Hacker News Reader

Comment by bryanlarsen | Hacker News Reader

13X is way more impressive than it seems at first glance.

Let's take a simplistic model of accidents: that the average driver is at fault in an accident 50% of the time. So a perfect driver would only halve the number of accidents -- they only eliminate the accidents where they would otherwise have been at fault.

But Waymo's numbers are better than the "perfect" driver above. How is that possible? Because in most accidents the blame is not split 0%/100%. You can avoid a lot of accidents with defensive and safe driving.

View on HN

> You can avoid a lot of accidents

More than 1/2 of roadway fatalities involve alcohol or drugs. An oversized fraction of fatalities are represented by young men under 24. 1/6 of all fatalities are motorcycles. 1/6 of all fatalities are pedestrians being struck by a vehicle.

Do you have a solution for preventing humans from driving while under the influence of alcohol or drugs? Or even preventing humans from driving while texting?

The insurance industry will solve that on their own as technology advances. It simply won't be worth insuring a human that uses alcohol/drugs.

Most accidents are caused by drunk drivers, distracted drivers, jackass drivers and others, who together still constitute a minority, and there are plenty single-vehicle crashes. 13x improvement over average may still constitute diminished safety for a diligent sober driver.

Yes, but Waymo also has to drive on the road with those drivers, and these stats include crashes that are their fault. Diligent drivers get hit by drunk/distracted drivers all the time.

13x better than an average that includes drunk, fatigued and distracted drivers would not be very impressive on a closed course.

On a real world course where the only way to achieve those kind of numbers is to avoid getting hit by those drunk, fatigued and distracted drivers? Very impressive.

13x compared to what ? The average driver is such a bullshit statistic - accidents are highly correlated with stuff like alcohol/drugs/lack of sleep/lack of experience/physical issues, then the other huge behavioral factor distraction and driving style, and on top of that car performance matters a lot too. I don't see any attempt to correct for that in their "human benchmark". Heck the least they could have done is compare to human taxi drivers which would be apples to apples. If it's 13x compared to that I'm sold for using it as a taxi service !

But individual driving - you can eliminate all those factors assuming you're a healthy, expericed driver with a new car. Nothing against self driving in principle but the failure cases I've seen look so bizarre - I'm way more comfortable with my limitations.

Is it a bullshit stat though? it's not like you or I can go to a different dimension where all drivers are healthy, fully awake, undistracted, sober, competent, etc.

You don't need to. You just need to be in good condition yourself and actually paying attention. Professional delivery drivers routinely achieve seemingly absurd mileages per incident.

Yes, but I don't actually trust most folks to be in a "good condition". Even the best drivers have their off days.

Well, if we were talking about forcing people to stop driving and transition to current waymos it's plausible that diligent sober drivers would be facing greater risk. Would that be acceptable to improve average statistics?

> if we were talking about forcing people to stop driving and transition to current waymos

We're not

If you're arguing to the regulator that "look these cars are not as bad as humans overall" I guess that's fine, but if you're trying to sell me on using self-driving by comparing to something that's completely unrelated to my use-cases that's just BS.

Unless their message is "if you're drunk turn on self driving" which I could get behind, I sincerely doubt current self driving is better than humans - simply because of the data they chose to compare to. If they were better than professional taxi drivers I'm sure they would tout that data widely.

This perspective makes the statistic even more impressive. Every day driving they interact with probably thousands of people who are impaired in some way and manage to drive safely far more often

No not really - because it doesn't compare to other comparable drivers. You're assuming that human drivers can't drive defensively.

Like a realistic comparison would be comparing it to taxi drivers/ride share drivers - then you'd see the risk vs using a cab/uber/whatever. Not a drunk 18 year old with a clunker with no brakes.

Reply on news.ycombinator.com