> When you look at airplane crashes, it’s not equipment failure. It’s human error.
A lot of the crashes are due to equipment failure. Take a look at the series "Aviation Disasters". A pilot is needed to figure out what is wrong, and to compensate to bring the airplane back. A computer program can't do that.
Such is far, far more complex and nuanced than dealing with a failure on a car. Just take Sully's landing on the Hudson after a flock of geese took out his engines. What computer program could have figured out why the engines failed, and what was possible for the airplane to do? Evaluate the weather conditions, and make the best choice of landing spot? Phooey.
Well, that's more or less how it works today. Airline pilot's job is basically hours of boredom punctuated by minutes of terror. Computers do most of the work, pilots are there for takeoffs and landings (which could be automated) but more importantly for when things really go wrong. Airline pilots flying scheduled flights (Part 121) are in constant communication with the dispatcher on the ground. Dispatcher is trained to pretty much the same standard as pilots and helps with the planning, safety and can help with workload during flight. Having all three in place is probably why airline flying is so safe these days.
On the other hand, I could see freighters (e.g. likes of FedEx fleet) fully automated in the near future (5 years? less?)
Here is a less radical question: if the plane was capable of flying itself, do two pilots have to be in the plane? Could we get away with just one pilot and still be at sufficient redundancy?
And if the pilot had nothing to do, how the heck would they not go mad from boredom?
Even if computers never save the passengers in these emergencies, as long as they do save them from more pilot error emergencies then it's worth it. Sure the Hudson river passengers may have died, but at the same time the 9/11, Air France, Malaysia Airlines, etc ones may have lived. Hell, you could even have no improvement in safety anywhere and it might still be a good idea. Flying is already so safe that there's no harm in doubling the fatality rate. New technology doesn't always have to improve everything to be worthwhile.
This is more a political problem - customer fears, unions, regulations, etc.
As a pilot who loves flying, I am sad to say I must agree with the article. Computers are ideally suited to fly planes, statistically will be safer and we like it or not, will be flying most large aircrafts in a future not to distant. (My guess.. 20 years tops)
Air travel is already safe. Statistics show 1.27 deaths per 100,000 flight hours, or another stat in a single year you have the odds of dying as 1 in 4 million or so. You are more likely to die driving to the airport. How would having no pilots decrease those odds significantly? I would think that any decrease due to a computer doing it better would be offset by bugs in the software or unanticipated situations.
You don't need me to agree but I do agree. Plane travel is immensely safer than car travel. The only reason plane crashes make the news more than car crashes is because of the rarity and the massive casualties (sometimes).
would be offset by bugs in the software or unanticipated situations
Indeed, it's always the "unanticipated situations" that cause problems both for humans and computers - and in the case of flying, I think it's better to have some level of automation to reduce human error, but there must always be a manual override. Completely replacing the pilot with innumerable lines of code executed by a computer would be even more dangerous as it's far more likely that a pilot can react intelligently to assess the situation and take manual control of the plane than someone be able to immediately find the bug and fix it, without introducing any new ones, then update the software "on-the-fly" (literally).
It's similar to the issues that abstractions cause in software: the more we rely on them to hide details, and the more complex the systems we build as a result, the harder it is to reason about the problems that arise due to something lower down in the stack. "Out of sight, out of mind."
I do not know how I feel about this. I realise that a modern day plane basically flies itself once it reaches cruise altitude what is referred to as fly-by-wire. However, the notion that pilots are not needed or do nothing on planes is ridiculous.
Are we forgetting that it was a pilot in 2009 on US Airways Flight 1549 that safely landed the plane after striking a flock of geese during its initial climbout only to be ditched into the Hudson River shortly after. If a pilot was not in the plane at the time, no computer would have been able to save the plane.
Typically a pilot will turn off the autopilot around 1000 feet and then do the rest for descent and subsequent landing. On takeoff, most pilots will fly to around 10,000 feet, some to cruise altitude before turning on the autopilot. There are some tasks on modern day planes that cannot be currently done without a pilot (at least not safely).
We have to remember things like autopilot still require manual input from the pilot and cockpit crew to get from its departure point to its destination. A pilot doesn't just fly a plane out of an airport, press a couple of buttons and then do nothing for the rest of the flight, this is a misconception perpetrated by the media and so-called experts who have never flown a plane in their life. Autopilot is merely a tool, much like a machine keeping a patient breathing and their blood flowing in the operating theatre is a tool for surgeons. The machines need to be calibrated, have their data input and be closely monitored much like a pilot does on a plane, even with autopilot engaged.
