Tesla and Uber Unlikely to Survive (Vol. 2)
Discussion
Interaction with pedestrians will be interesting too. The way the car picks up pedestrians, bicycles and motorbikes looks quite erratic in those videos. I would be nervous crossing the road in front of an 'FSD' Tesla. At least with a human you can make eye-contact and establish if they're engaged and have seen you. With a Tesla in FSD you have no clue if the car is paying attention or not, you can only look at the driver and hope he's covering the brake pedal. And if it does go, what do you do? With a human driver you or someone else can shout or bang the car or do something, but what if a Tesla decides to steamroller you while the 'driver' is looking at his phone?
NDNDNDND said:
Interaction with pedestrians will be interesting too. The way the car picks up pedestrians, bicycles and motorbikes looks quite erratic in those videos. I would be nervous crossing the road in front of an 'FSD' Tesla. At least with a human you can make eye-contact and establish if they're engaged and have seen you. With a Tesla in FSD you have no clue if the car is paying attention or not, you can only look at the driver and hope he's covering the brake pedal. And if it does go, what do you do? With a human driver you or someone else can shout or bang the car or do something, but what if a Tesla decides to steamroller you while the 'driver' is looking at his phone?
If you are crossing at a designated point (zebra crossing etc.) one assumes this will be factored in. If you aren't then the onus is on you a a pedestrian to take all precautions as you are the vulnerable party?Already a rising number of accidents caused by drivers not paying attention to the road hence all of the legislation around mobile phone usage, so you could argue having the car monitoring what is going on and automating actions would be an improvement?
Completely agree there will be certain circumstances where the AI is not as quick to spot an impending issue as quickly as a human driver but I do wonder how that balances out with the above mentioned general lack of attention shown by large numbers of drivers I'm sure we all come across on a daily basis?
SWoll said:
Completely agree there will be certain circumstances where the AI is not as quick to spot an impending issue as quickly as a human driver
Why do we assume an AI neural network monitoring x8 cameras, radar, and ultrasound sensor which never gets tired or distracted would be WORSE than a human with one pair of eyes and ability to be distracted by everything.The Uber pedestrian fatal crash showed the car had 'seen' the pedestrian crossing the road with multiple different sensors up to 6 seconds before the collision. If the car was allowed to react by it self it would have stopped/avoided the crash. The weak link was the human operator who was given the responsibility of authorising an stop procedure but instead was too busy watching TV.
If the AI system can demonstrate reliable safety measures such as pedestrian detection, human operators will quickly be seen as the danger.
AI algorithms out perform humans at pretty much every procedure/process based task, without ever been tired or needing a break. If these algorithms can now tackle quasi complex tasks like driving than we are heading very rapidly towards a different world where human input is seen as the cause of unwanted variation that needs removing from the system for better safety/efficiency/productivity.
Edited by gangzoom on Saturday 24th October 12:12
gangzoom said:
SWoll said:
Completely agree there will be certain circumstances where the AI is not as quick to spot an impending issue as quickly as a human driver
Why do we assume an AI neural network monitoring x8 cameras, radar, and ultrasound sensor which never gets tired or distracted would be WORSE than a human with one pair of eyes and ability to be distracted by everything.The Uber pedestrian fatal crash showed the car had 'seen' the pedestrian crossing the road with multiple different sensors up to 6 seconds before the collision. If the car was allowed to react by it self it would have stopped/avoided the crash. The weak link was the human operator who was given the responsibility of authorising an stop procedure but instead was too busy watching TV.
If the AI system can demonstrate reliable safety measures such as pedestrian detection, human operators will quickly be seen as the danger.
AI algorithms out perform humans at pretty much every procedure/process based task, without ever been tired or needing a break. If these algorithms can now tackle quasi complex tasks like driving than we are heading very rapidly towards a different world where human input is seen as the cause of unwanted variation that needs removing from the system for better safety/efficiency/productivity.
Edited by gangzoom on Saturday 24th October 12:12
You can spot someone not paying attention either whilst driving or walking down the pavement via body language and certain tells where I don't think AI will be able to for a long time yet .
As I say it's a balancing act between the importance of this ability versus general attentiveness and driving skills of the average person on the road.
gangzoom said:
Why do we assume an AI neural network monitoring x8 cameras, radar, and ultrasound sensor which never gets tired or distracted would be WORSE than a human with one pair of eyes and ability to be distracted by everything.
The actual intelligence of the human brain is a million years more advanced than any current processor could hope to be.Order66 said:
The actual intelligence of the human brain is a million years more advanced than any current processor could hope to be.
The social cues aspect is a relatively small part of driving. Reaction times are arguably more important, if pedestrian safety is your number one concern?lothianJim said:
Order66 said:
The actual intelligence of the human brain is a million years more advanced than any current processor could hope to be.
