Hacker News Reader

Comment by hawkharris | Hacker News Reader

In the 1950s, the rate of commercial plane crashes caused by human error was 71 percent. By the 1990s, with the introduction of new automated flight systems, the rate had decreased to 52 percent. [0]

Yes, there are horror stories about "dumb" transportation software that linger in the forefront of our memories. But ask yourself, What is the burden of proof for this technology? Do self-driving cars and aircraft automation systems have to work 100 percent of the time before we accept their usefulness?

If that's the burden of proof, we will remain stuck with human drivers who are much more "dumb," by a measure of the car accidents and plane crashes they cause.

The author's argument against automation in these life-saving contexts is weak. His article would have been stronger if it stuck with points about watches and similar consumer products.

[0] http://www.statisticbrain.com/airplane-crash-statistics/

View on HN

The stastic you quoted doesn't actually tell us there are less crashes due to human error, though. We could have just as many crashes due to human error, plus more automation-related ones, and that'd still reduce the percentage of crashes that are due to human errorm

"The author's argument against automation in these life-saving contexts is weak."

I think you missed the point of the article. Its not that existence of automation is bad, its the implementation of it that (in a lot of cases) is wanting, useless, distracting, or even detrimental to the items purpose.

"The notion that we can have automated or semi-automated cars as long as the driver is watching over them is a dangerous myth."

If we accept the author's argument that a human driver can't watch over an automated car, the question becomes fairly black-and-white: do we rule out the driver or rule out the automated vehicle?

Within this rigid self-imposed framework, the author seems to choose Option B: ruling out vehicle automation. Throughout this post (in the aircraft analogy, for instance), he notes the importance of human supervision.

To this, I say, Where is the evidence that human-supervised vehicles are safer than fully automated cars? Where is the evidence that humans are better equipped than good software to react to split-second emergencies?

The only point in favor of human drivers is that we have more stats to describe them, but the stats are horrific. By contrast, Google's autonomous cars have logged over 300,000 miles in various conditions without an accident (minus getting rear-ended by a human driver).

The current fatal crash rate for human drivers is about 1 per 100,000,000 miles, and the injury crash rate is around 1 per 1,200,000 miles. 300,000 miles with 1 accident doesn't prove much.

Where did you find that statistic?

Best presentation, but just for one state at a time. Utah is typical: http://s3.amazonaws.com/zanran_storage/des.utah.gov/ContentP...

National data, but only for fatal accidents: https://www.census.gov/compendia/statab/cats/transportation/... Articles 1103 and 1106.

More detailed source of national fatality data: http://www-fars.nhtsa.dot.gov/Main/index.aspx

That's not what the author is saying. Their saying it's bad that the automated system didn't warn the pilot when something went wrong. Or that self-driving cars can't reasonably depend on humans to monitor them at all times. He's not saying that all automation is inherently bad.

Reply on news.ycombinator.com