RE: Autonomous cars: do you feel lucky
Discussion
otolith said:
DonkeyApple said:
If you take somewhere like Oxford street, pedestrians only stop crossing the side junctions because cars drive deliberately at them to force them to stay on the pavement. Remove that key ability from the car and in fact reverse it so that pedestrians know the cars won't even move off the lights and all those junctions cease to work and the impact of flow being halted in the side streets will kick back throughout Town. You already see this happening when a visitor from the regions doesn't understand and they stop for pedestrians. They can't move forward and they trap everyone behind them.
Maybe if that becomes a problem we can change the law so that the pedestrians are no longer perfectly entitled to do that?A change of the law wouldn't be enough. If you watch side junctions at places like Oxford Street the law already says the pedestrian must stop but it's a 'herd' thing that sees the pedestrians just endlessly flowing into the road until the lead vehicle moves forward and carves a path through by snapping the humans out of the herd and into a personal, self preservation mode.
I think that what you will actually need is big physical changes. Either ban vehicles altogether or installing barriers for total segregation.
Or, you program the car to deliberately drive at pedestrians to maintain the essential fear of personal injury that allows the junctions to work currently. But that doesn't seem likely.
Fury1630 said:
What happens when you're being driven somewhere in your AI car - on a minor road, let's say over Dartmoor or the Peak District & it snows heavily? Does the car just stop on the basis that it's too dangerous to go on? Does it head off into a snowfield on the basis that there's no edge to the road & no obstructions, so no reason not to? Does it stop & call for the emergency services & thus take recourses away from the inevitable crashes?
There are plenty of occasions where humans can adapt their behaviour & do irrational, illogical things to get out of potentially difficult situations. I don't see AI being able to cope in anything other than highly defined town / motorway situations any time soon.
What does a human do in that situation that’s SO clever? You just drive slowly and carefully and sort of hope for the best that the road roughly goes where it looks like it goes. It’s not witchcraft. There are plenty of occasions where humans can adapt their behaviour & do irrational, illogical things to get out of potentially difficult situations. I don't see AI being able to cope in anything other than highly defined town / motorway situations any time soon.
Here are some ideas for the many many ways this problem could be solved:
- proceed slowly with caution based on GPS map data for the road layout and some visual cues like hedgerows.
- could even have various sensors that can penetrate the snow to read for the terrain underneath to ensure you don’t drive into a snow filled ditch - already you’re better off than if you were self driving your 2002 Fiesta which can’t do those things
- In a really sticky situation it could stop and request you take over manual control - or if it’s a totally control-less vehicle, request your authorisation to hand over control to a remote driver service run by the manufacturer.
otolith said:
aaron_2000 said:
I think Top Gear explained it well. A plane can take off, fly and land without pilot interaction, but would you get on a plane if it didn't have a pilot? I know I wouldn't, and as much as I'd use Autopilot on a long dull drive down a motorway, I wouldn't get in a car that didn't have a driver.
I would get in an autonomous plane, if I trusted the engineering. At least it isn't going to have a psychotic break and deliberately fly the plane into a mountain, Andreas Lubitz style.Fury1630 said:
What happens when you're being driven somewhere in your AI car - on a minor road, let's say over Dartmoor or the Peak District & it snows heavily? Does the car just stop on the basis that it's too dangerous to go on? Does it head off into a snowfield on the basis that there's no edge to the road & no obstructions, so no reason not to? Does it stop & call for the emergency services & thus take recourses away from the inevitable crashes?
Radar, Lidar, ultrasound and 360 x 360 degree multi-spectral imaging are pretty old tech now and a bit of an upgrade from the Mk1 eyeball.Fury1630 said:
There are plenty of occasions where humans can adapt their behaviour & do irrational, illogical things to get out of potentially difficult situations. I don't see AI being able to cope in anything other than highly defined town / motorway situations any time soon.
I would say we do counter-intuitive things to solve problems rather than irrational or illogical.The software will evolve and learn over time, humans 150 years ago couldn't drive cars either, somewhat ironically then, most vehicles were autonomous by and large.
imagine where the sensors, software and networks will be in another 50 years.
We all talk like it matters whether we like driving or not, whether we like V8 petrol engines, I am sure there were horse owners that didnt like cars, but they are all dead and progress happened.
Automation and electrification will happen, I dont think non autonomous cars will be banned, like someone said, you can still ride horses on the road and they are really unsuitable, a car will still operate as it always has, even the lawmakers wont really want to ban normal cars, so many classics and other exceptions, plus it will take decades to get to a point of full viability and market penetration, I suspect most will just be like normal cars with controls, but offer different levels of assistance from fully autonomous, through "assisted" where you feel like you are driving but the systems will do most of the work, to complete control.
