Hacker News Reader

Comment by freejazz | Hacker News Reader

Musk said "unequivocally". Besides, the person you responded to was making a different point. They were making the point that the crashes would likely be those easily avoided by humans. A) I think this is a distinct point from yours, but still answers it anyway as far as it being a refutation that FDS is safer, and it's not pointing to a single anecdote. the poster was calling for ALL footage to be released... B) if your point is to say that there is always a human that could have made such a mistake in the car, what is that supposed to mean? not all drivers should be driving, many are even unlicensed...

View on HN

My point was NOT say that a human could have made that mistake. Obviously a human would not have (except for all the times a distracted human would do exactly this).

My point was that the standard "Release all footage of accidents and look at how many accidents humans could have avoided" is not productive.

The method of illustrating this is to show how this standard would have performed with human drivers, who will generate hundreds of thousands of incidents that other humans could have easily prevented.

Those incidents will have some categorical differences, obviously. But incidents they will be.

If robots make 1 mistake in 10000, and monkey makes 1 mistake in 1000, robot's are unequivocally safer than monkeys.

It's that simple.

The question is how many mistakes robots make, and how many mistakes humans make.

If humans wouldn't have made that mistake... it's certainly not safer than humans in that regard. You are saying that FDS might be safer on the whole. Apparently, it would be because it FDS doesn't make other mistakes that humans do make, if it was safer. Not sure how that supports you saying that having footage of the mistakes that FDS does is not productive. It'd be productive even if FDS WAS actually as safe as humans (I seriously doubt it is, and no doubt the article outright disputes such a notion) because FDS could be made safer. The reality is that you are saying that FDS should not be any safer than a typical driver. While I certainly understand the rationality behind such an argument, I think we should strive for better.

> If humans wouldn't have made that mistake... it's certainly not safer than humans in that regard.

I don't think you understand how this works. If you prevent more accidents than you cause, then it's safer. Period. It DOESN'T MATTER if those accidents you caused would not have been caused by a human. If you cause less accidents than humans on average, then it is safer. And Tesla has shown that data that even among it's own cars it's safer.

We are just pushing back and forth where you measure safety at. The site of the incident? At scale? You are saying at scale. You are saying this is unequivocally true but do not provide any reasoning as to why. It might be safer measured on a cross country trip as opposed to driving in the city. I don't really leave the city that much. So, it might not actually be safer for me. I hope you can genuinely engage with this as opposed to being so purposefully obtuse.

> And Tesla has shown that data that even among it's own cars it's safer.

I'm not going to trust the data from people who designed a system that immediately turns off the second it becomes apparent it would generate an adverse statistic. Get real.

> You are saying at scale.

That's all anyone has been talking about in this thread from my reading. There was no pushing back and forth.

> I'm not going to trust the data from people who designed a system that immediately turns off the second it becomes apparent it would generate an adverse statistic.

You're going to need citations for that. There's no demonstrations I've ever seen showing that.

>That's all anyone has been talking about in this thread from my reading. There was no pushing back and forth.

Except for my post... that you responded? I'd have thought you read that, although I could be mistaken.

>You're going to need citations for that. There's no demonstrations I've ever seen showing that.

https://www.motortrend.com/news/nhtsa-tesla-autopilot-invest...

dang - all I had to do was google "Tesla FDS turns itself off"

That link you provided doesn't show any demonstrations showing it.

So? It's clearly happening, that's what the NHTSA found.

Tesla counts any incidents that occurred within five seconds of FSD/Autopilot disengaging as being "under autopilot". Given the driver is meant to be paying attention and ready to take over at any time 5 seconds is plenty of time to take over (and come to a complete stop from ~70mph)

Oh cool, it'll point your car at a child and give you 5 seconds to fix its own mistake! How generous! What a great system! Fully autonomous driving except for when it makes a mistake...

The burden of proof is on the technologists, requiring them to release footage keeps them honest where human life is at stake.

It's not an unreasonable ask in those circumstances.

All they have to do is statistically have a smaller rate of accidents than human drivers, and his statement would pass. And honestly I wouldn't be surprised if that's true. Doesn't change the fact that people hate the idea of machines accidently killing people more than people accidently killing people, but it's not that crazy a thing to say

Humans get into accidents otherwise avoidable by most other driving humans every day.

Reply on news.ycombinator.com