3 ms·

Your comment prompted me to look at stats of injuries per km driven. [1] https://en.wikipedia.org/wiki/List_of_countries_by_traffic-related_death_rate https://

by thesz 6y ago

Your comment prompted me to look at stats of injuries per km driven. [1] https://en.wikipedia.org/wiki/List_of_countries_by_traffic-related_death_rate https://en.wikipedia.org/wiki/List_of_countries_by_traffic-r... Australia is precisely at the middle. Right now it is 13th. So, I would like your opinion on driving in UK which has Isle of Man in it. It has much lower accidents rate per km driven. And also three times as much population. As for AI-reeve theme, AI is inherently biased - due to training sets used. Imagine being one with face expression that is a little bit closer to what AI considers not normal (slower blinking, less eye movement, etc - something like that). You certainly will be held to a higher standard than others just because chances are you will be considered not normal by AI. "I don't like your face, so I'll punish you for that" but from a robot. Russia have (and had from beginning of century if not earlier) a system that reads vital signs of train locomotive operators. It is consisted of special bracelet, pedal and series of signals. Bracelet reads heart rate and skin conductance and determines whether person has good health and is attentive. Pedal is used to tell the system that operator is attentive - it should be pressed after a signal is issued. If operator does not respond to series of signals, emergency braking is activated. That system works very well, we have rather low train accidents rate. But recently I heard about its replacement and it is all about face recognition, all for the sake of making it cheaper. And I guess it is as not good as the system employed by Amazon.

tomhoward 6y ago
Yeah I noticed that fatalities-per-billion VKT stat. One thing is that it's improved since the 2016 year that the Wikipedia data is from. In 2018 Australia's figure was 4.3, 7th on the list (though the UK isn't included in that report - recent UK figures seem hard to find). Australia's figure was better than the OECD average and much of Europe. But a reason for Australia's fatalities per billion VKT figure being so much higher than the UK would be its vastness, and the distances some people need travel in remote ares. According to this report [1] (see fig 1.1 - page 3), Australia's fatality rate increases dramatically the more remote you go. Presumably it's a combination of fatigue, the quality of roads and low police presence that leads to more dangerous conditions in remote areas. As for the rest of your comment, there's a huge assumption in all this that Amazon's primary goal here is punishing drivers, rather than just detecting if they're tired and activating a warning signal for their own safety. And surely as the training data includes all the drivers in their network it will become more accurate. [1] https://www.bitre.gov.au/sites/default/files/documents/international_2018.pdf https://www.bitre.gov.au/sites/default/files/documents/inter...

by thesz 6y ago

tomhoward 6y ago

Yeah I noticed that fatalities-per-billion VKT stat. One thing is that it's improved since the 2016 year that the Wikipedia data is from. In 2018 Australia's figure was 4.3, 7th on the list (though the UK isn't included in that report - recent UK figures seem hard to find). Australia's figure was better than the OECD average and much of Europe. But a reason for Australia's fatalities per billion VKT figure being so much higher than the UK would be its vastness, and the distances some people need travel in remote ares. According to this report [1] (see fig 1.1 - page 3), Australia's fatality rate increases dramatically the more remote you go. Presumably it's a combination of fatigue, the quality of roads and low police presence that leads to more dangerous conditions in remote areas. As for the rest of your comment, there's a huge assumption in all this that Amazon's primary goal here is punishing drivers, rather than just detecting if they're tired and activating a warning signal for their own safety. And surely as the training data includes all the drivers in their network it will become more accurate. [1] https://www.bitre.gov.au/sites/default/files/documents/international_2018.pdf https://www.bitre.gov.au/sites/default/files/documents/inter...