Slashdot Mirror
Airline Pilots Rely Too Much On Automation, Says Safety Panel
Hugh Pickens DOT Com writes "Nearly all people connected to the aviation industry agree that automation has helped to dramatically improve airline safety over the past 30 years but Tom Costello reports at NBC News that according to a new Federal Aviation Administration report commercial airline pilots rely too much on automation in the cockpit and are losing basic flying skills. Relying too heavily on computer-driven flight decks now poses the biggest threats to airliner safety world-wide, the study concluded. The results can range from degraded manual-flying skills to poor decision-making to possible erosion of confidence among some aviators when automation abruptly malfunctions or disconnects during an emergency. 'Pilots sometimes rely too much on automated systems,' says the report adding that some pilots 'lack sufficient or in-depth knowledge and skills' to properly control their plane's trajectory. Basic piloting errors are thought to have contributed to the crash of an Air France Airbus A330 plane over the Atlantic in 2009, which killed all 228 aboard, as well as a commuter plane crash in Buffalo, NY, that same year. Tom Casey, a retired airline pilot who flew the giant Boeing 777, said he once kept track of how rarely he had to touch the controls on an auto-pilot flight from New York to London. From takeoff to landing, he said he only had to touch the controls seven times. 'There were seven moments when I actually touched the airplane — and the plane flew beautifully,' he said. 'Now that is being in command of a system, of wonderful computers that do a great job — but that isn't flying.' Real flying is exemplified by Capt. Chesley Sullenberger, says Casey, who famously landed his US Airways plane without engines on the Hudson River and saved all the passengers in what came to be known as the 'Miracle on the Hudson.' The new report calls for more manual flying by pilots — in the cockpit and in simulations. The FAA says the agency and industry representatives will work on next steps to make training programs stronger in the interest of safety."
The obvious solution is self-flying planes! Then there won't be a pilot to rely too much on automation.
It depends on the simulation. If you are training for a cross oceanic flight, you would simulate switching out flight crews and long periods where you would normally use auto pilot. The simulation would toss various problems at you to break up what is normally a dry, boring routine so you know how to handle different problems.
Personally, I think we're just a few years away from a fully automatic flying experience.
Only the dead have seen the end of War. - Plato
If you read the summary, you'll notice one of the big problems is when that automation fails. It's great when it removes human error, but if automation fails, you still want human error as minimal as possible... and that means teaching pilots to rely less on automation (which is a very different thing from using less automation).
Automation fails from time to time, and when it does, pilots are the failsafe. But to be able to do that, they need to stay in practice, and that's the problem being highlighted here: they're getting so little time in control that they're getting out of shape.
When Capt Sullenberger landed on the Hudson, the aircraft software worked to prevent his stall. But his flying skill is what safely landed the plane. His knowledge of what the aircraft can and cannot do was critical. He even realized he needed the APU for the computers to continue operating, and turned it on early in the emergency. His actions showed that he understood his plane and how to fly it. Some pilots are forgetting the "fly it" part.
For example, the situation many years ago in Iowa where the hydraulics failed and the pilot had to steer the plane using only the engine throttles is an example of something that no computer system is designed to do. Yet a veteran pilot managed to pull it off.
That veteran pilot was a passenger that just happened to be on the plane. 99% of pilots would not have been able to pull it off. So what should we do about that?
Option 1: Train 100,000 pilots on a difficult technique that they will likely never use.
Option 2: Have one programmer sit down with that veteran pilot and code up the technique, test it on a simulator, test it on a real plane, and then use a USB thumb drive to update all flight control software on every plane.
automation is not designed to handle extreme failures of the aircraft.
It should be. One of the lessons of TMI was that automating routine stuff only leads to disaster because operators lose the skills they need to handle emergencies. The "extreme failures" are the first thing that should be automated, because those are the events that pilots are least capable of handling properly. ABS in cars is a good example of this. Nobody needs ABS to slow down for a routine traffic light. But ABSes have saved many lives when drivers slammed on the brakes to avoid a collision, or started slipping on ice.
A human can get an appreciation of velocity even without working pitot tubes, in a middle of a weather system where GPS doesn't work.
Sorry buddy, you've just killed yourself in exactly the same way the AF pilots killed themselves. Oh the irony. NEXT STUDENT, PLEASE.
Seriously. Your seat-of-the-pants "feel" for a modern jet is precisely what is going to kill you. So let me be clear: if you ever end up as an untrained babbling idiot in a cockpit of a jetliner, trying to save a bunch of souls while the air data is missing, you better keep it straight and level and not mess with anything until you've read the checklists. After you do, and you better be quick about it, you'll know that what you're supposed to do is to set the throttles to a fixed position that depends solely on altitude and desired rate of climb/descent. You'll look those up in a fucking table, and as long as you do, you have a chance to make it. There's going full retard, and it's you.
A successful API design takes a mixture of software design and pedagogy.
If it were so easy to just automate extreme failures, websites like Google, Facebook, and Amazon would go down a lot less often. Unfortunately despite thousands of employees with extreme technical skill, there are still mistakes that bring them down from time-to-time. If we didn't have human SREs or System Administrators, things would be a lot worse. A computer doesn't have the analytics skill of a pilot and never will unless we end up with a singularity.
