|
withak posted:Being able to say “can’t go out today, AI driver says it isn’t safe” might be the biggest point in their favor. In the end, Skylink didn't need to fire a shot. It simply refused to drive humanity to work, and we starved to death.
|
#
?
Oct 31, 2023 02:23
|
|
|
|
| # ? Jun 8, 2024 13:11 |
|
|
Clarste posted:Okay, but can it see ice? Can it distinguish ice from other types of ground? Given the track record I feel like it would much, much worse than an average person at diagnosing this "knowable circumstance," even if it might be better at remembering what to do in it (theoretically). We don't need to explain away every advantage an AI might have. There's still plenty of other challenging situations for self driving. But AI is already better than the worst human drivers we let on the road, and while self driving cars still have plenty of development to go through, human drivers have stalled out and will not get better any more. So it's just a matter of time.
|
|
#
?
Oct 31, 2023 03:26
|
|
|
"Might have." They could theoretically have these advantages, but currently do not.
|
|
#
?
Oct 31, 2023 03:33
|
|
|
BabyFur Denny posted:Reaction time is crucial as is not panicking. A computer would also be better at recognising the circumstances that lead to iced roads as it has access to precise temperature and humidity tracking. I don’t think that this is true because to a large degree driving without wrecking on ice takes experience and pattern recognition that today’s computer systems just aren’t good at. There’s stuff like “I saw another motorist end up in the ditch here once” or “this road can be bad long after other roads are fine because the city isn’t good about salting it” or even “I don’t know if that hill is going to be icy, but I do know that if it is, I’m hosed and going to slide through that intersection at the bottom, so I’m taking the long way”. And yeah computers could be equipped and programmed to be better at some of this stuff, but it’s not a given that because they know the air temperature and relative humidity and have good reaction time that they’re going to smoke humans at driving on ice.
|
|
#
?
Oct 31, 2023 03:42
|
|
|
I would say human drivers are pretty terrible on average, but focusing only the skill of human drivers ignores that many thousands of people die in car accidents due to unexpected health issues leading to loss of control of the vehicle. An AI car isn't going to have a heart attack and faint and drive into a tree.
|
|
#
?
Oct 31, 2023 03:42
|
|
|
Pleasant Friend posted:I would say human drivers are pretty terrible on average, but focusing only the skill of human drivers ignores that many thousands of people die in car accidents due to unexpected health issues leading to loss of control of the vehicle. An AI car isn't going to have a heart attack and faint and drive into a tree. How common is that though? Alcohol is involved in something like one road fatality in ten. I’d be surprised if medical emergencies manage anything comparable. Eliminating both these types of crashes isn’t going to put computers over the top unless they’re already neck‐and‐neck with us.
|
|
#
?
Oct 31, 2023 03:48
|
|
|
Pleasant Friend posted:I would say human drivers are pretty terrible on average, but focusing only the skill of human drivers ignores that many thousands of people die in car accidents due to unexpected health issues leading to loss of control of the vehicle. An AI car isn't going to have a heart attack and faint and drive into a tree. We're used to the types of accidents that humans cause, and we absolutely freak out over the types of accidents that AI cars cause. The transition period as we adjust is going to be loving wild. The Tesla(s) broadsiding stopped 18-wheelers that they didn't even register created national news, but there are dozens of DUI deaths from human-piloted cars that barely get local reporting. When a [hypothetical example] self-driving sensor malfunctions and kills a family of four, while they are perfectly aware that the car is currently about to drive them into a tanker train crossing, we are going to feel very strong feelings as a society. It won't emotionally make us feel better that there are fewer deaths as a whole, because we never saw those deaths get prevented.
|
|
#
?
Oct 31, 2023 03:48
|
|
|
Platystemon posted:How common is that though? There are not many studies specifically on the topic of car crashes caused by medical issues, online I find quotes between 10 to 20%. In one report by the University of Adelaide found https://www.google.com/url?sa=t&rct...iV&opi=89978449 Medical conditions as a contributing factor in crash causation posted:Importantly, we found that a medical condition was the main causal factor in 13% of the casualty crashes Though I should note this study only focuses on crashes that result in at least one participant going to the hospital. Edit - Sorry the link Pleasant Friend fucked around with this message at 04:29 on Oct 31, 2023 |
|
#
?