While I applaud the forward thinking from people like Steve Jurvetson, statements like "Planes are mostly already flown on autopilot already" would be strongly disagreed with any pilot you ask. No two flights are the same and all flights are made up of individual and manual decisions from the pilot and crew. If a pilotless plane were something that could be a reality, the airline industry which has been struggling for years now to keep costs down amidst rising fuel costs would have already partially integrated this or started to cut down on the number of pilots they hire.
I would like to see a pilotless plane decide what to do when the landing gear fails and an emergency landing needs to be made. I would like to see what happens when a tire blows on a high-speed take off and the plane has to make an emergency landing, or what happens when a plane has to make an emergency detour over rocky terrain.
You would still need people to operate these planes remotely, what happens when contact is lost with a plane and it finds itself flying without a pilot at 35,000 feet who can manually take control of the plane and land it? Is Steve forgetting that a pilot would effectively be replaced with someone in a call centre type building sitting at a computer terminal remotely programming commands and programming things like runway codes and other things into the plane?
Some things just can't and should never be replaced. Is modern day technology meant to replace doctors? Every time a new device intended to help keep a patient alive is created, is its intent to aide or replace a doctor? Of course not. People would never agree with replacing doctors with machines completely.
Looking to eliminate drivers on the ground is one thing, because ground transportation is far more dangerous than air transportation. I agree with driverless cars because we already have the data and vehicle trips thanks to mapping data and technology mean they are predictable to a certain degree. Air travel is the safest kind of travel there is, fatalities and accidents have been consistently going down year after year, the number of accidents with fatalities in the last 10 years on Western commercial airlines can be counted with both hands.
Replacing pilots in commercial aircraft is something that will not happen in my lifetime, or perhaps ever. It is one of the worst ideas I have ever heard. Not everything can or should be automated for the sake of auto...
We should also remember that it was pilot error that doomed Air France flight 447. That being said, remotely piloting a plane doesn't give the feedback that sitting in the captain's seat does and although planes may not need pilots to take off and land in the near future, a pilot (or two) will still be on board if not to at least reassure the passengers but also in the case of uncertainties that software may not be able to account for.
Edit: Can anyone tell me about gyroscopes in the plane and why there weren't any or did they cease function? Would they have helped if they somehow gave indications that the plane was in fact not stalling? From what the data showed, the air speed sensors (pitot tubes) clogged with ice and triggered a stall alarm so the pilot reacted by pulling up but since the plane was actually level but the alarm reported a stall due to lack of airspeed reardings from the sensors it made the situation worse [1].
In the Air France 447 case, the pilot error came after the automation failed due to unreliable / wrong airspeed indication. This is exactly the scenario why we need a human in the control loop: to handle unexpected and unpredictable failures. Unfortunately the pilots of Air France 447 had not been able to handle the situation either.
His point is that while the pilots failed, they were only called upon because the autopilot had already failed. In other words, Air France 447 is not evidence that autopilot is the superior choice.
I don't disagree there was a pilot error. However, this error came after automation failed.
And I wouldn't call pilots incompetent. I am an instrument rated pilot myself and flying on instruments in a lousy weather at night is hard by itself. Add an unexpected and unexplained failure on top of it and you have a situation that can simply "overload" your brain. There are good lessons from this tragedy and Air France adjusted its training program almost immediately after the crash details have been uncovered.
I don't think it is that the pilots did or did not realize the stall. The stall alarm went off 75 times. The problem was they were not aware of their pitch for some reason. They must have thought they were heading towards the ocean due to the stall alarm combined with seeing very low air speed reports from the sensors when in fact it was the opposite as the airspeed was fine but the junior pilot kept pulling back on the stick which induced and worsened the ascending stall.
This is why I was asking about gyroscopes in the planes in enough locations to accurately tell a pilot "hey, I know your airspeed reads very low and the stall alarm is going off but you're not stalling. Lock the throttles and hold the plane level until the airspeed returns."
It is called "attitude indicator" and there are multiple AIs installed on larger planes. The problem is that a pilot need to determine what exactly failed: either airspeed indicator, or horoscopes, or both, or something else entirely.
Ish123, I can't reply to your comment but you're exactly right[1] and I can't help but wonder if pilots are trained this way, someone like me (I have never flown any kind of plane my entire life) and you came to this conclusion then why did they react the way they did? The whole situation just seems so simply avoidable. Then again, I'm not flying a huge plane with hundreds of people on it in darkness with pounding thunder/rain/lightning and all sorts of gizmos blinking and making noise at me.
It almost seems like something else should be added to the training curriculum and routinely tested that hardens pilots nerves and prepares them for some of the most stressful situations they could ever encounter such as before grunts go off to combat.
