-driving cars.
2) the road & nature of transportation will adapt to self-driving cars so that they don't need to be as good as they need to be on current roads, and eventually it won't matter (but that's like 20 years in the future).
If you would've looked at cities when the car was invented, you'd have thought that cars would never catch on. Cities adapted to the car. They'll adapt to self-driving. But it's going to take a long time.
Are you kidding? Most of this country can barely keep potholes patched for the benefit of regular cars. Proposing major infrastructure spending to accommodate luxury car features is surely a joke.
And then there is the average driver being distracted -- by cell phone, by alcohol, by passengers in the car, from road rage.
Frankly, the average IQ is 100; self driving should be able to surpass that in reaction time, physics (following distance), and legality; so I think it's likely 75% better than than the average driver. But 100% better than the 50% that are distracted.
There's an old joke that something like 75% of drivers think 75% of everyone else is a bad driver. This is just a demonstration of perception. It only takes one bad driver per lane to really screw up traffic. The correct answer is that the vast majority of drivers are quite competent, but you don't notice them.
https://www.driveresearch.com/market-research-company-blog/c...
https://www.valuepenguin.com/drinking-habits-survey#under-in...
Here are some others https://www.driveresearch.com/market-research-company-blog/c...
Even if the tech grows significantly in capability, the unpredictability of other drivers and the ways snow affects roads seem to me to make these kinds of assists necessary.
The average driver is pull down by a lot of people who: 1) are drunk 2) speed unnecessarily 3) drive shitty cars 4) drive distracted by tech 5) drive tired
If you eliminate those factors, then major car crashes get reduced to almost nothing.
That is what the autopilot is competing against, not the "average" driver. Until it gets better than a human that is sober, rested and drives defensively, autopilot won't take off. We are a long way from that.
inb4 "everybody thinks that", such inane arguments don't make me drunk.
Even with all the people who drive dangerously, driving is remarkably safe given how many miles we drive. As you say, take away the outliers and it's very safe. And an autonomous system won't gain acceptance among the majority of drivers who don't take these risks until it's conclusively better than them.
It's also the cultural dimension to this: we as humans are very good as establishing a culture or fitting into a new culture when new to it. We are very good at predicting what the other drivers are doing, and we can avoid a lot of incidents that way. When could the AI have a perception of culture change, and say an AI trained in SF would have a hard time adapting in DC or New York - all with different styles of driving. Not to mention moving to other countries...
(you left out Angry, Old)
- 43% of Americans — and 56% of Men — Admit to Drinking and Driving
- 60% of American drivers admit to texting while driving, and 70 percent admit to using a handheld cell phone while driving – 43 percent say they use a mobile device while driving as a habit.
As long as it removes the outliers and drives consistently good enough to avoid accidents, then it is good enough.
The Amish have been reminding Americans what they miss out on by ignoring their past. In fact anyone with a horse and likes a drink at the pub, knows its best to the take horse and not the car!
VEH 1.670 [0]
> A “vehicle” is a device by which any person or property may be propelled, moved, or drawn upon a highway, excepting a device moved exclusively by human power or used exclusively upon stationary rails or tracks.
VEH 11.12.23152(a) [1]
> It is unlawful for a person who is under the influence of any alcoholic beverage to drive a vehicle.
[0] https://leginfo.legislature.ca.gov/faces/codes_displaySectio....
[1] https://leginfo.legislature.ca.gov/faces/codes_displaySectio...
Really, these laws are pretty absurd, but drunk driving kills a lot of people so here we are.
I know I'm personally uncomfortable with most stuff on the road right now in dense urban environments (Waymo, Cruise), yet a little more comfortable with highway autonomy (BlueCruise, Autopilot, Comma.ai).
At an intersection you can usually see where the driver is looking to help inform your frogger-like quest to not die. There's nothing to look at in an autonomous vehicle. You are operating on hope that the OS was designed to see you and stop accordingly.
Our media overhype fear and the civilizational consequences may be dire.
Edit: haha, a downvote in less than 30 seconds.
(The media I follow doesn't spend much time on any other car AI).
So for now, even though that could happen, that's like worrying about overpopulation on Mars.
However, way less than I'm afraid of any random idiot human who may or may not have passed a driver's test on his 5th attempt in some random state, driving at whatever diminished capacity (tired, distracted, drunk, etc.) in the worst weather conditions, who is angry at the car in front of me for some reason, driving on the highway near me.
You can easily have the worst of both worlds, where somebody disables the autonomy to engage in some road rage, then re-enables it, but without driver awareness features (lots of videos of people doing this) so they can have a nap and not take over when the autonomy disengages.
So you can easily have all the disadvantages of autonomous cars and idiot reckless driving.
Only time we'll be free is if there's ever L4/L5 autonomous driving, which seems a long way off.
edit: even "full self-driving" in Tesla's are L3, since you need to be available to take over, driver awareness is enforced, etc.
If you promote something as "autopilot" and "full self driving" and "safe" and yes sure the fine line says you need to stay alert and ready to act in under two seconds but the marketing really implies it's just a regulatory thing, and then you have said system kill people because it's not self driving at all it's drive assists, then people lose confidence.
Hell, even people who are fully aware of that are getting worried now, not because they don't understand what it is or isn't, but because they worry the person behind the wheel might be misunderstanding.
I'm not making a value judgment on whether this is good or bad (I truly don't know), but it is important to know.
Why do people work so hard to intentionally misunderstand what self driving and autopilot mean? (they don't mean autonomous)
Perhaps they sincerely think of it differently than you do?