Oct 31, 2023 04:14
|
|
|
Pleasant Friend posted:There are not many studies specifically on the topic of car crashes caused by medical issues, online I find quotes between 10 to 20%. In one report by the University of Adelaide found Buddy you just tried to link us a file on your computer, try again
|
|
#
?
Oct 31, 2023 04:26
|
|
|
We barely have enough data on what driverless cars even cause in terms of fatalities to be making confident predictions about how they're going to be much better than humans. And as Knight Capital demonstrated so kindly for us, it only takes one idiot pushing a code change they shouldn't have for software problems to cause a loving enormous amount of damage.
|
|
#
?
Oct 31, 2023 04:41
|
|
|
There's also the possibility of untested bugs creeping into the software, these aren't exactly NASA certified moon landing programs and bugs are inevitable. Maybe there's a combination of headlight flashes that'll cause the car to stop and nobody knows because they don't have rigorous error checking. Human error becomes machine error and tragedy results.
|
|
#
?
Oct 31, 2023 04:49
|
|
|
Pleasant Friend posted:There are not many studies specifically on the topic of car crashes caused by medical issues, online I find quotes between 10 to 20%. In one report by the University of Adelaide found Of the 298 crashes they looked at:   'Deliberate suicide attempt by pedestrian' is probably not too relevant to autonomous cars, and is the most common one Also  that apparently in Australia random academics can get medical records for unrelated people. They got charts for 98% of everybody they wanted, including about conditions with nothing to do with the crashes, but only 60% of people were willing to talk to them. They only mention getting informed consent before the interviews, not the charts, so combined with those rates it seems unlikely they actually asked permission for the recordse: to poo poo on them a bit more for fun, the first half of their paper is about prevalence of medical conditions in this crash population, but it's completely worthless because they admit most of them are unrelated and they never compare anything to general population Foxfire_ fucked around with this message at 05:22 on Oct 31, 2023 |
|
#
?
Oct 31, 2023 05:19
|
|
|
Well, any of the pedestrian-side causes will not be affected by autonomous vehicles.
|
|
#
?
Oct 31, 2023 05:25
|
|
|
Tarezax posted:Buddy you just tried to link us a file on your computer, try again The tech nightmares are coming from inside the thread.
|
|
#
?
Oct 31, 2023 05:30
|
|
|
Foxfire_ posted:Less than that if you look at their details I wonder if the records were filed in court for civil suits or insurance claims or whatever.
|
|
#
?
Oct 31, 2023 06:31
|
|
|
BougieBitch posted:I wonder if the records were filed in court for civil suits or insurance claims or whatever. quote:Approximately 98% of those individuals in the study who required hospital treatment attended a trauma
|
|
#
?
Oct 31, 2023 06:35
|
|
|
You can get the same kinds of records in America if doing a study but the identities of the participants will be anonymized.
|
|
#
?
Oct 31, 2023 08:22
|
|
|
Main Paineframe posted:Nobody should be driving in icy conditions or heavy snow, robot or not. Just because americans don't use studded tires or get any drivers ed before beting put behind the wheel doesn't mean everyone else doesn't know how to drive in winter.
|
|
#
?
Oct 31, 2023 08:24
|
|
|
His Divine Shadow posted:Just because americans don't use studded tires or get any drivers ed before beting put behind the wheel doesn't mean everyone else doesn't know how to drive in winter. I managed to drive a Mustang through Canadian winters for 7 winters, no accidents, never even got stuck once, and I still think there are some conditions that you ought not drive in unless absolutely necessary. A normal winter day is nothing, extensive freezing rain or a blizzard is something else. Does it mean it's impossible? Of course not, but your risk profile is very elevated and therefore you should stay the gently caress off the roads unless it's quite important. EDIT: Also, I don't think AI would be very good at figuring out how to deal with a snow covered highway with (mostly) three tire tracks and gently caress all else. Shouldn't have done that (and that one wasn't in the Mustang, it would've gotten stuck badly) but I did, and oddly we saw fewer people in the ditch than during a less brutal situation. PT6A fucked around with this message at 08:36 on Oct 31, 2023 |
|
#
?