Handling emergencies is a big part of initial and recurring training for pilots. I can't speculate "why" pilots didn't handle the situation the right way but based on the Air France response to this crash I believe the main cause is too much reliance on automation and lack of actual hand-flying. Pilots found themselves way out of their comfort zone and it impacted their ability to make good decisions.
If you are interested to learn about "too much automation" problem there is a very good (and funny) talk given 15+ years ago:
Uhh Fly-by-wire has nothing to do with autopilot. Fly-by-wire is just that the controls are dealt with in software rather than physical connections to the controls.
He doesn't have a strong argument. Perhaps he knows more than he's saying but most of those tasks could be done in advance or by remote (not time critical, like navigating around storms), or by suitably trained flight attendants (medical emergencies and miscellaneous busy work he alludes to).
If 1% of landings are on autopilot, why not all of them? He doesn't answer this except "setting it up is hard". But setup can be done remotely, if not in advance, by someone who's also controlling several other aircraft, so there are efficiency gains.
We have had driverless trains for decades - see London's DLR. The ticket inspectors can ask the passenger sitting at the front to please change seats so they can unlock the control panel and handle abnormal situations.
Yet we are only still trialling driverless trucks on mine sites. Even with the sites modelled entirely (and frequently), and with LiDAR, radar and cameras on the trucks, it is still far from perfect. I would argue that an aircraft is in an even less controlled environment (you can stop the trucks when it all goes wrong or comms fails!), which is why it's such a hard problem.
Trains, relative to trucks and planes, are literally on rails. This removes one of the greatest challenges (navigation).
Rail systems aren't really comparable because, well, they're on rails. You have five more degrees of freedom when flying an aircraft.
A train driver's main job is to have good reflexes. Aside from opening the doors in case of fire, I can't think of a railway accident that couldn't have been prevented by simply stopping the train(s) in time. A driverless train simply needs an emergency stop button and a way for passengers to get out in an emergency.
In that particular case there were faulty brakes on one of the cars. Had they been working (or the fault communicated properly) and the train not been overloaded, the derailment would likely not have happened. There was no other alternative - if no braking had happened, the disaster would almost certainly have been worse.
In a computerised system, the brakes would most likely have fault reporting. Actually in this case they did - the logs indicated that the dynamic braking system wasn't producing any current. This should have been indicated to the driver.
There's also plenty of human error there in not reporting the brake failure and 'eyeballing' weights. Passenger trains are presumably designed such that even with severe overcrowding and lots of luggage there is small chance of the brakes being underpowered. I have no idea why an emergency brake system wouldn't activate all possible means to slow things down.
I agree that in this instance braking actually worsened the problem, but the solution was still that the train should have been braked properly. A computerised system would likely have performed similar steps although it would know the braking force provided by each technique and respond accordingly.
Why do you believe we can train automated cars to deal with emergency situations or adverse conditions, but not airplane autopilots? What, fundamentally, prevents innovations in processing power and automation systems from being able to do this?
Your listing of current problems is fairly accurate as to the road blocks, however, your insistence that technology can't overcome them seems knee-jerk and unreasoned.
> Replacing pilots in commercial aircraft is something that will not happen in my lifetime, or perhaps ever. It is one of the worst ideas I have ever heard. Not everything can or should be automated for the sake of automating them.
This in particular is obviously hyperbole (at least, being charitable), since the obvious case for computers taking over from pilots is that they solve several issues: their reaction times are considerably faster, they're considerably better at reasoning about the physics of a situation, they learn from all situations encountered by any artificial pilot (or actual pilot) we have records of, they have greatly larger numbers of training hours, and they don't suffer fatigue.
Arguing that computers will never take over for pilots or that it's a bad idea requires that you raise an argument about the limits of computers that is in the /future/.
As an aside, your comment about doctors is also wrong: I'd trust Watson to do diagnosing over most doctors I've met, and only expect my comfort with robotic doctors to rise as technology improves. I expect that the next major improvement we'll see is fully robotic surgeons.
Again, your post reads as partly just unreasoned, anti-technology ranting, but I feel I should give you a chance to explain those points.
I usually agree with most of Steve Jurvetson's writings, but I've got a few doubts about this one.
Factually, he's largely correct. For example, A F/A-18F Super Hornet can be flown between waypoints and use Terrain Following Radar coupled to the autopilot. The jet can also fly a carrier approach "coupled up" on autopilot.
Could a datalink control the mission computer on a F/A-18F? Quite possible. Could a X-47B fly autonomously onto carrier deck? It's been done.
Could a F/A-18 WSO(eg. non-pilot) program the mission computer and land the jet without manual flying? If the runway is equipped for ACLS (Not ILS) approaches, then yes.
