[astazangasta]

The thing I keep emphasizing in this conversation is how terribly good human drivers are, on average. In the US, there are three fatalities for every 10 billion passenger miles traveled. That's tremendous. A machine that is currently incapable of driving in rain and unable to see a child in the street has a long way to go to hit this mark.

Musk is here to sell his car. He'll say all sorts of bullshit to make it look sexier.

[ohsnap]

What could be tremendous about that statistic? It seems like you are trying to contrast a small number with a big number to give a sense of some false magnitude. I could say people only murder other people once in every 100 billion seconds - thats tremendous!!

[Retric]

That's a load of FUD.

Modern Self driving cars track people on sidewalks not just the street. As to driving in the rain, they can do it, but like poeple there less safe at highway speeds in the rain. https://www.youtube.com/watch?v=YXylqtEQ0tk (6:45, 9:00) Also, 3:52 demos stopping for someone in the road.

They even read stop lights (8:00).

PS: If you actually look at that left turn on @9:00 the car has a much better idea what's around it than people do.

[FranOntanaya]

Miles per passenger depend on too many factors. I'd stick with 1 death per 6,200 driver's licenses (US, 2009)

[jeffdavis]

Deaths per mile is more useful because it compares the cost (deaths) with the benefit (miles traveled per passenger). Additionally, we can compare directly against alternatives, such as buses, air travel, and self-driving cars.

Deaths per driver's license doesn't tell us the benefit we are getting, and doesn't give us a way to make comparisons with alternatives.

[apendleton]

I'd argue that deaths per mile is also misleading, however, as it misstates the actual benefit; people don't directly benefit from a particular number of miles, but rather from a particular set of trips, whose length will tend to vary by mode. People who choose a car-oriented lifestyle will tend, on average, to make longer trips than those who choose lifestyles built around transportation by walking, biking, or using transit, but I would argue that the urban cyclist biking two miles to work derives the same benefit (getting from home to work) as the suburban car commuter driving twenty miles. For comparing cars to bikes, in particular, the relative safety depends on whether you go with a per-mile or per-trip metric, with the winners being car and bike, respectively.

[jeffdavis]

I mostly agree, but that's true any time you wrap a complex problem up into a single number. I am merely arguing that, if we pick a single number, deaths-per-passenger-mile is much better than deaths-per-licensed-driver.

I guess you are saying that deaths-per-trip is an even better number. That would be interesting to see, I agree, but as an additional number, not a replacement.

[maxerickson]

I haven't looked to see where they got it, but Wikipedia gives a figure of 7.6 deaths / billion km, in terms of vehicle miles, not passenger miles.

http://en.wikipedia.org/wiki/List_of_countries_by_traffic-re...

That's ~ 12 deaths / billion miles. So there is a big discrepancy with the figure stated by astazangasta. Passenger vs vehicle will make up for some if it, but even an assumption like 4 people per vehicle mile still puts it at 3 vs 30.

(this isn't really a response to your comment, but it seems like a good place in the thread to put it)

[Retric]

Self-driving cars seem to be ~1:1 with driver licenses. A licenced driver drives at most one car at a time.

[jeffdavis]

There's obviously a connection between driver's licenses and miles traveled, but if you have the latter number, why use the former? It makes the benefit more obscure and reduces the accuracy (for instance, if passengers of self-driving cars drive more or less than passengers of human-driven cars).

Also, it doesn't offer a good way to compare against things like air travel.

[maxerickson]

There is a sense in which the part of a self-driving car that is licensed will be driving many cars at once.

(There will of course be something like 1 instance of it per vehicle, I'm commenting on the semantic awkwardness, not trying to be pedantic)

[omellet]

A better measure would be deaths per driver decision, if you could estimate such a thing.

[jeffdavis]

That may measure the average driving ability; but it doesn't give you any hint as to the benefits (more driver decisions don't seem like a benefit to me), and it also doesn't give a good way to compare with alternatives.

[GraffitiTim]

That statistic is orders of magnitude off. There are 30,000 driving fatalities per year. At the ratio you stated the number of miles driven per person per year would be absurd.