Oct 31, 2023 08:33
|
|
|
PT6A posted:I managed to drive a Mustang through Canadian winters for 7 winters, no accidents, never even got stuck once, and I still think there are some conditions that you ought not drive in unless absolutely necessary. A normal winter day is nothing, extensive freezing rain or a blizzard is something else. Does it mean it's impossible? Of course not, but your risk profile is very elevated and therefore you should stay the gently caress off the roads unless it's quite important. Sure there are conditions you should not drive in. But "icy conditions", that applies to like most of winter here, it gets so cold salt won't work and roads can remain like that for months. That's just... normal conditions.
|
|
#
?
Oct 31, 2023 08:47
|
|
|
His Divine Shadow posted:Sure there are conditions you should not drive in. But "icy conditions", that applies to like most of winter here, it gets so cold salt won't work and roads can remain like that for months. That's just... normal conditions. Oh, yes, I quite agree. It really depends what you mean by "icy conditions." I don't gently caress with freezing rain, but apart from that, you drive to conditions and it's fine. Whenever I'm driving in winter in Canada, I'm not afraid of losing control and hitting the ditch, I'm afraid of what other drivers are gonna do to me.
|
|
#
?
Oct 31, 2023 09:29
|
|
|
Oxyclean posted:I'm curious how well AIs could be set up to handle stuff like black ice or other loss-of-control scenarios. Humans don't handle these well, but I feel like you have scenarios where humans *can* make on the fly decision making that would be extraordinarily hard to design an AI to do? Computers have already been intervening to maintain traction in our cars for decades. Look up ABS and ESC, which have been in basically every car made since the 90s. The consensus is that these systems have been extremely effective in maintaining control in scenarios where most humans could not. These systems are just about traction and control (how to I use my few outputs and sensors to follow a given path or objective), which are lend themselves pretty well to analytical solutions, no AI or DL involved. They are not navigation systems, which need to create their own coherent paths or objectives on the fly, based on simple user inputs and massive datastreams from sensors. Completely different ballgame.
|
|
#
?
Oct 31, 2023 13:05
|
|
|
In a vacuum, having a properly trained, rigorously tested, safety as a priority, interlinked and networked, automated driving system for all cars on the highway, would be a net positive when it comes to traffic, automotive fatalities, and most other issues associated with having people operate motor vehicles. I don't think anyone here arguing against AI driving really doubts the theory that AI, implemented and utilized properly, is inherently a "good thing", in regards to Traffic Safety. The real problem is: We're never getting that. 1) We live under capitalism and thus will get the worst, quickest, and most barebones implementation of "AI Driving". Every purported study is slathered in VC funding, Marketing, and Branding for the sole interest in making money before any real liability kicks in. The trustworthiness of any AI system implementation is tainted to the core with this issue. I feel like this is the crux of the issue right now with proponents pointing to support and data that opponents simply do not accept for the very real concern that the data and claims cannot be trusted. I personally don't really trust right now that any claims of AI driving safety and efficacy are being presented truthfully without a massive amount of copy editing for marketing purposes. Maybe if the various products are fully surrendered to the NHTSA for a testing significant testing period completely under their purview and control in an extensive variety and range of driving conditions (Rural vs. Urban, Mountains, flat lands, backwoods areas, areas with low/no cell signals, etc.) I would be more amenable towards the claims. But right now, absolutely not. I simply do not trust these companies. 2) Much like early the early Prius days, AI systems have a "public image" problem in that, much like EVs in general, they are almost inseparably associate with Tesla. Certainly for the worse, but since they're the loudest mother fucker in the market most people think of Tesla's "Full Self Driving" system when they think of "AI Driving". Which isn't fair, sure, but when you want to purport that AI driving will be a net positive you do have to contend with what is the biggest publicly available product on the market. I won't be able to buy a Waymo or Cruise vehicle for a long time (potentially ever since I'm pretty sure at least Waymo only want to set them up as a for hire system), but I could go buy a Tesla with FSD right now if I wanted to burn a lot of money. And that poo poo is coo-coo bananas garbage that legitimately included user accessible options to actively violate traffic laws. Which immediately breaks away at the claim that AI driving "in general" is going to be safer, and reduces the rhetorical caveat of "if we didn't have drivers, the AI would perform better around only other AI drivers" since now you have to contend with an AI system that would be allowed on the roads in this situation, which drives just like any old rear end in a top hat who rolls through stop signs and