Hacker News Reader

Comment by downandout | Hacker News Reader

Well.....what could possibly go wrong there? There have been several air disasters over the years that resulted from a failure to flip the right switch or check certain settings, either in-flight or pre-flight. A Greek crash where the pilot forgot to flip the auto-pressurization switch, killing most on board before the crash itself, comes to mind. This video leaves little wonder why these things happen.

Having spent 18 years trying to design software and websites to be idiot-proof, to me it looks like they could seriously use some human-friendly UI and procedure design. When the stakes are as high as they are during a packed commercial flight, it would seem to me that maximum possible simplicity would be the goal. Perhaps a "flight checklist wizard". It is possible to enable the granular control that pilots need without making them this complex.

View on HN

I don't know. At work, I'm responsible for the regression test. I wrote the regression test. And I have a check list with 52 items required to run it (step 50 is actually running the test, the next item is making sure it's running, and the 52nd item is "walk away for the next five hours"). Could it be simpler? Yes, but it would require even more setup (since it involves generating several datasets that are copied to four machines), making sure configuration files are up to date (less an issue now than when we first started), and could involve restarting named (since one of the datasets is a particular zone file), and then starting up the various programs across all four systems.

And then, the testing environment changes. This time, it's two machines (can you tell this happened?). Not four. At least with doing this huge checklist each time, I know the steps well enough to adapt the procedure (okay, I need to mentally translate machine "foo and bar" to "snafu"). Automated, it would require extensive testing just so we can run our tests ...

The difference is that if you accidentally skip one of these steps, you won't destroy a $50M+ piece of equipment, devastate 100+ families, and/or disable an airport or neighborhood for days while rescuers collect the body parts of all the people you killed. Complexity is the enemy of humans in stressful situations. Where lives and tens or hundreds of millions of dollars are on the line in a zero-fail environment, I would strive for simplicity.

Not saying you're entirely wrong, but there's a difference between incidental vs necessary complexity, and a web UI is not even remotely similar to an avionics dashboard. There might be something we software people could teach those in avionics, but I'd imagine the reverse is actually much more the case.

In fact, ergonomics was started by research during WW2 to reduce airplane accidents caused by human error. There's a reason flying is the safest mode of transportation today, and efforts in ergonomics play a big role. For example, the gears are placed in standardized locations / have standardized shapes to prevent pilots from using the wrong ones while in panic mode.

Some modern airliners have computer-based checklists that will verify that all checklist steps were completed correctly for you. They also have alarms that will go off if the aircraft configuration is egregiously wrong on takeoff (eg, http://www.youtube.com/watch?v=hOqMttyD_Zc). With older or smaller jets, however, some of these safety features might be missing or less sophisticated, so it's more on the pilots to correctly execute the checklist.

Automation sounds great, but it must be followed-up as to how it's used or abused in the real world in order to maintain air safety.

A primary concern of automating error detection is that it leads to laziness by cognition of least resistance. In the real world, pilots will simply rely on the machine to give them feedback rather than knowing what's fundamentally wrong themselves. This is dangerous because mastery goes away and you end up with experienced, routine button-pushers that slap a 777 into a seawall or trim some trees.

The other concern is zero incident complacency. If there were a "chaos monkey"-like system that could continuously introduce fire drill challenges, while being able to reveal themselves as authentic or not, would keep pilots alert and grade reactions before something larger happens. This is an instance where automation by purposeful fault injection might be able to maintain reaction fitness more consistently over time.

What this video doesn't show is the multiple redundant checks that are done, and the warning systems that sound if something is out of place (like the one that the Helios pilots misunderstood). Newer airplanes do have more human friendly designs and more automation, including electronic checklists. For existing cockpits, they are usually only upgraded when serious design flaws are discovered. But that doesn't happen often.

I was also thinking about this. Even if the controls and procedures remain completely unchanged, it should be possible to issue warnings when things were done wrong. In fact I'd be surprised if this wasn't at least partially implemented already.

Next to some planes having electronicaketen checklists; some "older planes" have a somewhat simpler form of digital checks. For example: the 737-800 has a take-off configuration warning (a Horn) When the thrust levers are advanced without having set a proper flap setting for take-off.

(Unfortunately: this Horn is the same Horn as the cabin altitude warning Horn, having a big part in the mentioned greek airline incident)

Some modern airplanes do have integrated electronic checklists.

Reply on news.ycombinator.com