Could a flight attendant program a 737-700 FMS and fly a coupled ILS landing? Probably.
Could the X-47B mission systems be integrated on a 737-700 and fly autonomously? Yes. Its relatively rare for a tower controller to wave-off a 737 because the pilot didn't fly a proper and stabilized approach. Its up to the flight crew to fly a go-around if they are not stabilized on approach. Are they going to start putting LSO (Landing Signal Officers) at JFK to wave off an automated airliner? Maybe...
If an X-47B looses its engine the chance of the jet ditching intact is improbable. If an F/A-18 has a dual engine failure its almost certain the crew will eject. If an A320 looses both engines due to a birdstrike, thats when Capt. Sully earned his paycheck and then some. An autonomous flight computer cant factor the rare emergencies.
There is probably large public support for getting rid of the TSA and having remote override for the FMS might work. Taking the pilot out of the cockpit isn't the solution to the TSA problem.
Small correction : fly-by-wire doesn't refer to autopilots, but to having control surfaces be hydraulically driven and controlled by electrical signals. Previously, control surfaces were mechanically connected to the pilot's controls.
> Typically a pilot will turn off the autopilot around 1000 feet and then do the rest for descent and subsequent landing.
This hasn't been true anymore for about half a decade. Most landings are executed by autopilots. It depends on the airfield and weather and ...
A famous example was an event in 2006 for London Heathrow, when there was a period of about 3 weeks where the airport was autoland-only (has happened several times since then too). No human was allowed to land a plane there, but autopilots were. It resulted in two things, firstly a capacity reduction of about 40%, and a large amount of pressure on airlines to incorporate autoland features into their planes.
> Some things just can't and should never be replaced. Is modern day technology meant to replace doctors?
Yes, definitely. AI algorithms result in better patient outcomes.
> Every time a new device intended to help keep a patient alive is created, is its intent to aide or replace a doctor? Of course not. People would never agree with replacing doctors with machines completely.
Completely, perhaps not. We'll see. I don't think we'll really get the choice though.
Not yet, but a decade or two out, maybe. It requires much better sensors. For one thing, you need all the gear for automatic driving to get from the runway to the gate. Landing on an airport without ground aids may be necessary in an emergency.
What we'll probably see first is larger one-pilot aircraft, able to land themselves in case of pilot failure. Also, unpiloted fighter jets are almost inevitable, simply because modern airframes can pull more Gs than pilots can stand. Boeing is modifying old F-16 aircraft into the QF-16, which is unmanned. These were originally intended for target practice, but DARPA is interested in using them
for other purposes.
The QF-16s are unmanned but they are not computer-controlled. I believe the argument here is for computer-controlled airliners not remotely piloted ones. After what happened to one of our Scan Eagle UAVs[1], the last thing we need is the tragic loss of life and PR nightmare of an unmanned commercial airliner getting hacked and crashed.
QF-16 is "remote piloted" not unpiloted. I don't think the QF-16 has more than basic autopilot (hold heading and altitude) integrated. This is partly due to the QF-16As being Block 15 jets.
I focused on scenarios for removing all security checkpoints and delays from a customer perspective.
Imagine checking in at a kiosk to get your boarding pass, and going though no security lines to board the plane. Bring anything you want with you, but know that the flight is under video surveillance, like retail stores today.
I assumed technologies that work in rudimentary form today and will benefit from Moore’s Law. (The only 12-year forecast that I felt confident about is the continuation of the 100-year abstraction of Moore’s Law, bringing a 256x computational advance by 2018).
So, I started with the assumption that computer-controlled flight would be possible. It’s a pretty safe assumption given what we already have today.
With no cockpit, everything changes. The potential for harm is greatly reduced if the plane cannot be navigated from within. No hijacking. No use of the plane as a weapon.
Bombs become the only threat, and a reduced one.
Personal weapons? A gun or knife-fight could do more damage in a restaurant, or many large group gatherings. Why bother with a plane where criminal activity will be recorded, and the only people harmed are on board?
As for bombs, passive sniffers in a free flowing airport gateway are more plausible than detecting improvised weapons than could be used against a pilot.
At the airport, a quick fingerprint biometric would be a natural way to get a boarding pass (as 12 million people have already done in Florida to get access to an amusement park). So even smuggled bombs would have more capture and downside risk for a terrorist cell than other targets.
Pie-in-sky ideas: hardening a UAV to bombs should be easier than current planes; smaller planes could lower risk; luggage could fly separately; biologic weapon sensors could trigger a flight path to quarantine, etc.