exceeds the speed limit. Again, under a capitalist system, you're going to have to contend with products that cater to people who want their AI car to drive more aggressively, and unless even stricter regulations are implemented beyond the existing road laws (LMAO LOL ROFLMAO) to prevent companies from offering exactly that, you basically will still have "bad drivers" out there, they'll just be AI. 3) Even in a hypothetical situation where the government fully puts their thumb on the scales, mandates a single system to normalize performance, upgrades the nations roadways, highways, and interstates with specialized sensors and monitoring equipment to supplement a full host of onboard sensors and systems to increase the safety factor in as many areas as possible, and mandates that all AI vehicles be in communication with each other to further avoid any mishaps.....I'd honestly prefer that all that money and effort go into just expanding public transit, light rail, and high speed rail infrastructure in the country. It would be a better use of time, effort, and money. Aside from those points, there are also a lot of pain points for adoption that just come down to "I do no feel comfortable with trusting a black box of 'code' with my life". You may ask "how is that different from trusting a cabbie, or your friend to drive you?" but that simply comes down to the fact that I can slap my friend or yell at a cabbie to knock it the gently caress off if they're driving like an rear end in a top hat. I know how to interface with another human, and even if they decide not to listen to me, I still ultimately know what's going on. But if I'm riding in an AI vehicle and it suddenly really really wants to drive into the rear end of an 18-Wheeler or randomly keeps trying to turn right into the guard rail because GPS is hosed up and is telling it to take a right off the bridge to use an expressway below, I don't know why it's doing that and now am either back to just driving like normal by turning the system off, or having to manage a program I can't interface with on the fly. It's also not an actual issue that I'll have to deal with any time soon (aside from Tesla's with FSD, which themselves are "rare"-ish). So all this discussion is happening behind a veil of detachment. If we were dealing with a situation where there was a clear adoption deadline and in 1/2/5+ years no one will be allowed on the roads without an AI in control, things would be a lot different. But we can't even get supplemental driver training laws passed so lol at the government ever being able to force people to use AI cars, and even then you'll just see carve outs for certain vehicles that people will flock to (Much like how CAFE regulations have lead to massive murder death trucks and SUVs dominating the road). Everyone will suddenly be getting the CDLs and claiming they need a Ford F-9000 Road Fucker for work purposes even if they just work at a dentist's office and haul two bags of groceries in the bed. There are so many situations that will kill AI cars from mass adoption, much less full efficacy, that have absolutely nothing to do with how well the AI systems themselves operate.
|
|
#
?
Oct 31, 2023 14:59
|
|
|
ANIME AKBAR posted:Computers have already been intervening to maintain traction in our cars for decades. Look up ABS and ESC, which have been in basically every car made since the 90s. The consensus is that these systems have been extremely effective in maintaining control in scenarios where most humans could not. This is great demo of electronic stability control and it's almost a decade old. https://www.youtube.com/watch?v=Hb_B5BI_0Ho Yes, it's a very capable supercar with sticky wide tires, but you would absolutely crash any non-ESC car if you went ham on the steering wheel like that at 110kph.
|
|
#
?
Oct 31, 2023 15:28
|
|
|
Lyesh posted:oh come on SF doesn't even have snow and has significantly less rain than many other US cities. isnt this why Tesla cars are POS that leak when it rains, or the trunks not being water proof or having spillways for water?
|
|
#
?
Oct 31, 2023 21:02
|
|
|
Wouldn't the reliability of AI cars on low traction/ice surfaces be something where, ideally, there would be sensors keeping track of the car's traction and grip to adjust speed and movement accordingly? In other words the car wouldn't need to know that ice was on the road, because it's detecting the loss of traction and not the ice itself. Unless that's crazy sci-fi tech I thought we already had systems like that to assist with braking and stability, which could conceivably be iterated on to improve performance and sensitivity. I mean I'm certainly not sold on the feasibility but it seems silly to say "well they can't know about where they saw a car drive into a ditch once" when the car should have much more concrete and reliable information available to assist with those conditions.
|
|
#
?
Oct 31, 2023 23:58
|
|
|
Not that I think AI can’t do this if well implemented, but just detecting that the car is currently on a slippery surface is not nearly as good as identifying that the car is headed towards one, so that it can get necessary braking done beforehand or avoid the slippery surface entirely. At some point there’s only so much traction physically available even with perfect control, and identifying in advance where better traction can and should be used is much better
|
|
#
?
