Let me make an apology, because that comment wasn't fair. You only referenced the picture in the article.
This is what it comes down to. A woman entered a roadway with her bike and was hit by an automated car which did not slow down so we can conclude the car did not see her at all. I find it hard to believe that there was not some indication that there was a person there that a human would have noticed. Even with a tree in the way you usually see a bike tire or something that would indicate that there is a human there. How did this car not see any sign of a human that would walk onto the road?
No, when you have every opportunity to make the choice before it happens then it IS a case of ethics. Perhaps the AI is too simplistic for that kind of reasoning, in which case it is too early to have automated cars on public roads.
Let me simplify this, because I had some bad information before. The woman was walking a bike across the road. It is not clear to me how a car could approach and not once see any evidence that there was a human there that may walk onto the road. It comes down to how long she was behind the tree whether it fully obscured her for the entire time that the location was in view of the car. If a human saw a bike tire moving from behind a tree, they would slow down because they would come to the conclusion that there is a person there. Why did the self driving car not make the same inference and slow down?
Ok keep dreaming the dream, that automation will ever match humans. Not sure how that will happen if they are missing people with bikes stacked with bags. Doesn't sound very hard to miss.
Ok so that changes the number to 0.000024 accidents per year per licensed driver. Is automation looking better now? Should we go with hours driven by humans? That's bound to look a lot worse for automation.
Furthermore, she was in the CENTER MEDIAN, which means at one point she walked clear across the oncoming lane. Was the car not around when she did that? Is there a bend just up the road?
The driver is in a car that is driving itself, she was probably not attentive as a result. How big is this freaking tree? Why was the car driving faster than it takes a person with a shopping cart loaded with bags to walk from one side of a tree to the other? Was the other side of the tree obscured?
Somehow this girl managed to be totally invisible to the car. That doesn't happen in a road; sure people get partially obscured by other vehicles and traffic controls and the like but they have been designed not to obscure vision. It is important to know why the car didn't see the girl.
That's not really the point though. A computer is not a human. They have a chance to program the choice in a computer, unlike a human. So what are they programming?
There are 254,639,386 registered vehicles in the US which amounts to 0.000019 pedestrian deaths per year per registered vehicle. How is the death to automated vehicle ratio looking so far? And they don't even drive in all conditions yet.
Ok well let's count the days until that full set of recordings is produced. As the days go on, it looks more and more like they either aren't there or are pretty damning.
My point about the videos is that there should be more avaliable than the ones being used to drive the car. There needs to be a driver's view to keep these companies honest. Sure, if the cyclist shows up in the driving feed then is not avoided that is completely damning. We should also have a separate video in order to determine whether a human would have seen this coming.
- Ok well if they have tested that extensively then I guess there will be a completely believable reason how the Uber car completely failed to miss a person running in an open road. I guess she is invisible.
- There are a lot of people on slashdot who know these things, but normally they are the type of people that speak up only if it supports automated cars. So I'm thinking they don't look ahead at all, or are very bad at recognizing shapes that are far ahead or obscured. Oh! I think there have actually been articles on how AI is easy to fool recently.
- I honestly think they are bad at recognizing objects from far away, especially when partially obscured. I hope they are eventually forced to demonstrate that they are actually catching everything that happens in front of them on the road, maybe then I will be proved wrong.
There has to be some definition of how good they should be before they go on a public road though. I'd like to force them to submit a high definition video of a driver's line of site from the car and use that to determine if a human would have seen something like this accident coming.
Slashdot Mirror
I'm sure serveral automation fanbois got a hard reality check today. Tough love.
The question is, would a human have slowed down before the Uber car? oooohhh never mind, the Uber car DIDN'T EVEN SLOW DOWN.
Automated defensive driving skills are too expensive.
Just two different flavors.
Let me make an apology, because that comment wasn't fair. You only referenced the picture in the article.