I've always wondered why we don't have more sophisticated autopilots for trains. I recognize the importance of having staff on a train to deal with a lot of the engineering, break, track and other equipment issues, but It's unclear to me why the actual driving (braking/acceleration) of the trains is still done by (often bored out of their skulls) train operators.
I'm guessing the big element that was missing was sensing track obstacles and braking accordingly - in Singapore/Dubai they have doors limiting access to the track, and in vancouver they have (tending to failure in the winter) track sensors to see if anyone has hopped onto the track.
(Automation systems typically tell the pilot where to go with a magenta line, on the G1000 systems I am used to it's an indicator of GPS navigation)
From the opening line: "as we look at this accident history, we find that in 68% of these accidents, Automation Dependency played a significant part".
So automation certainly is not the solution currently, at least not by itself. It handles a lot of the routine very well, but not non-routine situations, which happen frequently enough. Big events like Captain Sullenberger's Hudson landing have been mentioned, but there are many other smaller situations that occur on a regular basis.
On the other hand, automation has been a crucial part of making commercial aviation so incredibly safe and accidents so rare that we no longer really have a statistically relevant sample base. That's an enormous achievement, but automation is only part of the story.
The current amazing safety record has been achieved with the current consensus/compromise setup: humans + machines.
Most of the problems that happen nowadays seem to happen due to problems at the man+machine interface. Asiana 214 for example would have been fine as a fully automated approach just as much as a fully hand-flown approach, but confusions between the automation and the pilots caused a crash.
However, looking at those statistics and saying "we must remove one element" is fallacious, simply because the statistics do not show how many crashes were prevented by having that element aboard.
So while these sorts of radical proposals sound good and appear to make a lot of sense when not examined too closely, I would be very surprised if the objectively best solution is not the combination that we have, refined further to smooth out problems at the computer/human interface.
Automated trains and cars have been mentioned as examples. Planes are fundamentally different, because with any surface bound vehicle you can, in an emergency, just stop and wait for help. For a train, that's literally it. For a car, you'd probably want to pull over somewhere safe, but if worse comes to worse you can stop in place and hope everyone around you stops as well or avoids you.
If you "stop" a plane in flight, everyone on board dies. You have to actively land the plane on a suitable surface at suitable speed and orientation, arguably the hardest part of flying. That's also one reason why requirements for airplane engines are so different than for cars: if a car engine dies, you pull over. If a plane engine dies, you also die unless you find a good place to land in the next couple of seconds or minutes.
53 comments
[ 2.9 ms ] story [ 117 ms ] threadA lot of the crashes are due to equipment failure. Take a look at the series "Aviation Disasters". A pilot is needed to figure out what is wrong, and to compensate to bring the airplane back. A computer program can't do that.
Such is far, far more complex and nuanced than dealing with a failure on a car. Just take Sully's landing on the Hudson after a flock of geese took out his engines. What computer program could have figured out why the engines failed, and what was possible for the airplane to do? Evaluate the weather conditions, and make the best choice of landing spot? Phooey.
We saw this on Air France - where the pilots' panic caused them to ignore readily available information that would have saved their lives.
On the whole of it, I'd rather the plane be flown by a computer, with a remote pilot jumping in for extreme edge cases.
Someone should be able to physically take some kind of control when the computer pilot breaks and the comm link to the remote pilot doesn't work.
On the other hand, I could see freighters (e.g. likes of FedEx fleet) fully automated in the near future (5 years? less?)
And if the pilot had nothing to do, how the heck would they not go mad from boredom?
This may provide some useful insight.
207 crashes - human error; 82 crashes - weather conditions; 74 crashes - mechanical failures; 23 crashes - crime
And besides, what do you do at 30,000 feet with a planeload of people and an emergency and nobody in charge?
This is more a political problem - customer fears, unions, regulations, etc.
The marketing campaign writes itself!
Indeed, it's always the "unanticipated situations" that cause problems both for humans and computers - and in the case of flying, I think it's better to have some level of automation to reduce human error, but there must always be a manual override. Completely replacing the pilot with innumerable lines of code executed by a computer would be even more dangerous as it's far more likely that a pilot can react intelligently to assess the situation and take manual control of the plane than someone be able to immediately find the bug and fix it, without introducing any new ones, then update the software "on-the-fly" (literally).
It's similar to the issues that abstractions cause in software: the more we rely on them to hide details, and the more complex the systems we build as a result, the harder it is to reason about the problems that arise due to something lower down in the stack. "Out of sight, out of mind."
Here is a related article about this: http://spectrum.ieee.org/computing/software/automated-to-dea...