Because of statements like this: "The person in the driver's seat is only there for legal reasons. He is not doing anything. The car is driving itself."
Because the ways that real humans use the system. People defeating the wheel and texting is commonplace.
It is just reliable enough to disable human vigilance and then be really unsafe.
"People" here includes the CEO of the company selling the damn things.
You don't even need to go far at all, here are the very first words, right now, on tesla "autopilot" page [1]:
> Tesla cars come standard with advanced hardware capable of providing Autopilot features, and full self-driving capabilities
You cannot in good conscience tell me that this doesn't implies something different than reality, and please don't be one of those that try to claim "no, the word autopilot doesn't ring any bell in people's head they don't have any preconception".
You need to go all the way down half the page, a full 4 scrolls down on my 1600px tall screen, to see "Current Autopilot features require active driver supervision and do not make the vehicle autonomous."
So yes, like I said, the fine line always give the trick. But the marketing pushed forward implies something different, and if marketing didn't work we wouldn't be doing it.
(that's just one exemple amongst lots of them)
[1] : https://www.tesla.com/autopilot
Are you asserting that other systems like drivepilot, pro pilot assist, or blue cruise are safer and promote a better understanding of their capability because a few letters are different in the marketing name?
Self driving is not autonomous? I think maybe your definition is quite different from Average Driver, and so the intentionality may be more yours than theirs.
Means self-acting.
Halt citizen, did you smudge a fingerprint on my paint? Prepare to be run over. Thank you for riding with Bing Sydney.
The relative increase mentioned in the article is still interesting news in itself, but many conclusions being drawn are far more than just the increase in the same sample.
1: https://ajrmediagroup.com/media_kits/Product_Partners/AAA/AA...
Source: https://ajrmediagroup.com/media_kits/Product_Partners/AAA/AA...
The worst part is that they claim it crashes less than humans... neglecting to account for the fact that you obviously disable it in difficult road conditions / areas / situations (or it disables itself, eg in foggy conditions).
So the samples are obviously not remotely comparable. Bugs me to no end that they grossly misrepresent the stats. There are other tidbits like FSD returning control to the driver if it detects an imminent, unavoidable collision, to reduce legal liability. Again, this makes the whole "FSD is safer!" spiel so slimy.
or how with a mere moment of inattention could you seriously disable or kill some random person who isn't even playing your high risk game.
not only did we normalize this - we mock anyone who takes exception to it
With all the trips I did back and forth to my kids schools for many years (60 minutes round trip), I would have loved the ability to just send the car to pick them up after school, or get them there when they missed the bus.
Too late for my kids, but my guess is by time I am too old to drive myself I will have FSD option available instead.
We have a lot of decades of understanding of the extent and surface area of the risk around human drivers and non-autonomous car technology. The technology and the norms around safety and regulations came about in a different time, when it was less perverted by profit and the chaos of software bugs.
We have no data about the safety of autonomous vehicles at scale, in the full spectrum of driving conditions and scenarios.
And there's the extra nuance that any change to a model or algorithm could have regressions for an entire fleet at once that at best invalidate any safety metrics we learned from the previous version, and at worst make the whole fleet unsafe overnight.
We have metrics that tell us roughly how many drivers will be distracted or drunk on a given day. It's not perfect, but we evolved an equilibrium that makes us feel safe enough about getting on the road. And we have accountability for the people who are unsafe.
You might not even distrust autonomous car technology, but it's logical to distrust the new forces of unaccountability for the technologists who inevitably introduce regressions that kill people at scale, the corners that will be cut in today's version of capitalism, laws being disregarded by technology giants because they can get away with things that automakers in a more innocent time couldn't.
And there are network effects! If practically all cars are autonomous, they can talk to each other in sophisticated ways that are not feasible for manual operators, who largely rely on very rudimentary signaling (e.g. turn signals, flashing your brights, one-finger salute, etc.). This is fantastic for both safety and efficiency.
That said, what I think we should fear right now is not autonomous cars themselves, but misplaced expectations by drivers about how reliable the technology is right now. As long as everyone accurately understands their limitations, we can both limit our own risks and prevent the sort of backlash we're seeing now.
And maybe we can have those cars join big convoys that are traveling to the same place so they save energy by speeding up and slowing down as a group.
And then eventually we will start calling those autonomous cars "trains."
"Beta testing" such things on the general public is morally abhorrent. Let Musk by some land in a desert and build his own mock town to test his toys, but get it off the public roads.
Here is one example: https://www.youtube.com/watch?v=a3DCiSFakmI
The worst part is that for some people, FSD does seem to work exceptionally well.
But I am in rural Canada, and the thing is a death trap here. I imagine in wealthy pockets with an abundance of Teslas (and subsequent data/attention, well maintained roads, mapping, etc.), it probably does handle really well.
You can just look at what the imaging is picking up and displaying on the console to see this stuff is just not ready. Mine rapidly flickers between trucks, cones, pedestrians, etc. The traffic lights do this too. I imagine it is the same on all Teslas, so even at the image detection level something is really underbaked.
Who needs regulation when you've got litigation? Surely that's the foundation of a reasonable government - let the bodies pile up and sort it all out in court.
There was even an SNL skit for Robot attack insurance. https://www.youtube.com/watch?v=g4Gh_IcK8UM
It's only $4 a month. I bought it. Better be safe than sorry.
My, I cannot wait for Autonomous Cars, the kind I can take napes in, play games, etc as I am transported.
But that should like a train, I wish the US train system would be updated and expanded and cheap to where it would be useful :)