It's only really 9 9s of reliability if you assume somebody will definitely be killed as soon as you have a 10-second lapse in concentration, which is obviously not true.

In motorway driving you could probably shut your eyes for 10 seconds every 20 seconds and be perfectly fine almost all of the time. Especially at night time, or other quiet times of day.

If people actually did that, then someone would crash, and that crash would be factored into the final accident rate.

So that final reliability rates includes any and all things humans do while driving. (Not what they could do, what they actually do.)

Yes but what I'm saying is you don't need 9 9s of good driving in order to get 9 9s of accident-free driving. You only have an accident when you're driving badly and are exceptionally unlucky (I suppose the amount of bad luck required decreases as the badness of the driving increases).

In the scenario I mentioned you'd have at best a 50% duty cycle of good driving, but the accident rate would be substantially lower than 50% every 10 seconds.

> There's this belief that humans are horrible at driving, etc, etc. Until you run that math and realize, no, they're not. There's just a lot of driving going on.

I've been hammering this concept for a while :) Unimpaired, non distracted humans are actually really good at driving. We're optimized for processing visual inputs and making fuzzy decisions, have high dexterity and coordination, and fear death. All of that is alien to software.

Oddly enough, even impaired drivers are probably pretty good compared to some current self-driving cars. I'd give good odds that someone with a 0.10% BAC would have swerved or stopped for that woman that the Uber car hit.

Yep sadly too the Uber car didn't even slow after the hit, so even a drunk person probably would have done different things... Enough maybe to be non-fatal. Not saying we should replace Ubers sdv fleet with drunk drivers obviously the entire program is scrapped now.

> If the cause of a fatality is 10 seconds of failure, not 5 minutes, humans have greater than 9 9s of reliability.

If you're driving by the textbook, then you have two seconds to react to the car in front braking; the assumption being that you take one second to react normally, and get one extra second for safety.

On the other hand, in something like a play street law grants you exactly zero seconds of response time.

The question is now, of course, whether only those <<10 s intervals count as the failure, or your entire approach to driving (e.g. driving while texting).

> then you have two seconds

Which takes us to 10 9s of reliability. Are there any computers with that level of reliability?

> The question is now, of course, whether only those <<10 s intervals count as the failure, or your entire approach to driving (e.g. driving while texting).

I thought about this all day. I feel that from the moment you can no longer avoid to accident (no matter what you do), until it occurs, is the correct timeframe to measure. Not sure what that number is though.

Say you texted, got in a dangerous situation, then corrected. Is that really a failure? It's a risk of course, and enough people doing that will increase the final death tally. But for each individual driver it's not a failure, if it's not a failure then you can not count it.

i.e. doing it any other way would be counting it twice: Once for those situations that actually caused an accident, and again for taking the risk.

Or put another way, taking a risk and winning is not a failure. (Otherwise where do you draw the line on what a risk is?)

Fatalities seems like a pretty low bar to measure failures by... I would prefer to measure reliability by reported accidents per 100k miles. This stat seems to be 183 per million miles in 2009, compared to 1.13 for fatalities per million miles. So quite enough to shave 7 9's down to 5ish...

The problem is that a lot of accidents are not reported and even injury severity can be very subjective (especially if there's an interest to do so[1]). I prefer counting fatalities because it's a hard endpoint - bodies are hard to hide.

[1] https://www.youtube.com/watch?v=_ogxZxu6cjM

Collisions with trees and with other vehicles? Or collisions with pedestrians? Maybe we're fine with five 9s of reliability with inanimate objects and with other 2-ton metal crash protection cages but prefer seven 9s with fragile unprotected human bodies (which are far more likely to end up as fatalities than just crashes).

> So quite enough to shave 7 9's down to 5ish...

From other comments in this thread, the correct number is actually 10 9's, not 7. So using the accident rate instead of fatality takes you to 8 9's.

I agree with the basic point, but does the 1.13 per 100 million miles figure include freeway driving?

Yes.

freeway driving causes less accidents than street driving

I think you were downvoted to dead because people saw this as an objection to the parent; more charitably, I assume you're just adding more color - but you might want to telegraph your intentions a bit clearer.

He was not downvoted to dead, his account is banned, all his posts show up dead.

Gotcha