Well, yes, the easy bits. The problem is that you still need pilots when things don't go exactly to plan, and they never do: http://www.askthepilot.com/questionanswers/automation-myths/
This is not to say that passenger planes can't be automated someday, but we're nowhere near there yet.
Are we forgetting that it was a pilot in 2009 on US Airways Flight 1549 that safely landed the plane after striking a flock of geese during its initial climbout only to be ditched into the Hudson River shortly after. If a pilot was not in the plane at the time, no computer would have been able to save the plane.
Typically a pilot will turn off the autopilot around 1000 feet and then do the rest for descent and subsequent landing. On takeoff, most pilots will fly to around 10,000 feet, some to cruise altitude before turning on the autopilot. There are some tasks on modern day planes that cannot be currently done without a pilot (at least not safely).
We have to remember things like autopilot still require manual input from the pilot and cockpit crew to get from its departure point to its destination. A pilot doesn't just fly a plane out of an airport, press a couple of buttons and then do nothing for the rest of the flight, this is a misconception perpetrated by the media and so-called experts who have never flown a plane in their life. Autopilot is merely a tool, much like a machine keeping a patient breathing and their blood flowing in the operating theatre is a tool for surgeons. The machines need to be calibrated, have their data input and be closely monitored much like a pilot does on a plane, even with autopilot engaged.
While I applaud the forward thinking from people like Steve Jurvetson, statements like "Planes are mostly already flown on autopilot already" would be strongly disagreed with any pilot you ask. No two flights are the same and all flights are made up of individual and manual decisions from the pilot and crew. If a pilotless plane were something that could be a reality, the airline industry which has been struggling for years now to keep costs down amidst rising fuel costs would have already partially integrated this or started to cut down on the number of pilots they hire.
I would like to see a pilotless plane decide what to do when the landing gear fails and an emergency landing needs to be made. I would like to see what happens when a tire blows on a high-speed take off and the plane has to make an emergency landing, or what happens when a plane has to make an emergency detour over rocky terrain.
You would still need people to operate these planes remotely, what happens when contact is lost with a plane and it finds itself flying without a pilot at 35,000 feet who can manually take control of the plane and land it? Is Steve forgetting that a pilot would effectively be replaced with someone in a call centre type building sitting at a computer terminal remotely programming commands and programming things like runway codes and other things into the plane?
Some things just can't and should never be replaced. Is modern day technology meant to replace doctors? Every time a new device intended to help keep a patient alive is created, is its intent to aide or replace a doctor? Of course not. People would never agree with replacing doctors with machines completely.
Looking to eliminate drivers on the ground is one thing, because ground transportation is far more dangerous than air transportation. I agree with driverless cars because we already have the data and vehicle trips thanks to mapping data and technology mean they are predictable to a certain degree. Air travel is the safest kind of travel there is, fatalities and accidents have been consistently going down year after year, the number of accidents with fatalities in the last 10 years on Western commercial airlines can be counted with both hands.
Replacing pilots in commercial aircraft is something that will not happen in my lifetime, or perhaps ever. It is one of the worst ideas I have ever heard. Not everything can or should be automated for the sake of auto...
Edit: Can anyone tell me about gyroscopes in the plane and why there weren't any or did they cease function? Would they have helped if they somehow gave indications that the plane was in fact not stalling? From what the data showed, the air speed sensors (pitot tubes) clogged with ice and triggered a stall alarm so the pilot reacted by pulling up but since the plane was actually level but the alarm reported a stall due to lack of airspeed reardings from the sensors it made the situation worse [1].
[1] http://www.popularmechanics.com/technology/aviation/crashes/...
The sad thing is you can hear the cockpit alarm say STALL STALL STALL and the pilots never even acknowledge it.
And I wouldn't call pilots incompetent. I am an instrument rated pilot myself and flying on instruments in a lousy weather at night is hard by itself. Add an unexpected and unexplained failure on top of it and you have a situation that can simply "overload" your brain. There are good lessons from this tragedy and Air France adjusted its training program almost immediately after the crash details have been uncovered.
This is why I was asking about gyroscopes in the planes in enough locations to accurately tell a pilot "hey, I know your airspeed reads very low and the stall alarm is going off but you're not stalling. Lock the throttles and hold the plane level until the airspeed returns."
It almost seems like something else should be added to the training curriculum and routinely tested that hardens pilots nerves and prepares them for some of the most stressful situations they could ever encounter such as before grunts go off to combat.
[1]http://www.technologyreview.com/view/424125/flight-447s-fata...
If you are interested to learn about "too much automation" problem there is a very good (and funny) talk given 15+ years ago:
http://vimeo.com/64502012
If 1% of landings are on autopilot, why not all of them? He doesn't answer this except "setting it up is hard". But setup can be done remotely, if not in advance, by someone who's also controlling several other aircraft, so there are efficiency gains.
