- cross-posted to:
- technews@radiation.party
- hackernews@derp.foo
- cross-posted to:
- technews@radiation.party
- hackernews@derp.foo
Tesla braces for its first trial involving Autopilot fatality::Tesla Inc is set to defend itself for the first time at trial against allegations that failure of its Autopilot driver assistant feature led to death, in what will likely be a major test of Chief Executive Elon Musk’s assertions about the technology.
Although it’s far from perfect, autopilot gets into a lot less accidents per mile than drivers without autopilot.They have some statistics here:https://www.tesla.com/VehicleSafetyReportEDIT: As pointed out by commenters in this thread, autopilot is mainly used on high ways, whereas the crash average is on all roads. Also Tesla only counts a crash if the airbag was deployed, but the numbers they compared against count every crash, including the ones without deployed airbags.
Oh yeah, potentially cherrypicked statistics straight from Tesla. I’ll believe those statistics when they come from someone not with a horse in the race to adopt autonomous vehicles.
I think it’s been reported that the FSD statistics they put out are worthless because it tends to disable itself right before collisions.
What’s the motivation to cherrypick though?
Human drivers are bad enough that I don’t think there’s any doubt that autopilot puts them to shame with regards to safety, so they can either look way better and not be suspicious, or look way better and be suspicious… Sounds like an obvious choice to me
They have a financial motivation. You ucould also just Google self driving car safety, and one of the first Google hits is an article that calculated the safety of human drivers from data collected in 2021. Turns out humans are already pretty damn safe, there’s roughly 99.9998 of driving with zero accidents.
I assume you mean accidents with a fatal injury, given there is a ~1% chance that any given death will be from a car accident (17.4 deaths per 100k per year * 70 years = 1.2%) - using your statistic yields closer to 2.5% however this works with only one driver dying.
Turns out you’ve been tricked by statistics, driving is fucking lethal and chances are most people know or are friends with someone who has died or will die in a car accident (assuming ~80 friends/acquaintances per person)
Why the arbitrary number of 70 years?
If you calculate the chance of having an accident per route traveled (about 2.57 per person per day), you get a number much closer to NocturnalMorning’s statistics.
A rough estimate for global life expectancy. It’s actually slightly over 73, so the chances of dying in a car accident are marginally higher than I said.
The data I used wasn’t related to driving frequency or age, it was purely the number of people in a random global sample of 100,000 people you would expect to die in a car accident in a given year. That of course includes people of all ages and people who never drive at all, but also taxi & HGV drivers. Even if we say people aren’t in cars so much under the age of 5 or over the age of 60, that would push up the deaths per 100,000 people per year between 5 and 60 by the exact amount to keep the chance per year over a human lifetime at 17.4/100000.
Where are the numbers to back this up?
Start with the numbers on humans driving drunk, tired, on their phone, while having a conversation, bored or in practically other state and work backwards. Driving is dangerous as fuck and it’s pretty much universally accepted that the biggest challenge for autonomous vehicles is humans doing unpredictable and stupid shit
The biggest challenge when I’m driving is humans or AI doing unpredictable and stupid shit.
You still have not given any numbers to back up your claim. While we all expect that AI will one day be much better than humans in driving, there is no data to say that it currently is.
Ok so sure there’s nothing on Tesla’s autopilot, however that’s not to say there’s nothing on autonomous systems…
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8431415/
https://www.orsa.org.uk/reducing-occupational-road-risk/reducing-driver-error-accidents/
There’s your stats on humans being reckless and dangerous when driving cars, and of course there’s nothing concrete for fully autonomous cars because they aren’t legal anywhere, but here’s some stats on pretty much every existing driver assist - notably they all prevent accidents compared to just a human driving: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8431415/
It really isn’t a stretch from the 3 most frequent crash causes being human error and human assistance tools reducing accident frequency a bunch to say that all these systems coming together (as they cover near enough everything to do with driving a car) would be safer than a human driver, but I don’t doubt you’ll deny it as you’re asking for something impossible to give (as governments haven’t allowed full autonomous driving cars yet, so there’s no statistics on their use) and so aren’t actually looking for information but to confirm your biases and feel like you’ve “won”, despite the fact there’s no objectively unsuspicious data on the exact situation you’re asking for meaning that you can’t prove yourself right either beyond “I’m a little suspicious of this company so I must be right”
Those stats are misleading though. Autopilot only runs on highways, which are much safer per mile even for human drivers.
Tesla are basically comparing their system, which only runs in pristine, ideal conditions, against an average human that has to deal with the real world.
As far as I’m aware they haven’t released safety per mile data from the FSD cars yet, and until they do I will remain skeptical about how much safer it currently is.
