[1]https://en.wikipedia.org/wiki/Motor_vehicle_fatality_rate_in...

Also, the link to the NHTSA doesn't work, but I'd be interested to read the primary research. Concern is they're just taking accident rates for sober drivers and BAC 0.08 drivers, which isn't a fair comparison because they aren't time matched. As in, drunk driving is a lot more likely to happen very late at night/early in the morning when bars close, when traffic is extremely low and accident rates are also very low. If your chances are 3x the chances of a person driving any time at all, you're much more than 3x the chances of someone driving unimpaired at 2 in the morning.

Of course, even if they time match the rates, there's still the reality that a lot of sober people driving at 2 in the morning are still impaired. Tired driving is also quite dangerous.

I could have gone for a shock stat - 100000 people a year are injured in accidents involving drunk drivers. 10000 people are killed.

I could have taken the position that a typical drunk driver is probably actually 10x as likely to have an accident as a sober driver, based on those NHTSA numbers.

But I didn’t need to to make the point that a self driving car could be literally twice as likely as an average driver to crash and that would still be ‘better than a drunk driver’.

Granted, it becomes an issue when people who wouldn’t drive drunk decide to use self-driving functionality simply because is there. But, there will be a point where it crosses over into being a net improvement, and that point may very well be a fair bit before meeting parity with humans. Obviously it’s not quite that simple (not all failure modes are equal) but it’s not that far off, either…

People who may not have chosen to drive (e.g., tired, drunk, impaired, or otherwise incapable of driving) might choose to do so since they can "lean on" the car doing it for them, and if the car encounters a situation it can't deal with the driver is completely incapable of handling it.

I think the answer right now is no, we're still requiring the driver to be able to take over, and our thought experiment presupposes they're inebriated.

On the other hand, it's obviously better to have the computer take a shot (no pun intended) at getting home than letting the driver do it themselves, even today. At what point does that moral balance tip? It seems like we're not too far off from the answer being "obviously let the car drive, whatever the drunk person was going to do is probably riskier for both themselves and everyone else".

I'm way more interested in self-driving (and, more broadly, driving assistance) for emergencies like someone having a heart attack while at the wheel. For example, I feel a lot better now that, should I somehow die or lose consciousness while driving (but not due to being drunk!), my car with autonomous breaking will mostly keep going in whatever direction I had it but with a much better chance of stopping before causing someone else harm.

I think a lot of people are unfamiliar with rural America -- the nearest bar might be 10 miles away down a country road with nothing else around (which was the case in my hometown... well more of a "village" than a town), and your friend may live 10 miles on the other side of that bar, so he may not want to drive that far out of his way so you can get drunk and he can't. And taxis and buses aren't an option.

Relying on humans to make the correct and safe choice to not drive while drunk is a losing proposition, our DUI death rate shows that.

Even if a smart car can't be trusted to drive you, it should be able to notice that you're drunk and refuse to let you drive. My car notices when I'm losing focus and gives me an alert reminder, so it should be able to notice when I'm drunk.

How is the bar not sued out of existence? There is no plausible way to travel to and from the bar, other than a private automobile, and no taxi service within an hour's drive. There are no houses nearby with clientele that could walk to the bar. Why isn't that considered an attractive nuisance like a swimming pool?

So, hopefully not very often :) But it's one of those things where, since I could afford to change the car, I felt I had to go ahead and make that investment in safety.

I think they're wrong to do this, because the worst potential outcome of drunk driving is so devastating and catastrophic that it's personally optimal to eliminate that risk to zero. But the fact is that this isn't really how humans weigh risk.

Obviously anti-drunk-driving campaigns have been somewhat effective. People take this problem more seriously than they did 40 years ago. But I wonder if we aren't at a point of diminishing returns. Some people are going to drive drunk, no matter how many times you tell them to take an Uber.

Or he can just stop on his way home.

We both know what the right thing to do is, but one has significant costs the other doesn't.

Or imagine he does stop at the bar on his way home and tells himself he'll just have one beer, but he ends up having four and shouldn't drive. Now he has to get an Uber, take it home, get another Uber before work in the morning, and take it back to the bar to get his car. And then, finally, head to work.

Or he can just drive home.

Again, we both know what the right thing to do is, but one has significant costs the other doesn't.

Even if that were true, I have a hard time imagining that the cost of self driving packages won't continue to decrease until it's included in literally every new car. Just a matter of time (assuming someone is able to make a fully working version).

I don't live in an area that has Uber, so I'm not that familiar with it, only used it a few times when traveling :)

My impression of the the way drivers are compensated is that the hours, fuel, and wear on your vehicle make it kind of slim margins.

Other than when Uber deeply discounts to drive growth, it's always been cheaper to drive yourself than to pay Uber. The reason you pay Uber is for other reasons (e.g. convenience, cost of parking, being away from your vehicle, intoxication).

Now we're onto something!

Self-driving cars which are merely as safe as driving with a 0.08BAC will still greatly reduce traffic accidents. Of course, no one will report it as such. Having something that's paying attention 100% of the time is going to be a huge boon for traffic safety. Spending 15 seconds picking a song on your phone is all it takes to rear end a car or blow through an intersection.

A 100% increase would be 2x as likely.

I gave the ‘better than a drunk driver’ claim the benefit of the doubt and allowed it to be merely 2x as likely to crash than a sober driver, since that is still ‘better’ than a driver who is exactly at the federal US legal limit (and who will crash 3x as often).

What are the odds of a sober driver crashing a car?

[1] https://www.allstate.com/resources/allstate/attachments/best... Page 4

[2] https://www.bts.gov/content/motor-vehicle-safety-data