We have had driverless trains for decades - see London's DLR. The ticket inspectors can ask the passenger sitting at the front to please change seats so they can unlock the control panel and handle abnormal situations.
Yet we are only still trialling driverless trucks on mine sites. Even with the sites modelled entirely (and frequently), and with LiDAR, radar and cameras on the trucks, it is still far from perfect. I would argue that an aircraft is in an even less controlled environment (you can stop the trucks when it all goes wrong or comms fails!), which is why it's such a hard problem.
Trains, relative to trucks and planes, are literally on rails. This removes one of the greatest challenges (navigation).
A train driver's main job is to have good reflexes. Aside from opening the doors in case of fire, I can't think of a railway accident that couldn't have been prevented by simply stopping the train(s) in time. A driverless train simply needs an emergency stop button and a way for passengers to get out in an emergency.
In a computerised system, the brakes would most likely have fault reporting. Actually in this case they did - the logs indicated that the dynamic braking system wasn't producing any current. This should have been indicated to the driver.
There's also plenty of human error there in not reporting the brake failure and 'eyeballing' weights. Passenger trains are presumably designed such that even with severe overcrowding and lots of luggage there is small chance of the brakes being underpowered. I have no idea why an emergency brake system wouldn't activate all possible means to slow things down.
I agree that in this instance braking actually worsened the problem, but the solution was still that the train should have been braked properly. A computerised system would likely have performed similar steps although it would know the braking force provided by each technique and respond accordingly.
Your listing of current problems is fairly accurate as to the road blocks, however, your insistence that technology can't overcome them seems knee-jerk and unreasoned.
> Replacing pilots in commercial aircraft is something that will not happen in my lifetime, or perhaps ever. It is one of the worst ideas I have ever heard. Not everything can or should be automated for the sake of automating them.
This in particular is obviously hyperbole (at least, being charitable), since the obvious case for computers taking over from pilots is that they solve several issues: their reaction times are considerably faster, they're considerably better at reasoning about the physics of a situation, they learn from all situations encountered by any artificial pilot (or actual pilot) we have records of, they have greatly larger numbers of training hours, and they don't suffer fatigue.
Arguing that computers will never take over for pilots or that it's a bad idea requires that you raise an argument about the limits of computers that is in the /future/.
As an aside, your comment about doctors is also wrong: I'd trust Watson to do diagnosing over most doctors I've met, and only expect my comfort with robotic doctors to rise as technology improves. I expect that the next major improvement we'll see is fully robotic surgeons.
Again, your post reads as partly just unreasoned, anti-technology ranting, but I feel I should give you a chance to explain those points.
Factually, he's largely correct. For example, A F/A-18F Super Hornet can be flown between waypoints and use Terrain Following Radar coupled to the autopilot. The jet can also fly a carrier approach "coupled up" on autopilot.
Could a datalink control the mission computer on a F/A-18F? Quite possible. Could a X-47B fly autonomously onto carrier deck? It's been done.
Could a F/A-18 WSO(eg. non-pilot) program the mission computer and land the jet without manual flying? If the runway is equipped for ACLS (Not ILS) approaches, then yes.
Could a flight attendant program a 737-700 FMS and fly a coupled ILS landing? Probably.
Could the X-47B mission systems be integrated on a 737-700 and fly autonomously? Yes. Its relatively rare for a tower controller to wave-off a 737 because the pilot didn't fly a proper and stabilized approach. Its up to the flight crew to fly a go-around if they are not stabilized on approach. Are they going to start putting LSO (Landing Signal Officers) at JFK to wave off an automated airliner? Maybe...
If an X-47B looses its engine the chance of the jet ditching intact is improbable. If an F/A-18 has a dual engine failure its almost certain the crew will eject. If an A320 looses both engines due to a birdstrike, thats when Capt. Sully earned his paycheck and then some. An autonomous flight computer cant factor the rare emergencies.
There is probably large public support for getting rid of the TSA and having remote override for the FMS might work. Taking the pilot out of the cockpit isn't the solution to the TSA problem.
> Typically a pilot will turn off the autopilot around 1000 feet and then do the rest for descent and subsequent landing.
This hasn't been true anymore for about half a decade. Most landings are executed by autopilots. It depends on the airfield and weather and ...
A famous example was an event in 2006 for London Heathrow, when there was a period of about 3 weeks where the airport was autoland-only (has happened several times since then too). No human was allowed to land a plane there, but autopilots were. It resulted in two things, firstly a capacity reduction of about 40%, and a large amount of pressure on airlines to incorporate autoland features into their planes.