It actually would be really hard to get an unbiased estimate of safety given the current systems, because the data is inherently cherry picked by drivers who can switch the feature on/off depending on how complex the driving task is. What a simple number like crashes per mile really measures is really how likely FSD drivers are to overestimate the system’s ability plus some unknown base rate of unavoidable accidents.
Probably the only way to control for this is looking at cars that are fully autonomous door to door and aren’t limited to pre-selected roads/areas. I don’t know that anyone is even doing that sort of testing.
Hmm you’re right about autopilot mainly being used on highways and those roads are a lot safer. I’ll edit my main comment
Doesn’t auto pilot kinda work on normal roads now? Not saying I trust the stat either.
That’d be the fsd stats, not autopilot
Source
Does that report from Tesla include when autopilot turns off shortly before crashing into something?
That’s literally the only data we have so that’s what I’m basing my opinion to while being fully aware that while I doubt that these stats lie they may however be misleading as statistics often are.
My key argument still stands; autopilot/FSD is not as bad/dangerous as people here make them to be and they’re getting better all the time.
If one is going to make the claim that these systems are more dangerous than human driver then show me the data you’re basing it on. People surely don’t think that just because they don’t like the company/CEO, right?
Do you have statistics not by Tesla?
They’re probably the only ones who even has access to such statistics. If you’re simply just going to refute the stats because of the source then atleast provide some credible counter evidence.
https://www.forbes.com/sites/bradtempleton/2023/04/26/tesla-again-paints-a-very-misleading-story-with-their-crash-data/
Even according to that article autopilot and FSD seems to be about at the level as human driver. I’m willing to accept that - many others arent.
The narrative here is that these systems are dangerous and shouldn’t be allowed to be used on public roads. My argument is that they’re not as dangerous as reading stories about these individual incidents might make them seem like and they’re getting better all the time. If they’re not significantly better than human drivers now they will be soon and Tesla most likely is going to lead the way.
Tesla’s numbers are trash. Tesla have been caught again and again lying.
…then provide some more trustworthy stats because you just saying that is not it. This is literally like debating a climate change denier or flat earther.
“Here’s a picture of the earth from space”
Why would Tesla release any numbers that would make it look bad?
Why should we trust any numbers that comes from Tesla?
And when autopilot is at fault for an accident or fatality, who should be held responsible?
Just because it’s better, shouldn’t absolutely them of responsibility when it fails.
It’s an interesting question. But I would be disappointed if the self-driving was basically killed by the legal questions, since it has a huge potential to save lives.
The driver is always responsible for using the tools within the car correctly and maintaining control of the vehicle at all times.
Either way the driver would be at fault. However, the driver might be able to make a (completely separate) case that the car’s defects made control impossible, but since the driver always had the option to disable self-driving, I doubt that would go anywhere.
Just like you don’t get off the hook if your cruise control causes an accident… and it doesn’t matter how much Tesla lied about what it may or may not be capable of, because at the end of the day it’s always the driver’s responsibility to know the limitations of the vehicle and disable the feature and take control when necessary.
Which is exactly what this case is claiming, that the software is defective.
And what happens when we progress beyond Level 2 or 3 automation? Then the car is making choices for the driver, choices the driver may not have any say in or realistically be capable of reacting to in an emergency?
Deferring responsibility to the driver under any scenario is a cop-out. We have a long history of engineering qualifications and regulations to ensure safety of the populace, engineers and architects design structures to be safe, plumbers have to plumb to code, heck even cars themselves have a mile long list of compliance requirements. All to ensure the thing that companies build aren’t killing the population, and when they do someone is responsible.
Yet as soon as we start talking about software, “not my problem dawg.”.
This is a guy who was using a glorified cruise control (which is all AP is) at high speed whilst watching a DVD instead of looking at the road.
The software can only help so much. There’s a reason why there are laws requiring attentiveness checks now… people are reckless
People are only reckless because they believe Teslas false marketing claims.
The car doesn’t “just drive itself”, it isn’t even close to “just driving itself”. The advertising claiming so is much more at fault than the driving watching a movie.
So you’re correct to call it a tool, with this level of automation the driver is ultimately the operator. But you’re missing something
Did you misuse the tool, did they sell you a bad tool, or did their instructions cause the tool to be misused?
The first is as you said - if I make and sell you a circular saw and you cut your finger off being an idiot, that’s on you.
If the thing flew apart under normal use, that’s on me - it’s likely my responsibility, and possibly negligence.
If the box or user manual said it is for wood and metal use, and it’s actually entirely unsafe for metal use, that’s probably negligence on my part
Cruise control doesn’t unexpectedly jerk your wheel to the side, if it did and you could prove you were using it reasonably and in the recommended way, you’d almost definitely get off the hook
This reminds me when you google if a certain company or product is good or legit and the top one is posted from the companies website.