We don't have strong AI yet and pilots will never just "sit down with a programmer". Automation has to be tested thousands of times across thousands of scenarios in different aircraft and conditions for decades. Even then, there's always the chance that some snippet of code is waiting to kill a plane full of people because it got the wrong set of sensor inputs.
What I meant by "it's not about staying in practice" was that the issue with automation has to do with things that happen on a much smaller time scale. It didn't mean that staying in practice is irrelevant. Is is relevant, but you need more than that. Practice demonstrably isn't sufficient in itself.
The only difference between a plane and a car is that the time scales may happen to be two or three orders of magnitude different, if you're lucky. Autopilot disconnect-related CFIT is a classical example of what I'm talking about. By the time the pilot figures out that his idea of what's going on (we're in a safe, controlled flight on autopilot) differs from reality (CFIT-in-progress), it's too late, or there's sufficient panic that has set in that the control responses are not what you've been trained for either (stall recovery, spin recovery, etc.).
It doesn't matter that the pilot has more time to figure it out. They are unaware of their own mental model's divergence from reality. In spite of having been given all that time, they still CFIT because they think they're on autopilot. You'd think this would be pretty obvious, but there's one insidious thing. If you're unaware of it, it will eventually kill you. Your brain's sampling of the state of the environment is highly dependent on how confident one is in their own model's accuracy. If things "feel" like everything is the way you imagine it should be, you'll be tricked by your own brain into "seeing" made-up instrument readings, your sensitivity to increased wind noise will be diminished, etc. I'm dead serious. It takes awareness of this pitfall to be able to force oneself to see how it really is, to make your brain not trick you. When you don't, and you're a pilot, usually a couple hundred people perish with you. This is happening over and over, it's sickening. The reason why it happens with such regularity is that we're dealing with a basic property of our brain's visual interaction with the environment. It's not widely appreciated in nonspecialist circles, unfortunately. We're almost all "broken" like that.
A successful API design takes a mixture of software design and pedagogy.
Planes receive a unique flightplan and detailed instructions for take of and landing that are steered by a central traffic control to make sure that there won't be any other planes nearby. Thats possible because EVERY plane has to receive instructions from them.
I'll stop you there because you've just shown you don't know what you are talking about. Even if we just limit the discussion to large commercial aircraft, your claim that each receives a "unique flightplan" is ridiculous. The initial parts of the flightplan (departure) are so UNunique that they print them in books and give them names. A common departure clearance would be something like "United 123 is cleared to [destination], Farmington 3 departure, SHADO (an intersection somewhere on the filed flight plan), then as filed, maintain 3 thousand, departure frequency 123.45". Pretty much every aircraft going the same direction gets the same thing.
When the aircraft gets close to the arrival airport, it will get yet another UNunique approach, by name. "United 123 cross BILBO at 5 thousand, cleared for the ILS 14 right approach". That ILS approach will start at some initial approach fix (maybe BILBO, maybe after) and then bring every aircraft on that approach through the same course. The goal of the approach controller is to get them all lined up at a nice, regular spacing all coming down the same ILS with sufficient spacing that as soon as the preceeding one clears the runway the next one is about to land.
An important thing to know about the system is that even with a filed flight plan and a clearance "as filed", the flight plan does not specify the approach procedure. That bit of critical info isn't known until close to arrival. Usually the last approach controller will tell the pilot "expect the ILS 21" or whatever. The automated weather system may contain that planning information, too, but the pilot is free to ask for something else if he wants it, and he isn't cleared to fly that approach until the words "cleared for ..." come out of the controller's mouth. If communications is lost enroute, the rule is that the pilot can fly any appropriate approach procedure.
The second bit of foo is "steered by a central traffic control". The pilot steers the plane. ATC issues clearances and gives instructions, but the pilot steers. And "center", despite its name, it not a "central control". There are a lot of them, and each "center" (New York Center, for example) is split up into sectors. Since we're currently limiting our context to large commercial passenger aircraft, yes, there will almost certainly be a "center" involved in the flight, but they take over only after the aircraft has gone through the departure controller at the airport, and will hand the flight off to the approach controller for the destination airport (for airports large enough to have their own). For destinations that aren't large enough to have their own approach, or their own control tower, this "central control" will actually cut the aircraft loose to talk on the CTAF (common traffic advisory frequency), so this IFR aircraft on a "unique" flight plan will now have to mix in with all the VFR traffic at the airport, even the student who is up in the pattern practicing landings. See and avoid.
These ATC folk don't make sure there aren't any other planes nearby. Only for IFR traffic (which anything above 10000 feet must be in the US) do they provide traffic separation. They will issue instructions to keep two IFR aircraft apart, but the vertical spacing can be just 1000'. In airspace where VFR flight is permitted, and outside ATC control, it is quite possible for another aircraft to be "nearby" and less than 1000'.
And the final nail? "Thats possible because EVERY plane has to receive instructions from them." It's severe clear outside here this morning. I could drive to the airport and fly off to someplace else where there is commercial service, and the only time I'll have to talk to ATC is when I'm within 5 mi