Hacker News Reader

Comment by arde | Hacker News Reader

Well, I read that the current system for self-driving cars by Google relies on a lot of data from Google (i.e., its system does not purely work through in-car sensors). So if Google decided to pull the plug on that data, or even if they stopped keeping the data current, we'd be in precisely that scenario.

View on HN

I think you'll find that while Google is good at throwing a lot of money on expensive experiments, Google would be a terrible car manufacturer. Car companies are already doing their own research, and making better user-focused designs than Google's "people are stupid, take away their steering wheels" mindset.

People are stupid, and should have their steering wheels taken away. I trust an algorithm driving way more than a stupid, moody, aggressive human.

Which is an unfortunate position given that human drivers are still vastly safer than any algorithm. The way Google presents its marketing for Self-Driving Cars is meant to downplay that.

Their cars are deathtraps without human drivers.

According to what source? I've not looked into it, but everything I've seen shows that self-driving cars are much safer. Likewise: Self-Driving cars are going to get safer, whereas I see no reason for humans to do so.

According to Google, actually. They're just very deceptive in their marketing. Google reported to the California DMV that across 424,000 miles driven, Google cars would've been in 13 accidents (10 of which would've been their fault) without human drivers interceding. Even if you only count the at-fault accidents, that's an accident every 42,000 miles. Which is a terrible record.

Meanwhile, insurance companies estimate humans average about 250,000 miles between accidents. Factor in unreported incidents and we get a safe estimate that human drivers could be closer to 150,000 miles between accidents.

At the most generous factor I can provide, humans are three times better drivers on average than Self-Driving Cars.

...But wait. During the same 424,000 miles, the cars' automated driving system had a system failure, reverting to the human driver, 272 times. (Roughly every 1,500 miles.) Without a human driver to take the wheel, who knows how many more accidents they could've caused.

The secret to the Self-Driving Car's safety record isn't their software. It's humans. Google's "1.4 million miles driven" isn't a really useful statistic, because it ignores all of the "hard parts" of driving where humans had to take over and save the car. Bear in mind, Google's test drivers are trained, and are attentive and alert at all times to quickly take over control. The human test drivers, therefore, are better than the human average, and that's where your safety figure comes in.

Source: http://static.googleusercontent.com/media/www.google.com/en/...

I do think the potential for self-driving cars is vast. Personally I'd rather not own a car, but subscribe to a fleet of self-driving cars. Then the issue I made a sarcastic comment about would be moot.

This is appealing to me because I have no special attachment or interest in cars other than as a means to get from point A to point B. I understand many people enjoy car ownership and the culture surrounding it, but I just want a tool.

Reply on news.ycombinator.com