This is what it comes down to. A woman entered a roadway with her bike and was hit by an automated car which did not slow down so we can conclude the car did not see her at all. I find it hard to believe that there was not some indication that there was a person there that a human would have noticed. Even with a tree in the way you usually see a bike tire or something that would indicate that there is a human there. How did this car not see any sign of a human that would walk onto the road?
No, when you have every opportunity to make the choice before it happens then it IS a case of ethics. Perhaps the AI is too simplistic for that kind of reasoning, in which case it is too early to have automated cars on public roads.
Let me simplify this, because I had some bad information before. The woman was walking a bike across the road. It is not clear to me how a car could approach and not once see any evidence that there was a human there that may walk onto the road. It comes down to how long she was behind the tree whether it fully obscured her for the entire time that the location was in view of the car. If a human saw a bike tire moving from behind a tree, they would slow down because they would come to the conclusion that there is a person there. Why did the self driving car not make the same inference and slow down?
Ok well then don't use articles that are inaccurate to state your case then.
Ok keep dreaming the dream, that automation will ever match humans. Not sure how that will happen if they are missing people with bikes stacked with bags. Doesn't sound very hard to miss.
Ok so that changes the number to 0.000024 accidents per year per licensed driver. Is automation looking better now? Should we go with hours driven by humans? That's bound to look a lot worse for automation.
Furthermore, she was in the CENTER MEDIAN, which means at one point she walked clear across the oncoming lane. Was the car not around when she did that? Is there a bend just up the road?
The driver is in a car that is driving itself, she was probably not attentive as a result. How big is this freaking tree? Why was the car driving faster than it takes a person with a shopping cart loaded with bags to walk from one side of a tree to the other? Was the other side of the tree obscured?
Somehow this girl managed to be totally invisible to the car. That doesn't happen in a road; sure people get partially obscured by other vehicles and traffic controls and the like but they have been designed not to obscure vision. It is important to know why the car didn't see the girl.
That's not really the point though. A computer is not a human. They have a chance to program the choice in a computer, unlike a human. So what are they programming?
Then a reduction needs to be demonstrated. They aren't doing that by hitting pedestrians so soon in the game.
Ok so if I promise you a bridge, will you give me $10,000 now? Because that's basically what the "self driving will save lives" promise is.
It's funny how you can cut off half way through a sentence and make it say what you want to say! Classic manipulation! Classic!
There are 254,639,386 registered vehicles in the US which amounts to 0.000019 pedestrian deaths per year per registered vehicle. How is the death to automated vehicle ratio looking so far? And they don't even drive in all conditions yet.
But ultimately it will kill far less people in total than humans, and that's what's ultimately the most important.
You really have no way of knowing that before it happens. Whether humans will be able to string together sensors and 'AI' to be a successful driver.
Ok well let's count the days until that full set of recordings is produced. As the days go on, it looks more and more like they either aren't there or are pretty damning.
My point about the videos is that there should be more avaliable than the ones being used to drive the car. There needs to be a driver's view to keep these companies honest. Sure, if the cyclist shows up in the driving feed then is not avoided that is completely damning. We should also have a separate video in order to determine whether a human would have seen this coming.
- Ok well if they have tested that extensively then I guess there will be a completely believable reason how the Uber car completely failed to miss a person running in an open road. I guess she is invisible.
- There are a lot of people on slashdot who know these things, but normally they are the type of people that speak up only if it supports automated cars. So I'm thinking they don't look ahead at all, or are very bad at recognizing shapes that are far ahead or obscured. Oh! I think there have actually been articles on how AI is easy to fool recently.
- I honestly think they are bad at recognizing objects from far away, especially when partially obscured. I hope they are eventually forced to demonstrate that they are actually catching everything that happens in front of them on the road, maybe then I will be proved wrong.
No you're not picturing it. You're in a road, and there is a deer crossing the opposing traffic lane; you would see something like that.
Who knows what a human driver would have done.
If there isn't enough data available for submission in order to make that determination then there needs to be more requirements to collect that data.
There has to be some definition of how good they should be before they go on a public road though. I'd like to force them to submit a high definition video of a driver's line of site from the car and use that to determine if a human would have seen something like this accident coming.