5 ms·

Software is far from 100% trustworthy. With the current state of art we cannot be certain that any software is secure, hack-proof and provable correct under al

Software is far from 100% trustworthy. With the current state of art we cannot be certain that any software is secure, hack-proof and provable correct under all operating conditions. A self-driving car is a canonical example of lots of software, sensors and actuators working seamlessly and robustly at all times. The recent issues with VW, Jeep, Toyota clearly demonstrates that car manufacturers can't even get human controlled cars to be 100% trustworthy. What chance is there that they'll be any better with a self-driving car?

superuser2 11y ago
It's utterly ridiculous that "100% trustworthy" would be considered relevant here. All self-driving cars have to be is better than humans, who are approximately 0% trustworthy. If a self-driving car is better than a human who is drunk, eating, and texting all at the same time, then it's already a safety improvement.
- alexandrerond 11y ago
  And still a human seems to know how to deal with a 0% trustworthy human better on the road. We are still busy making self-driving cars follow the rules, but much of the driving is to predict how other people bend them...
  - GFischer 11y ago
    "still a human seems to know how to deal with a 0% trustworthy human better on the road" I don't think they cope that well. If you're under 35, a car accident is your main worry (other than suicide): Motor vehicle crashes are a leading cause of death in the U.S. More than 33,000 people died from motor vehicle crashes in 2013 alone http://www.cdc.gov/injury/wisqars/overview/key_data.html http://www.cdc.gov/injury/wisqars/overview/key_data.html http://www.ons.gov.uk/ons/rel/vsob1/mortality-statistics--deaths-registered-in-england-and-wales--series-dr-/2012/sty-causes-of-death.html http://www.ons.gov.uk/ons/rel/vsob1/mortality-statistics--de...
    - alexandrerond 11y ago
      In fact you're right. The thing is, can you imagine sitting on a self-driving car minding your business in a busy road shared with human drivers? Since not all cars will be automated, the success depends on convincing everyone that they're still safer in a world dominated by all the other shitty drivers. I read somewhere that Google cars accidents had always been caused by the other person. Question is, would a human have "seen it coming"? Can a human judge that someone is driving agressively, or shyly better than a computer? Can a human infer from a slow-driving car that it is probably trying to park despite not using the blinkers? And which such informations, make better decisions to avoid others being imprudent? Still, probably most accidents are to blame on the drivers themselves and their own mistakes and self-driving cars would fix that, but what probably is really going to count if you want to sell such car is probably the "sense of security" you can get rather than the hard stats. That's why I don't see fully self-driving cars being adopted by anyone anytime soon as some pretend, and it is much more realistic to think of forms of "assisted driving" (where you still have to drive and pay attention) and still need driving controls in the vehicle.
- impendia 11y ago
  > who are approximately 0% trustworthy. I disagree with this. There are many routine driving situations where if someone else does something stupid, you're probably dead. Given how many cars are on the roads, I'm amazed that there are as few accidents as there are. (I submit as anecdotal evidence: if you're reading this, then you're not dead.) Indeed, it is relatively uncommon for people to limit or reduce their driving on account of the risk of an accident. Accidents do happen, but for practical purposes most people round down the risk to zero. That said, I agree with your main point. If humnas are 99.99% trustworthy and automatic cars are 99.995% so, then it seems clear that we should allow them on the roads.

superuser2 11y ago

It's utterly ridiculous that "100% trustworthy" would be considered relevant here. All self-driving cars have to be is better than humans, who are approximately 0% trustworthy. If a self-driving car is better than a human who is drunk, eating, and texting all at the same time, then it's already a safety improvement.

alexandrerond 11y ago
And still a human seems to know how to deal with a 0% trustworthy human better on the road. We are still busy making self-driving cars follow the rules, but much of the driving is to predict how other people bend them...
- GFischer 11y ago
  "still a human seems to know how to deal with a 0% trustworthy human better on the road" I don't think they cope that well. If you're under 35, a car accident is your main worry (other than suicide): Motor vehicle crashes are a leading cause of death in the U.S. More than 33,000 people died from motor vehicle crashes in 2013 alone http://www.cdc.gov/injury/wisqars/overview/key_data.html http://www.cdc.gov/injury/wisqars/overview/key_data.html http://www.ons.gov.uk/ons/rel/vsob1/mortality-statistics--deaths-registered-in-england-and-wales--series-dr-/2012/sty-causes-of-death.html http://www.ons.gov.uk/ons/rel/vsob1/mortality-statistics--de...
  - alexandrerond 11y ago
    In fact you're right. The thing is, can you imagine sitting on a self-driving car minding your business in a busy road shared with human drivers? Since not all cars will be automated, the success depends on convincing everyone that they're still safer in a world dominated by all the other shitty drivers. I read somewhere that Google cars accidents had always been caused by the other person. Question is, would a human have "seen it coming"? Can a human judge that someone is driving agressively, or shyly better than a computer? Can a human infer from a slow-driving car that it is probably trying to park despite not using the blinkers? And which such informations, make better decisions to avoid others being imprudent? Still, probably most accidents are to blame on the drivers themselves and their own mistakes and self-driving cars would fix that, but what probably is really going to count if you want to sell such car is probably the "sense of security" you can get rather than the hard stats. That's why I don't see fully self-driving cars being adopted by anyone anytime soon as some pretend, and it is much more realistic to think of forms of "assisted driving" (where you still have to drive and pay attention) and still need driving controls in the vehicle.
impendia 11y ago
> who are approximately 0% trustworthy. I disagree with this. There are many routine driving situations where if someone else does something stupid, you're probably dead. Given how many cars are on the roads, I'm amazed that there are as few accidents as there are. (I submit as anecdotal evidence: if you're reading this, then you're not dead.) Indeed, it is relatively uncommon for people to limit or reduce their driving on account of the risk of an accident. Accidents do happen, but for practical purposes most people round down the risk to zero. That said, I agree with your main point. If humnas are 99.99% trustworthy and automatic cars are 99.995% so, then it seems clear that we should allow them on the roads.