> Some things just can't and should never be replaced. Is modern day technology meant to replace doctors?
Yes, definitely. AI algorithms result in better patient outcomes.
> Every time a new device intended to help keep a patient alive is created, is its intent to aide or replace a doctor? Of course not. People would never agree with replacing doctors with machines completely.
Completely, perhaps not. We'll see. I don't think we'll really get the choice though.
What we'll probably see first is larger one-pilot aircraft, able to land themselves in case of pilot failure. Also, unpiloted fighter jets are almost inevitable, simply because modern airframes can pull more Gs than pilots can stand. Boeing is modifying old F-16 aircraft into the QF-16, which is unmanned. These were originally intended for target practice, but DARPA is interested in using them for other purposes.
[1]http://rt.com/news/iran-drone-hacked-downed-353/
---
I focused on scenarios for removing all security checkpoints and delays from a customer perspective.
Imagine checking in at a kiosk to get your boarding pass, and going though no security lines to board the plane. Bring anything you want with you, but know that the flight is under video surveillance, like retail stores today.
I assumed technologies that work in rudimentary form today and will benefit from Moore’s Law. (The only 12-year forecast that I felt confident about is the continuation of the 100-year abstraction of Moore’s Law, bringing a 256x computational advance by 2018).
So, I started with the assumption that computer-controlled flight would be possible. It’s a pretty safe assumption given what we already have today.
With no cockpit, everything changes. The potential for harm is greatly reduced if the plane cannot be navigated from within. No hijacking. No use of the plane as a weapon.
Bombs become the only threat, and a reduced one.
Personal weapons? A gun or knife-fight could do more damage in a restaurant, or many large group gatherings. Why bother with a plane where criminal activity will be recorded, and the only people harmed are on board?
As for bombs, passive sniffers in a free flowing airport gateway are more plausible than detecting improvised weapons than could be used against a pilot.
At the airport, a quick fingerprint biometric would be a natural way to get a boarding pass (as 12 million people have already done in Florida to get access to an amusement park). So even smuggled bombs would have more capture and downside risk for a terrorist cell than other targets.
Pie-in-sky ideas: hardening a UAV to bombs should be easier than current planes; smaller planes could lower risk; luggage could fly separately; biologic weapon sensors could trigger a flight path to quarantine, etc.
fkd
I'm guessing the big element that was missing was sensing track obstacles and braking accordingly - in Singapore/Dubai they have doors limiting access to the track, and in vancouver they have (tending to failure in the winter) track sensors to see if anyone has hopped onto the track.
http://independentflightinstructors.com/instructors/2011/11/...
(Automation systems typically tell the pilot where to go with a magenta line, on the G1000 systems I am used to it's an indicator of GPS navigation)
From the opening line: "as we look at this accident history, we find that in 68% of these accidents, Automation Dependency played a significant part".
So automation certainly is not the solution currently, at least not by itself. It handles a lot of the routine very well, but not non-routine situations, which happen frequently enough. Big events like Captain Sullenberger's Hudson landing have been mentioned, but there are many other smaller situations that occur on a regular basis.
On the other hand, automation has been a crucial part of making commercial aviation so incredibly safe and accidents so rare that we no longer really have a statistically relevant sample base. That's an enormous achievement, but automation is only part of the story.
The current amazing safety record has been achieved with the current consensus/compromise setup: humans + machines.
Most of the problems that happen nowadays seem to happen due to problems at the man+machine interface. Asiana 214 for example would have been fine as a fully automated approach just as much as a fully hand-flown approach, but confusions between the automation and the pilots caused a crash.
However, looking at those statistics and saying "we must remove one element" is fallacious, simply because the statistics do not show how many crashes were prevented by having that element aboard.
So while these sorts of radical proposals sound good and appear to make a lot of sense when not examined too closely, I would be very surprised if the objectively best solution is not the combination that we have, refined further to smooth out problems at the computer/human interface.
Automated trains and cars have been mentioned as examples. Planes are fundamentally different, because with any surface bound vehicle you can, in an emergency, just stop and wait for help. For a train, that's literally it. For a car, you'd probably want to pull over somewhere safe, but if worse comes to worse you can stop in place and hope everyone around you stops as well or avoids you.
If you "stop" a plane in flight, everyone on board dies. You have to actively land the plane on a suitable surface at suitable speed and orientation, arguably the hardest part of flying. That's also one reason why requirements for airplane engines are so different than for cars: if a car engine dies, you pull over. If a plane engine dies, you also die unless you find a good place to land in the next couple of seconds or minutes.