4 ms·

I really don't understand the pervasive idea that we are going to hold robot cars to so much of a higher standard than human drivers. I'll take 1.2x . Human dri

I really don't understand the pervasive idea that we are going to hold robot cars to so much of a higher standard than human drivers. I'll take 1.2x . Human drivers kill a lot of people every year in what are largely preventable accidents. Lets use robot cars in places that it makes sense to do so and save some lives and they'll get better over time.

novok 5y ago
I think the issue is the kind of accidents will be very different from humans, uncomfortable and very fatal. You also have to compare against sober and awake drivers, not 'the typical driver', because you have a choice to be sober and awake if you're more safety minded, while you don't really have those choices with AI driving. Like tesla cars going straight into concrete barriers on highway exits instead of down the 2 lane options on autopilot mode.
fma 5y ago
It's usually the optics/media that will torpedo the effort. "Big scary robot car was in an accident". Proof? 13 American soldiers died in Afghanistan in a combat zone and people are up in arms. 1800 Americans died yesterday from covid while going about their daily lives and it's pretty mum.
joe_the_user 5y ago
Self-driving cars are nowhere as safe as humans per mile. People touting self-driving cars like to talk about human accidents but people are remarkably safe per mile. Self-driving cars don't have that many accidents attributed to them but they haven't driven very much compared to humans.
- soared 5y ago
  Source?
  - joe_the_user 5y ago
    Wikipedia gives a summary of the accident rate in the US [1], humans drive millions of miles without accidents. Filip Piekniewski has an ongoing series on self-driving car problems. [2] (I'd look up the earlier articles in the series if I had time). This article I googled [3] give some statistics: ") At the moment, self-driving cars have a higher rate of accidents compared to human-driven cars, but the injuries are less serious. On average, there are 9.1 self-driving car accidents per million miles driven, while the same rate is 4.1 crashes per million miles for regular vehicles." However, nearly all self-driving cars are driven in highly controlled conditions, the safest driving areas available and/or an operator ready to take over - yet they still have a higher rate of accidents than human drivers. There's a reason self-driving cars aren't being deployed yet. [1] https://en.wikipedia.org/wiki/Motor_vehicle_fatality_rate_in_U.S._by_year https://en.wikipedia.org/wiki/Motor_vehicle_fatality_rate_in... [2] https://blog.piekniewski.info/2021/05/12/ai-mid-2021/ https://blog.piekniewski.info/2021/05/12/ai-mid-2021/ [3] https://carsurance.net/blog/self-driving-car-statistics/ https://carsurance.net/blog/self-driving-car-statistics/
  - morpheos137 5y ago
    >source The real world. See also: https://blog.piekniewski.info/2021/05/12/ai-mid-2021/ https://blog.piekniewski.info/2021/05/12/ai-mid-2021/ Self-driving aint happening until we get AGI drivers. Reading these comments trivialising the driving task makes me wonder if such commenters have ever actually driven or if they are just prone to magical thinking. As of 2021 computers can not reliably and flexibly out reason people about open ended real world problems. Safe driving in the real world requires such reasoning. It is edge cases all the way down. Ergo self driving is not happening anytime soon. When I see HN commenters say the robocar sees or thinks I wonder if these people are for real. It would probably be easier to make a robo physician who gives more reliable diagnoses than a human physician than it would be to make a self driving car that out performs humans.
    - joe_the_user 5y ago
      Reading these comments trivialising the driving task makes me wonder if such commenters have ever actually driven or if they are just prone to magical thinking. I would say it's a testament to the human ability to reduce very complex tasks to very simple descriptions that can be kept out of rational awareness. The impressive thing is that even the wrong beliefs of a given person work for them. Remember, the people effectively arguing that driving is "easy" still understand that humans fail at it regularly. So it's mismatch in their understanding of how smart humans are, something that probably doesn't impact their daily lives.

novok 5y ago

I think the issue is the kind of accidents will be very different from humans, uncomfortable and very fatal. You also have to compare against sober and awake drivers, not 'the typical driver', because you have a choice to be sober and awake if you're more safety minded, while you don't really have those choices with AI driving. Like tesla cars going straight into concrete barriers on highway exits instead of down the 2 lane options on autopilot mode.

fma 5y ago

It's usually the optics/media that will torpedo the effort. "Big scary robot car was in an accident". Proof? 13 American soldiers died in Afghanistan in a combat zone and people are up in arms. 1800 Americans died yesterday from covid while going about their daily lives and it's pretty mum.

joe_the_user 5y ago

Self-driving cars are nowhere as safe as humans per mile. People touting self-driving cars like to talk about human accidents but people are remarkably safe per mile. Self-driving cars don't have that many accidents attributed to them but they haven't driven very much compared to humans.

soared 5y ago
Source?
- joe_the_user 5y ago
  Wikipedia gives a summary of the accident rate in the US [1], humans drive millions of miles without accidents. Filip Piekniewski has an ongoing series on self-driving car problems. [2] (I'd look up the earlier articles in the series if I had time). This article I googled [3] give some statistics: ") At the moment, self-driving cars have a higher rate of accidents compared to human-driven cars, but the injuries are less serious. On average, there are 9.1 self-driving car accidents per million miles driven, while the same rate is 4.1 crashes per million miles for regular vehicles." However, nearly all self-driving cars are driven in highly controlled conditions, the safest driving areas available and/or an operator ready to take over - yet they still have a higher rate of accidents than human drivers. There's a reason self-driving cars aren't being deployed yet. [1] https://en.wikipedia.org/wiki/Motor_vehicle_fatality_rate_in_U.S._by_year https://en.wikipedia.org/wiki/Motor_vehicle_fatality_rate_in... [2] https://blog.piekniewski.info/2021/05/12/ai-mid-2021/ https://blog.piekniewski.info/2021/05/12/ai-mid-2021/ [3] https://carsurance.net/blog/self-driving-car-statistics/ https://carsurance.net/blog/self-driving-car-statistics/
- morpheos137 5y ago
  >source The real world. See also: https://blog.piekniewski.info/2021/05/12/ai-mid-2021/ https://blog.piekniewski.info/2021/05/12/ai-mid-2021/ Self-driving aint happening until we get AGI drivers. Reading these comments trivialising the driving task makes me wonder if such commenters have ever actually driven or if they are just prone to magical thinking. As of 2021 computers can not reliably and flexibly out reason people about open ended real world problems. Safe driving in the real world requires such reasoning. It is edge cases all the way down. Ergo self driving is not happening anytime soon. When I see HN commenters say the robocar sees or thinks I wonder if these people are for real. It would probably be easier to make a robo physician who gives more reliable diagnoses than a human physician than it would be to make a self driving car that out performs humans.
  - joe_the_user 5y ago
    Reading these comments trivialising the driving task makes me wonder if such commenters have ever actually driven or if they are just prone to magical thinking. I would say it's a testament to the human ability to reduce very complex tasks to very simple descriptions that can be kept out of rational awareness. The impressive thing is that even the wrong beliefs of a given person work for them. Remember, the people effectively arguing that driving is "easy" still understand that humans fail at it regularly. So it's mismatch in their understanding of how smart humans are, something that probably doesn't impact their daily lives.