Hacker News Reader

Comment by ghostbrainalpha | Hacker News Reader

This argument seems so pedantic to me.

Human driving has basically capped out at its potential to improve. It may even be starting to get worse...

Automated driving, has barely begun and even if it is less safe today, it will clearly be much safer than human driving soon. And will only continue to get safer and safer as the technology improves.

Do you we need to be so micro-focused as to have an exact day/model/release where autopilot is safer than human drivers?

I for one have been hit by a drunk driver. My wife was in an accident where the other driver was ACTIVELY playing candy crush. We can all see that automated driving will make the roads a safer, better place even if that takes 20 years.

View on HN

That's the point: I'm pretty sure that somebody will create automated driving technology that's better than humans within the next decade or so. But it won't necessarily be Tesla. If I had to guess I'd say Waymo is likely to get there first. And as I say in the piece, there's good reason to think that Tesla's incremental approach—start with ADAS, evolve into full self-driving—may be an evolutionary dead end because humans are really bad at supervising a driving technology that works 99.99 percent of the time but gets in a fatal crash the other 0.01 percent.

> And as I say in the piece

Are you the author of the article?

Twitter handle is the same as his hackernews handle, so I'd assume it's likely.

Yes.

"humans are really bad at supervising a driving technology that works 99.99 percent of the time but gets in a fatal crash the other 0.01 percent"

I've tried to express the same concept but you said it in far fewer words. Nice.

Good article.

What about Tesla's choice to design its technology around cameras rather than lidar? They seem to be the only ones going that route. I've seen quotes from a number of people who build this kind of technology who say that's a fundamentally less-safe way to build self-driving vehicles. Did you consider addressing that point in the article?

But the counter-argument, to allow autopilot to catch on if it is NOT safer today is to cause deaths, even if over the long run (6 months? 6 years? 60 years?) it will be a net positive.

The stats need to compare a Tesla not to the "average" car, but the average car in that market segment driven by an owner of that socioeconomic class to tell us much of anything. And then we need to separate out when autopilot is active vs when it's merely functioning in a passive sense. Hell, my VW Golf has forward collision warning with automatic emergency braking. Even Corollas these days have laser cruise.

Further investigation required: are humans worse at driving or statistics when loss of life is a factor in each?

So it is quite unreasonable to ask if this system is safer right now, because some future system almost certainly will be?

If Tesla had not prominently promoted the exact claim that this system is safer right now (and by a remarkable margin), that claim would not be under scrutiny. Tesla's statements are now being subjected to an objective, reasoned analysis, even though they were not made in that spirit.

>Human driving has basically capped out at its potential to improve. It may even be starting to get worse...

But we can significantly improve on that. Augmenting human capability is far easier than converting a pile of transistors to perform at the same level as humans. Improving road safety infrastructure is also possible - today, Improving technology in cars to detect when the driver is impaired is also easily possible - today. We don't need pie in the sky "autopilot" technology to improve on the current situation.

>Automated driving, has barely begun and even if it is less safe today, it will clearly be much safer than human driving soon. And will only continue to get safer and safer as the technology improves.

You can say that about anything. One day a bacterium will evolve into a human. Also, its more than reasonable to say, we don't want to get killed because some dude in the valley pushed a firmware update. If the tech sucks, don't endanger other humans. Go fix it, and come back and we'll gladly take a look at it again since it does have value. As of now, people are better off staying away from any Teslas on the roads.

The argument is not pedantic.

>it will clearly be much safer than human driving soon

You cannot say really that. The technology at this current stage is much worst than human drivers when it comes to understanding its environment and detecting hazards.

The interesting question, which this article touches on is the following. Given that we don't have full autonomous driving--and won't depending on the environment for quite a while--what is the appropriate level of assistive driving systems? For example, as this article indicates, automated emergency braking seems to be a fairly obvious win. Adaptive cruise control probably is as well. But there may be issues with putting incrementally better systems into cars until they are really good enough to operate on their own.

It's hard to call it capped out when we could improve a lot by reducing drunk driving and discouraging phone use while driving. Don't you think so?

"Humans are stupid, tech will solve it" attitude is missing the point that humans need to build, maintain and take care of tech. If humans are stupid, tech built by their designs and under their control will be shit too.

Your first point is good. I should have stated that "Hypothetical Best Possible Human Driver" has been established. We can do a lot to improve safety in the average driver. But human reaction times, spacial awareness have hard limits. You can only react so fast, and you can only look at what your two eyes can show you at once. Once the average driver starts to approach the best possible human driver, improvements will be very hard to come by.

But your second point, if humans are dumb, their tech will be dumb is demonstrably false. Groups of idiots can collaborate to make 1 fantastic solution to a problem. Or a solution that improves over time. Software also tends to have fairly stable performance, except for specific outages. Where a human may have the capability to be a safe driver, but performance varies drastically based on mood, age, distraction.

I could definitely be wrong. But I would be comfortable betting every penny I have that automated driving will be safer even the best human drivers in 20 years. No one can know the future, but betting on human power instead of technology has been a losing strategy since the printing press was invented.

Humans can be pretty good if they're trained like jet fighter pilots. I'm sure automated cars will outperform them (or at least be as good as them) if they're built and maintained like jet fighters. If we can't make average human a jet fighter pilot, can we make them maintain their cars at that level? And expect to pay for them accordingly.

On the second point, I'm more talking about hacking scene and maintenance. I'm pretty sure today's jackasses will find a way to hack auto-piloted cars to act like jackasses. And drunktards will drive unmaintained cars with not working and counterfeit sensors too.

One of my early thoughts with some of the driver assist technologies was "oh cool, I can take off my jacket while driving, since the car can pay attention in that short time span." It was at that time that I realized that there was a problem with autonomous driving: there is a gap where it's short of allowing true inattention but is usually safe enough that it is going to train users to be even more inattentive. The natural consequence of this is that autonomous driving, at a certain point, will decrease safety, not increase it.

Waymo seems to realize that this is a problem, and is actively pursuing a development model that doesn't release self-driving cars except on the other side of the unsafe gap. Tesla seems to be trying for a constant incremental improvement model, which means they're going to be putting people into the unsafe gap... and their marketing department is going to be counterproductive.

"Human driving has basically capped out at its potential to improve. It may even be starting to get worse..."

That's not true at all. If you've driven in other states and especially other countries, you know there is a lot of variations on average driving abilities. California drivers are among the worst and German drivers are among the best I've personally encountered. It's not that CA drivers are dipshits but if you look around enough as a pedestrian or while riding a bus passing other cars, you'll notice MANY of them are on their phones. This isn't humans reaching our cap but a recent degradation that can be corrected.

No, automated driving won't be doing that. It will never make your car safer than human drivers. You think every human driver is out there playing Candy Crush?

And there are several car models that have ZERO fatalities. How is Tesla supposed to exceed that safety standard?

Right now Tesla is being criminal in comparing their safety rates to the average vehicle. That includes comparisons to motorcycles, which have death rates that are 50x-100x more than luxury passenger cars.

If Tesla is comparing death rates of their luxury vehicles to motorcycles, they have already failed.

And you know whats safer than cars? NOT cars. As in, mass transit.

What car models have zero fatalities?

At the bottom of this page: http://www.iihs.org/iihs/news/desktopnews/on-the-road-again-...

Reply on news.ycombinator.com