Hacker News Reader

Comment by rob74 | Hacker News Reader

Do you mean the 0.07 "fatality rate" for Boeing 737-600/700/800/900 vs. the 0.09 for the A320 family? I would say they are both very safe aircraft, but I wouldn't read too much into the 0.02 difference, because that includes accidents like https://en.wikipedia.org/wiki/Germanwings_Flight_9525 (where the copilot intentionally crashed the plane) or https://en.wikipedia.org/wiki/Helios_Airways_Flight_522 (which was 100% due to human error)

View on HN

> which was 100% due to human error

Not sure how familiar you are with the airplane industry but nothing is ever "100% due to human error" in these events. We know humans are controlling these planes and the equipment can always gets better. Usually it comes down to UX, checklists and more to have multiple stopgaps to prevent issues. Just a quick scan of the Flight 522 page shows there were in fact many UX issues like why can you take off when the pressure system is turned to manual? Forcing a manual override if so would have helped. Same alert sound was also used for "take-off configuration warning" and "cabin altitude warning", leading to the pilots making the wrong decision. Attributing Flight 522 to only human error is not what the courts did (check the "Lawsuits and criminal proceedings" section in the page you linked) so not sure why you would.

Ok, if you want to get into the details of this particular accident... it can be argued that UX issues made it easier for the pilots to make the fatal mistake they made (and the lawyers representing the victims' families of course made that argument, because going after Boeing is far more promising than going after a small airline which declared bankruptcy soon after the accident), but in the end it was still human error. Aircraft are complex systems which can't be made 100% foolproof, that's why they are flown by qualified and trained pilots who have checklists for things that have to be done before take-off, and one of them was checking the pressurization system ("the flight crew overlooked the pressurization system state on three occasions: during the pre-flight procedure, the after-start check, and the after take-off check").

...and the courts ultimately had nothing to do with it, because the suit against Boeing was settled out of court. Which is not an "admission of guilt" on the part of Boeing, just "give them some money to make this unpleasant thing go away".

I appreciate your point about nothing ever being "100% due to human error", especially in regards to air travel. I recently read "The Design of Everyday Things", a book that devotes a lot of its time to discussing this same principle. We should blame the UX wherever possible before assigning any blame to the individual.

I'm much more concerned by the operator than the aircraft on proven planes. Air France for example -- landing in Syria in a civil war due to a backup plan that prioritised money over safety [0], pilots ignoring stall warning for nearly a minute [1].

The Max does concern me about Boeing planes goign forward, but existing airbus and boeing planes are fine.

[0] https://www.independent.ie/life/travel/travel-news/air-franc...

[1] https://en.wikipedia.org/wiki/Air_France_Flight_447

> because that includes accidents like https://en.wikipedia.org/wiki/Germanwings_Flight_9525

> He had set the autopilot to descend to 100 ft (30 m) and accelerated the speed of the descending aircraft several times thereafter.

That could have been avoided if the autopilot refused to crash the air plane. To be fair, I bet Boeing's autopilot also accepts commands to crash the plane. However this accident could have been avoided if the autopilot had refused to comply with an obviously deadly commands.

You're onto something, keep on thinking about it...

https://en.wikipedia.org/wiki/Lufthansa_Flight_2904

Please excuse the snark, but there is more to flight safety than a Hacker's "AI's the best"-mentality. We actually have a lot of experience gathered from accidents and non-accidents how these systems should behave in a general sense, and "overriding pilot-input" has proven not to be a good idea.

The discussion above was about details and UX.

I'm skeptical that an AI for airplanes exists that can distinguish between "obviously deadly commands" and "trying to rescue the plane during a mechanical emergency".

A hypothetical system would actually function almost exactly the same as the much derided MCAS. If angle of attack sensors indicate an inappropriate angle of ascent/descent, overcompensate for or override the pilot inputs.

As we've learned at great cost, angle of attack sensors are wrong more often than pilots are homicidal though, which is part of the reason why the form of intelligence used to deter and attempt to prevent obviously deadly pilot actions is a lump of meat called a copilot.

> a lump of meat called a copilot.

Looks like that's how they fixed the issue.

> aviation authorities in some countries implemented new regulations that require two authorized personnel to be present in the cockpit of large passenger aircraft

https://en.wikipedia.org/wiki/Germanwings_Flight_9525#Regula...

An A320 already does much more crash-prevention than a 737 because of its FBW system. It’s impossible to stop someone who’s determined to crash the plane unless you make it fully autonomous.

Reply on news.ycombinator.com