They will have to co-exist on the road with millions of non-autonomous vehicles from inception, so will have to cope, as they get more prevalent it wont matter as the hard work has been done, they will be fully capable in all scenarios, ever incident will be scrutinised and the software will evolve.
The beauty is that from generation to generation, it will get better, where say we breed, 17 years later we have someone who has to learn from scratch, make mistakes, cause pain, death and expense and some never, ever learn, some people just cant cope with driving.
there will be this initial reticence, there will be issues,accidents and probably some deaths, but eventually AV's will be more trusted than driving yourself, after all they never get tired, agressive, drunk, angry, bored, have 360 degree vision, they will be massively consistent compared to humans, ok they will break but so do humans, you hear of drivers having strokes and heart attacks.
Still a way off but it is very interesting, I will still enjoy driving but sometimes it may be nice to let it get on with it, M6, wet winters evening, nose to tail traffic or having to find somewhere in a city centre.
imagine where the sensors, software and networks will be in another 50 years.
We all talk like it matters whether we like driving or not, whether we like V8 petrol engines, I am sure there were horse owners that didnt like cars, but they are all dead and progress happened.
Automation and electrification will happen, I dont think non autonomous cars will be banned, like someone said, you can still ride horses on the road and they are really unsuitable, a car will still operate as it always has, even the lawmakers wont really want to ban normal cars, so many classics and other exceptions, plus it will take decades to get to a point of full viability and market penetration, I suspect most will just be like normal cars with controls, but offer different levels of assistance from fully autonomous, through "assisted" where you feel like you are driving but the systems will do most of the work, to complete control.
They will have to co-exist on the road with millions of non-autonomous vehicles from inception, so will have to cope, as they get more prevalent it wont matter as the hard work has been done, they will be fully capable in all scenarios, ever incident will be scrutinised and the software will evolve.
The beauty is that from generation to generation, it will get better, where say we breed, 17 years later we have someone who has to learn from scratch, make mistakes, cause pain, death and expense and some never, ever learn, some people just cant cope with driving.
there will be this initial reticence, there will be issues,accidents and probably some deaths, but eventually AV's will be more trusted than driving yourself, after all they never get tired, agressive, drunk, angry, bored, have 360 degree vision, they will be massively consistent compared to humans, ok they will break but so do humans, you hear of drivers having strokes and heart attacks.
Still a way off but it is very interesting, I will still enjoy driving but sometimes it may be nice to let it get on with it, M6, wet winters evening, nose to tail traffic or having to find somewhere in a city centre.
Venturist said:
Fury1630 said:
What happens when you're being driven somewhere in your AI car - on a minor road, let's say over Dartmoor or the Peak District & it snows heavily? Does the car just stop on the basis that it's too dangerous to go on? Does it head off into a snowfield on the basis that there's no edge to the road & no obstructions, so no reason not to? Does it stop & call for the emergency services & thus take recourses away from the inevitable crashes?
There are plenty of occasions where humans can adapt their behaviour & do irrational, illogical things to get out of potentially difficult situations. I don't see AI being able to cope in anything other than highly defined town / motorway situations any time soon.
What does a human do in that situation that’s SO clever? There are plenty of occasions where humans can adapt their behaviour & do irrational, illogical things to get out of potentially difficult situations. I don't see AI being able to cope in anything other than highly defined town / motorway situations any time soon.
"ok, the wheels are going to spin, but at least I'm moving forward"
"that car over there looks damaged - it could've skidded on ice"
"I know this isn't sensible but we're only a mile from the hospital & my wife's gone into labour"
"I won't go left because I know there's a hill down to the river & back up the other side - I'll take the longer way round"
AI is entirely dependant on sensors & sensors tend to not like winter, salt spray, ice, deep water, will you 15 year old AI car's sensors cope?
The A330 that crashed off south America was highly automated immaculately maintained & had GPS, radar altimeters. all kinds of clever stuff, but when the pitot-probe froze it just shut down & handed the problem to the pilots - who were unprepared & lost control.
As I've said, I can see it working well in cities, I just don't think AI is nearly as clever as has been suggested - like the AI chatbot - was it Google? - that had to be taken off line after 12 hours because it learnt so be insulting.
The Vambo said:
culpz said:
I'd also argue that humans are much less susceptible to having such breakdowns.
That would be a pretty flimsy argument. Imagine a computer, machine or system where it's judgement capabilities, ability to perform non judgemental repetitive tasks and even the amount of force its actuators can produce reduces by 15% if it isn't switched off for 8 hours every 24 hours, and reduces by 35-40% if not switched off for 48 hours.
It wouldn't even get off the drawing board but it is a basic part of being human, what about reduced performance caused by prescription drugs? What about reduced performance caused by low level illness? What about the million things that affect peoples emotions and that effects their capabilities?