The social cues aspect is a relatively small part of driving. Reaction times are arguably more important, if pedestrian safety is your number one concern?Burwood said:
i'd argue the exact opposite in certain situations. Granted a super computer could see things I can't but conversely I've seen some things coming no computer could (today). Like the kid you know is going to probably wander into the road, not looking.
With supervised FSD beta you potentially have more attention spare to look out for such kids as your focus is not preoccupied with lane keeping etc.FSD beta does seem to slow down preemptively when pedestrians are about.
It's still unclear in your scenario that faster reaction times won't reduce accident rates on average.
lothianJim said:
Order66 said:
The actual intelligence of the human brain is a million years more advanced than any current processor could hope to be.
The social cues aspect is a relatively small part of driving. Reaction times are arguably more important, if pedestrian safety is your number one concern?The neural nets have been trained that "anything with this sorta shape is a road sign" (as a very crude explanation), and have no intelligence to interpret something that subtly fails to match their expectations. When a road sign is partially covered by trees, or discoloured by lichen we thing "Ah, there's a road sign there that's hard to read, best figure it out" - the computer just thinks "nothing there that fits my idea of a road sign".. This sort of flat out failure, or wrong choice is not protected by a brain thinking about the consequences of it's actions - it's just a set of weights saying "if I don't see this, I can keep doing this".
When we approach some unexpected temporary roadworks, our brain is figuring out the intent of other human beings when they put "that sign there, those cones there". A computer has no concept of intent. And when the diversion sign has annoyingly blown over in the wind, we see it on the ground and figure out "ah, might have to take a turn here".
lothianJim said:
Burwood said:
i'd argue the exact opposite in certain situations. Granted a super computer could see things I can't but conversely I've seen some things coming no computer could (today). Like the kid you know is going to probably wander into the road, not looking.
With supervised FSD beta you potentially have more attention spare to look out for such kids as your focus is not preoccupied with lane keeping etc. If the car is "doing all the driving", we stop concentrating. Nasa has studied this and says that the driver does not sit there acting like a wingman to the computer - they relax and stare out the window. That means we don't see things, and when we do it takes a measurable time to do anything about it.
As we get closer and closer to cars being able to self drive, this becomes more of an issue, not less.
FSD beta has quite specific capability and isn’t claiming to be able to handle these edge cases. The car isn’t doing ‘all’ the driving with fsd beta. And likely won’t for years. It makes turns. Navigates highways. Avoids parked cars. Etc.
Time will tell re accident rates and the narrator likely be insurance premiums. I believe there are more recent studies that counter the nasa findings? Even with limited functionality it's still compelling product, that will earn Tesla billions every year from here on? It doesn't seem likely regulation will stop FSD beta rollout in USA but can see germany etc pooping the party?
Robotaxi is another thing entirely.
Time will tell re accident rates and the narrator likely be insurance premiums. I believe there are more recent studies that counter the nasa findings? Even with limited functionality it's still compelling product, that will earn Tesla billions every year from here on? It doesn't seem likely regulation will stop FSD beta rollout in USA but can see germany etc pooping the party?
Robotaxi is another thing entirely.
Edited by lothianJim on Saturday 24th October 16:13
lothianJim said:
FSD beta has quite specific capability and isn’t claiming to be able to handle these edge cases. The car isn’t doing ‘all’ the driving with fsd beta. And likely won’t for years. It makes turns. Navigates highways. Avoids parked cars. Etc.
Time will tell, if you belive these slides (24 minute mark), its taken Tesla's current neural net 6 months to go from having less than 1/4 of the codes rewritten in new 2.0 format to now clearly superseding the previous code, which has been in development since end of 2016 so nearly 4 years.Tesla was also clearly head hunting the best engineers they could in Feb to help develop/improve distance judgement using cameras with LIDAR been the benchmark.
The progress in Tesla has achieved with FSD this year appears to have taken a big step forwards. So who knows whats coming?
HW4.0 been already readied for mass production maynot be a joke after all, if there is now a clearer understanding of CPU demands needed?
I guess we'll just have to wait and see, after all its been 4 years since the original FSD announcement so whats more time
.https://youtu.be/hx7BXih7zx8
gangzoom said:
That’s fascinating. One advantage of having a Comp Sci degree is that it firmly places me into ‘conscious incompetence’ watching that. I’d love to work on something like that but I’m nowhere near clever enough. Tuna said:
lothianJim said:
Burwood said:
i'd argue the exact opposite in certain situations. Granted a super computer could see things I can't but conversely I've seen some things coming no computer could (today). Like the kid you know is going to probably wander into the road, not looking.
With supervised FSD beta you potentially have more attention spare to look out for such kids as your focus is not preoccupied with lane keeping etc. If the car is "doing all the driving", we stop concentrating. Nasa has studied this and says that the driver does not sit there acting like a wingman to the computer - they relax and stare out the window. That means we don't see things, and when we do it takes a measurable time to do anything about it.
