Over-regulation is delaying the automation of flight: Time was when a uniformed man would close a metal gate, throw a switch, and intone, “Second floor- men’s clothing, linens, power tools …” and the carload of people would glide upward. Now each passenger handles the job with a punch of a button and not a hint of white-knuckled hesitation. … And back in the day, every train had an “engineer” in the cab of the locomotive. Then robo-trains took over intra-airport service, and in the past decade they have appeared on subway lines in Copenhagen, Detroit, Tokyo, and other cities. …
Automated cars are so much more important than automated flight, just because we have so many cars and some largish fraction are piloted by angry, bored, distracted, stupid, tired to the point of sleeping, and yes, drunk amateurs. Automation is not a switch, it is a continuum. Probably long before we have automation we will have aids in the car that make it much harder to fnork up and die. Just like planes make loud obnoxious noises when the plane seems to be flying in to the ground, the smarter the car the more usefull will be its bitching us out when we are headed off the road or into another car.
I am so looking forward to getting in to my car and taking a nap or surfing the web while it takes me where I am going. Why don't I use public transportation you may ask? It doesn't go from my house to where I want to go in anything like the comfort and speed of a car, even to the point where my having to pilot the car with all my defects still makes the car a good deal for me.
I submit that the transition to automated flight is going right on schedule.
Automation should, and did, start with small planes. These can't exist if they need to be big enough to carry a pilot. Their small size means they pose much less risk if they crash than larger planes.
Automation should be used when it provides a real advantage over human piloted. For very large planes, the advantage is less as the ratio of the cost of the pilot to the rest of the plane is much less than with smaller, lower value flights.
Technodweebs like us virtually always underplay the risks and overplay the advantages because our bias is to love that flash gordon star trek stuff. In real life, things go a lot slower, and you can't prove that this is a bug, not a feature.
The FAA has must more stringent requirements for aviation software (incl testing) than you will find for any other application, except maybe nuclear weapons.
The question that excerpt doesn't answer is "How many incidents, that would have been accidents but-for the presence of a pilot, occurred, so that we may compare?"
Today, having a pilot on-board is probably safer, though more expensive. I imagine most of the cost is in engineering and verifying the user interface and control linkages, rather than salary or even the fuel cost of carrying extra people, equipment, and structure. Machines are slightly easier to incrementally improve, so presumably safety would shift in favor of machines with time and experience.
A related question is whether the level of safety we have today is sensible. I rather think we spend too much on airline safety, not too little.
I think there are two important factors in automation of control systems for cars and for aeroplanes which are often overlooked by casual observers. (But not by the people who are actually doing the job, mercifully.)
First, there's the nature of the problem being solved. Having driven cars and flown aeroplanes, I can attest that there are far more conflicting hazards that you have to track and react to per minute while driving than flying. On the other hand, the basic task is easier for driving than flying - there are far fewer controls and critical instruments for driving.
Second, there's what to do about exception handling. What should the system do if the instruments indicate a situation the coders thought should never happen (typically indicating some instrument malfunction), or the situation is beyond what the internal model yields sensible results for? In a car, it's quite easy to imagine a fail-safe mode that cuts all power, maybe attempts to apply brakes (if they're working), and brings the vehicle to a stop on the ground. In a plane, not so much. The survival prospects for an out-of-envelope aircraft do not typically improve if you cut engine power.
It's no coincidence that we have driverless trains widely deployed but not driverless cars or planes. If planes are a 3D problem and cars 2D, trains are 1D (very crudely!). And driverless trains do indeed respond to out-of-band situations by simply stopping.
All this said, I think the main barrier to wider development and deployment of driverless cars and planes is socio-political rather than technical.
Flight is not well modeled in all cases because fluid dynamics is a nonlinear system. It is thus difficult to design automatic control systems. There will always be some cases where the control system will fail. The human mind is far better at solving unusual and novel problems when trained for it. Most expert systems are designed to perform routine actions and they do that very well. The best solution is to automate as much as possible and to train the pilots in worst-case scenarios. This way we have the best of both worlds. There is absolutely no need to do away with human beings, just to aid them as much as possible by using machines.
A far more important problem is that of automated cars. Car driving with fixed roadways is a far more predictable system. For example we have train routing systems that are very good and nowadays train accidents are very rare. An easy solution is to remove all roads and instead replace it with tracks and some kind of smart routing to let personal bogies seamlessly get on and off. This need not be done everywhere, but only in dense areas where the payoff is likely to be huge.
During 2004 in the United States, pilot error was listed as the primary cause of 78.6% of fatal general aviation accidents, and as the primary cause of 75.5% of general aviation accidents overall.[1] For scheduled air transport, pilot error typically accounts for just over half of worldwide accidents with a known cause.
This isn't totally definitive, but it definitely seems like pilots are doing more harm than good. I suspect pilots will have to be very harmful (not just slightly harmful or "merely" wasteful) to overcome status quo bias, and spur the government to allow safer, automated flights.