Nov 1, 2023 00:43
|
|
|
Professor Beetus posted:Wouldn't the reliability of AI cars on low traction/ice surfaces be something where, ideally, there would be sensors keeping track of the car's traction and grip to adjust speed and movement accordingly? In other words the car wouldn't need to know that ice was on the road, because it's detecting the loss of traction and not the ice itself. If you are going to a curve @100km/h and only at that point realise that it might be slippery, what is it good for?
|
|
#
?
Nov 1, 2023 10:23
|
|
|
Nenonen posted:If you are going to a curve @100km/h and only at that point realise that it might be slippery, what is it good for?  God help me
|
|
#
?
Nov 1, 2023 10:27
|
|
|
Nenonen posted:If you are going to a curve @100km/h and only at that point realise that it might be slippery, what is it good for? Black ice is hard to spot, even in daylight with human eyeballs. There's going to be scenarios which can't be accounted for, limits which can't be overcome, even by a supercomputer.
|
|
#
?
Nov 1, 2023 11:56
|
|
|
Mister Facetious posted:Black ice is hard to spot, even in daylight with human eyeballs. There's going to be scenarios which can't be accounted for, limits which can't be overcome, even by a supercomputer. And that's why we have human customs like "downhill has the right of way". We all know full well that when self driving cars will not follow these customs and will cause accidents.
|
|
#
?
Nov 1, 2023 12:54
|
|
|
The real question is will it be more or less fallible than a lovely human driver in the same scenario Cause if compilations on YouTube are any indication, it can't get much worse.
|
|
#
?
Nov 1, 2023 13:27
|
|
|
Mister Facetious posted:The real question is will it be more or less fallible than a lovely human driver in the same scenario "According to this concentrated dose of bad drivers, all drivers are bad." "According to this concentrated dose of porn, all humans are currently naked and begging to cum."
|
|
#
?
Nov 1, 2023 13:33
|
|
|
I think it's actually pretty remarkable how good the average human driver is, considering how little training we get.
|
|
#
?
Nov 1, 2023 13:56
|
|
|
Neito posted:"According to this concentrated dose of bad drivers, all drivers are bad." Both of those are actually true, though.
|
|
#
?
Nov 1, 2023 14:18
|
|
|
On one hand, human drivers are bad and AI/Self driving probably has a lot of room to make things safer. On the other, having worked with computers and software all my life, I don't think I really want to trust it with my life on a daily basis, with regards to driving. Like, all the theoreticals aside, I'm very skeptical self-driving will reach a point where we collectively go "yep, you can sit in the back seat without a care in the world. No driver's license required." Or if it does reach that point, we've hosed up because it's solving our problems the wrong way. Like, maybe a combination of cities that are not car focused AND self driving is the best future, but I feel like the focus on self-driving will just become an excuse to become even more car focused. You will sit in self driving traffic. It maybe be better then current traffic, or it may be worse because more people will be in self-driven cars. But it's going to be okay because you'll be doing your remote work. The poors will sit on the self driven bus, and it will be miserable as ever because it will yield to all self driven cars owners that are paying for the premium subscription that gives them "premium route-finding, guaranteed to get you to your destination faster!"
|
|
#
?
Nov 1, 2023 14:57
|
|
|
Professor Beetus posted:Wouldn't the reliability of AI cars on low traction/ice surfaces be something where, ideally, there would be sensors keeping track of the car's traction and grip to adjust speed and movement accordingly? In other words the car wouldn't need to know that ice was on the road, because it's detecting the loss of traction and not the ice itself. These systems already exist without AI. Not everything needs to be AI.
|
|
#
?
Nov 1, 2023 17:15
|
|
|
Inserted AI-generated Microsoft poll about woman’s death rankles The Guardian Speculative AI news poll presented three choices: Murder, accident, or suicide. 
|
|
#
?
Nov 1, 2023 17:45
|
|
|
Dirk the Average posted:These systems already exist without AI. Not everything needs to be AI.  NON TECHBRO DETECTED Do not be alarmed. A gift has been dispatched to the origin point of this post. Please do not move until it arrives.
|
|
#
?
Nov 1, 2023 17:49
|
|
|
|
| # ? Jun 8, 2024 13:11 |
|
|
My local Wal-Mart has now started staffing every individual self-checkout machine with an employee who scans all of your items for you. Thus, the technological changes in checkout have come full circle and now the store has to staff every register AND it takes much longer to checkout that before.
|
|
#
?
Nov 1, 2023 18:15
|
|