As we get closer and closer to cars being able to self drive, this becomes more of an issue, not less.
Didn't old Elon state, when being pressed about the lack of HUD in the M3, that when you're in a taxi you don't look at the speedo?
lothianJim said:
FSD beta has quite specific capability and isn’t claiming to be able to handle these edge cases. The car isn’t doing ‘all’ the driving with fsd beta. And likely won’t for years. It makes turns. Navigates highways. Avoids parked cars. Etc.
Time will tell re accident rates and the narrator likely be insurance premiums. I believe there are more recent studies that counter the nasa findings? Even with limited functionality it's still compelling product, that will earn Tesla billions every year from here on? It doesn't seem likely regulation will stop FSD beta rollout in USA but can see germany etc pooping the party?
Robotaxi is another thing entirely.
The narrator will be the bereaved and the lawyers with hefty lawsuits.Time will tell re accident rates and the narrator likely be insurance premiums. I believe there are more recent studies that counter the nasa findings? Even with limited functionality it's still compelling product, that will earn Tesla billions every year from here on? It doesn't seem likely regulation will stop FSD beta rollout in USA but can see germany etc pooping the party?
Robotaxi is another thing entirely.
Edited by lothianJim on Saturday 24th October 16:13
If the s
t hits the fan when FSD is operating then who is liable, Tesla or the driver?
There have been studies that examine how soon you should permit self driving cars on city and urban streets, which is where accidents occur most often. It's an active debate but there is one school of thought (not limited to Tesla) that even imperfect driverless cars with a supervising driver, will still save more lives, net. Therefore FSD systems should be permitted freely, with driver supervision bridging whatever parity gap exists between the system and a human driver.
I'm not saying this a slam dunk, just that there is an active debate about this stuff across the industry, it's not simply Musk being a dick. Afterall, if it were really that dangerous, FSD beta would not be allowed in any US state.
Re liability for accidents in a private vehicle, it's (very) obviously the driver.
I'm not saying this a slam dunk, just that there is an active debate about this stuff across the industry, it's not simply Musk being a dick. Afterall, if it were really that dangerous, FSD beta would not be allowed in any US state.
Re liability for accidents in a private vehicle, it's (very) obviously the driver.
Edited by lothianJim on Saturday 24th October 20:04
lothianJim said:
Re liability for accidents in a private vehicle, it's (very) obviously the driver.
Is it very obvious though. This is where my small issue with the programme is. They use the term "Full Self Driving". It isn't. It won't be for a very very long time. However they still sell a very expensive product called Full Self Driving. It should be called "AI driver assist" or "Assisted Driving" or something.In the US I can very easily see a court case of "I paid tesla for Full Self Driving" and the car killed someone while I was in the car.
US courts are can be harsh on supposed defects in products which then cause harm - attributing blame to the manufacturer. I think the term "Full Self Driving" is playing with fire.
Exactly.
It could be, most probably, the finest and most honed AI driver assistance product on the market. But instead Tesla have overpromised/oversold it as fully automated self driving system.
Neither Tesla or the lawmakers will accept the liability befalling Tesla and only an idiot would rely on a self driving system with the knowledge that they'll be responsible for any accidents it causes or doesn't prevent.
Tesla cannot apply any caveats if they want to insist it's fully automated.
It'll likely lead to Telsa having to climb down and accept defeat. Perhaps Musk will successfully mask that failure and redress the window.
I can see AI automation getting a free pass in the scope of military equipment because of the nature of the environment that it operates in and because government - military ties are the strongest. I believe that it's never going to be accepted in the scope of public driving.
Like I say, I hope Tesla succeed in 'completing' it as an assistance package but it represents a miscalculation, on the behalf of Tesla, about real life.
It could be, most probably, the finest and most honed AI driver assistance product on the market. But instead Tesla have overpromised/oversold it as fully automated self driving system.
Neither Tesla or the lawmakers will accept the liability befalling Tesla and only an idiot would rely on a self driving system with the knowledge that they'll be responsible for any accidents it causes or doesn't prevent.
Tesla cannot apply any caveats if they want to insist it's fully automated.
It'll likely lead to Telsa having to climb down and accept defeat. Perhaps Musk will successfully mask that failure and redress the window.
I can see AI automation getting a free pass in the scope of military equipment because of the nature of the environment that it operates in and because government - military ties are the strongest. I believe that it's never going to be accepted in the scope of public driving.
Like I say, I hope Tesla succeed in 'completing' it as an assistance package but it represents a miscalculation, on the behalf of Tesla, about real life.
Edited by anonymous-user on Saturday 24th October 20:48
Gassing Station | EV and Alternative Fuels | Top of Page | What's New | My Stuff