We seem obsessed with basically building machines that can do what we can. I find that odd when AI is, in fact, a contrast to human intelligence.
culpz said:
Not at all. AI, at the end of the day, is simply a replication of the brain and it's capabilities. Yet, it will never ever quite be as complex or as intricate.
We seem obsessed with basically building machines that can do what we can. I find that odd when AI is, in fact, a contrast to human intelligence.
I think that is the wrong way round, autonomous vehicles are the goal and if it could be successfully built using an on board abacus or Babbage difference engine powered Machine Learning, they would.We seem obsessed with basically building machines that can do what we can. I find that odd when AI is, in fact, a contrast to human intelligence.
AI is the solution, not the aim.
The Vambo said:
culpz said:
Not at all. AI, at the end of the day, is simply a replication of the brain and it's capabilities. Yet, it will never ever quite be as complex or as intricate.
We seem obsessed with basically building machines that can do what we can. I find that odd when AI is, in fact, a contrast to human intelligence.
I think that is the wrong way round, autonomous vehicles are the goal and if it could be successfully built using an on board abacus or Babbage difference engine powered Machine Learning, they would.We seem obsessed with basically building machines that can do what we can. I find that odd when AI is, in fact, a contrast to human intelligence.
AI is the solution, not the aim.
I think we really need to learn to walk before we can run. Otherwise, we're just pissing in the wind.
I think there is a perception that autonomous vehicles need the kind of AI that is currently science fiction. They don't. They don't even need the sort that is currently running in research labs and being taught to be racist by douches on the internet. The people working on it don't even call it AI (although to be fair that is kind of a persistent trope in the development of machine intelligence, in that we only really call it AI if we can't do it yet).
I think that AI research will gradually feed into making autonomous cars smarter - it's likely to be the kind of thing that gets you to upgrade to the newest model - but I think we're a long way from that. I think that in order to be deemed trustworthy and verifiable for public use, first generation models need to be highly deterministic in their behaviour, which a genuine AI will not be.
I think that AI research will gradually feed into making autonomous cars smarter - it's likely to be the kind of thing that gets you to upgrade to the newest model - but I think we're a long way from that. I think that in order to be deemed trustworthy and verifiable for public use, first generation models need to be highly deterministic in their behaviour, which a genuine AI will not be.
culpz said:
Not at all. AI, at the end of the day, is simply a replication of the brain and it's capabilities. Yet, it will never ever quite be as complex or as intricate.
Not wishing to sideline this into a discussion of AI (there is one here) but there is a reputable idea that once we can build smart enough AI to exponentially bootstrap its own development we could easily end up creating AI magnitudes more intelligent than us.otolith said:
Not wishing to sideline this into a discussion of AI (there is one here) but there is a reputable idea that once we can build smart enough AI to exponentially bootstrap its own development we could easily end up creating AI magnitudes more intelligent than us.
I could agree that AI, in the future, could become consistently more intelligent than us. I don't believe that AI can become more intelligent that us full-stop, though.Paul Dishman said:
J4CKO said:
I think the problem with Automomous cars will still be humans, ever seen those Boston Dynamics videos where they boot the "BigDog" robot, the robot doesnt react other than trying to stay upright, drivers will learnt hat autonomous cars will defer more than Mr thrusty in his RR or Miss texter in her Mini, they will be meek and mild, like an electronic version of my auntie in her 1993 Micra.
They'll be easy to overtake and pull out in front of, in that case. Chop in and the autonomous car will do an emergency stop. Great new game.The classic use-case is the old-fashioned highway hold-up. If an autonomous vehicle is utterly deferential, one only needs to stand in the middle of the road, in a rural area, to bring your quarry to a halt.
anonymous said:
[redacted]
My point was not to correlate those two things together, but more to prioritise rationale. The main focus at current, in the development of cars, is getting rid of the ICE in place of EV's. At least, that's where the research and money is being pumped into.However, we shouldn't really be getting into the specifics of the technology capable of creating autonomous vehicles, when we can't even get electric technology to power our cars. That's all i was really getting at.
BricktopST205 said:
Yes humans can make mistakes but cars can not scan the situation ahead like we can do. Can they see a child playing football 100 meters down the road or can they see that guy coming towards you on his mobile phone. We can be preventative whereas all the self driving cars can do is react.
The only way I can see it working is if 100% of the road is autonomous.
You know, its one thing for people to ask if computers can do x, y, and z. But did you ever stop to wonder if all people can also do the same x, y, and z?The only way I can see it working is if 100% of the road is autonomous.
Because there are a lot of people I see driving today who have absolutely s
t skills, and I wonder how the fk they got their drivers licenses. Not to mention all the drop kicks who drive around in an almost perpetual state of road rage. And don't even get me started on the numbers of people driving under the influence of a substance.Gassing Station | General Gassing | Top of Page | What's New | My Stuff