4 ms·

> After all, human errors literally kill over a million people a year, whereas the statistical likelihood of dying from a self-driving car is like falling off a

by nsgi 10y ago

> After all, human errors literally kill over a million people a year, whereas the statistical likelihood of dying from a self-driving car is like falling off a building and being struck by lightning on the way down. I don't see any evidence for this. While I'm sure self-driving cars have the potential to be this safe, the death from Tesla's autopilot suggested the risk of death from self-driving cars might currently be similar to that of human-driven cars.

neeleshs 10y ago
https://www.wired.com/2016/08/how-tesla-autopilot-works/ https://www.wired.com/2016/08/how-tesla-autopilot-works/ The human-driven fatality stats are here - https://www-fars.nhtsa.dot.gov/Main/index.aspx https://www-fars.nhtsa.dot.gov/Main/index.aspx Very near, indeed, at least for now
enraged_camel 10y ago
>> the death from Tesla's autopilot suggested the risk of death from self-driving cars might currently be similar to that of human-driven cars. You can't extrapolate from a single incident like that. If you look at statistics, they strongly suggest that self-driving cars are significantly safer, and will become even more so the more vehicles out there are self-driving. edit: so many tesla haters downvoting! :D
greglindahl 10y ago
However, the error bar is huge on the autopilot data. I wish they'd release stats for the much more common events of injury accidents, accidents with major repairs, and accidents with minor repairs.
SubuSS 10y ago
I think I mentioned this elsewhere: It is not about the exact stat comparison between self driving cars and humans. It is actually about 1) the perceived and real adaptability of humans. (given an 'unknown' situation, what does the machine do?) 2) Non epidemic nature of faults in humans. (one version update bug nuking all teslas - we have seen this happen in software, no reason they can't happen in self driving software) 3. Non epidemic nature of coordinated events in humans (GPS satellites fail or leap seconds occur and suddenly all self driving cars go nuts - usually doesn't happen amongst humans). 4. Ethical decision making concerns from machines which are considered 'cold' and 'calculating' and whether that can be construed as murder. (Would the machine choose to kill the pedestrian family over car's current occupants in a purely two choice scenario?) 5. Degree of 'control' considerations. (Would you enjoy a remote turn off switch that denies you entry to your house / mobility etc.) 6. Anonymity considerations. (That's why money economy is still a thing. Possibility of 3rd party tracking vs. knowing about tracking are two different things). FWIW, I am not against either situations, I am just collating info :)

by nsgi 10y ago

neeleshs 10y ago

https://www.wired.com/2016/08/how-tesla-autopilot-works/ https://www.wired.com/2016/08/how-tesla-autopilot-works/ The human-driven fatality stats are here - https://www-fars.nhtsa.dot.gov/Main/index.aspx https://www-fars.nhtsa.dot.gov/Main/index.aspx Very near, indeed, at least for now

enraged_camel 10y ago

>> the death from Tesla's autopilot suggested the risk of death from self-driving cars might currently be similar to that of human-driven cars. You can't extrapolate from a single incident like that. If you look at statistics, they strongly suggest that self-driving cars are significantly safer, and will become even more so the more vehicles out there are self-driving. edit: so many tesla haters downvoting! :D

greglindahl 10y ago

However, the error bar is huge on the autopilot data. I wish they'd release stats for the much more common events of injury accidents, accidents with major repairs, and accidents with minor repairs.

SubuSS 10y ago

I think I mentioned this elsewhere: It is not about the exact stat comparison between self driving cars and humans. It is actually about 1) the perceived and real adaptability of humans. (given an 'unknown' situation, what does the machine do?) 2) Non epidemic nature of faults in humans. (one version update bug nuking all teslas - we have seen this happen in software, no reason they can't happen in self driving software) 3. Non epidemic nature of coordinated events in humans (GPS satellites fail or leap seconds occur and suddenly all self driving cars go nuts - usually doesn't happen amongst humans). 4. Ethical decision making concerns from machines which are considered 'cold' and 'calculating' and whether that can be construed as murder. (Would the machine choose to kill the pedestrian family over car's current occupants in a purely two choice scenario?) 5. Degree of 'control' considerations. (Would you enjoy a remote turn off switch that denies you entry to your house / mobility etc.) 6. Anonymity considerations. (That's why money economy is still a thing. Possibility of 3rd party tracking vs. knowing about tracking are two different things). FWIW, I am not against either situations, I am just collating info :)