To clarify: yes, when people are commuting to work, an autonomously driven car doesn't eliminate their need to be in the car. But people already spend huge chunks of their day doing stuff they could do in a car if they weren't driving: TV, internet, office work, reading, etc.
Autonomously driving cars don’t eliminate the human element any more than planes with autonomous systems do.
The benefits of autonomously driving cars are too massively staggering to do justice to here, but the obvious benefits are (1) greatly reduced traffic fatalities and injuries, and (2) freed human time. Just to get numbers out there, there are 30,000 people killed in the US every year in traffic accidents (which are almost exclusively driver error). That's an economic loss of about a quarter trillion dollars per year. Also, the 220 million American adults average 90 minutes in their car per day, which works out to roughly a trillion dollars per year at minimum wage.
The question isn’t whether humans could outperform computers in any situation, but whether they do on net and so whether removing them from the loop entirely is a win.
But it's not an either-or question. You could just have the pilot sit in the cockpit doing nothing unless the rare situations where humans out-perform computers occurred. As an added benefit, there wouldn't be nearly so much fear among passengers due to not having a pilot.
> Now then, how would you like to be a passenger on a plane that was being controlled entirely by software, which itself was controlled remotely via various coded (but not absolutely 100% secure) messages?
If you had asked some airplane passengers on 9/11, I wonder what they would have said.
Yes, even hypothetically. Many ways to do it if someone cares enough. See http://cm.bell-labs.com/who... or http://underhanded.xcott.co... or look into firmware viruses or hardware backdoors. And heck, even if the bad guys never go anywhere near the code or components, there's always researching 0-day vulnerabilities (or simply buying them).
All this is true right now, incidentally. Open source doesn't make the situation worse, and it may make the situation better (but don't mistake it for security pixie dust).
An automated car could drive you to work, then go somewhere else and move other people around while you work.
Automated cars are so much more important than automated flight, just because we have so many cars and some largish fraction are piloted by angry, bored, distracted, stupid, tired to the point of sleeping, and yes, drunk amateurs. Automation is not a switch, it is a continuum. Probably long before we have automation we will have aids in the car that make it much harder to fnork up and die. Just like planes make loud obnoxious noises when the plane seems to be flying in to the ground, the smarter the car the more usefull will be its bitching us out when we are headed off the road or into another car.
I am so looking forward to getting in to my car and taking a nap or surfing the web while it takes me where I am going. Why don't I use public transportation you may ask? It doesn't go from my house to where I want to go in anything like the comfort and speed of a car, even to the point where my having to pilot the car with all my defects still makes the car a good deal for me.
I submit that the transition to automated flight is going right on schedule.
Automation should, and did, start with small planes. These can't exist if they need to be big enough to carry a pilot. Their small size means they pose much less risk if they crash than larger planes.
Automation should be used when it provides a real advantage over human piloted. For very large planes, the advantage is less as the ratio of the cost of the pilot to the rest of the plane is much less than with smaller, lower value flights.
Technodweebs like us virtually always underplay the risks and overplay the advantages because our bias is to love that flash gordon star trek stuff. In real life, things go a lot slower, and you can't prove that this is a bug, not a feature.
The FAA has must more stringent requirements for aviation software (incl testing) than you will find for any other application, except maybe nuclear weapons.
Well said, Mr. Mouse.
The question that excerpt doesn't answer is "How many incidents, that would have been accidents but-for the presence of a pilot, occurred, so that we may compare?"
Today, having a pilot on-board is probably safer, though more expensive. I imagine most of the cost is in engineering and verifying the user interface and control linkages, rather than salary or even the fuel cost of carrying extra people, equipment, and structure. Machines are slightly easier to incrementally improve, so presumably safety would shift in favor of machines with time and experience.
A related question is whether the level of safety we have today is sensible. I rather think we spend too much on airline safety, not too little.
I think there are two important factors in automation of control systems for cars and for aeroplanes which are often overlooked by casual observers. (But not by the people who are actually doing the job, mercifully.)
First, there's the nature of the problem being solved. Having driven cars and flown aeroplanes, I can attest that there are far more conflicting hazards that you have to track and react to per minute while driving than flying. On the other hand, the basic task is easier for driving than flying - there are far fewer controls and critical instruments for driving.
Second, there's what to do about exception handling. What should the system do if the instruments indicate a situation the coders thought should never happen (typically indicating some instrument malfunction), or the situation is beyond what the internal model yields sensible results for? In a car, it's quite easy to imagine a fail-safe mode that cuts all power, maybe attempts to apply brakes (if they're working), and brings the vehicle to a stop on the ground. In a plane, not so much. The survival prospects for an out-of-envelope aircraft do not typically improve if you cut engine power.
It's no coincidence that we have driverless trains widely deployed but not driverless cars or planes. If planes are a 3D problem and cars 2D, trains are 1D (very crudely!). And driverless trains do indeed respond to out-of-band situations by simply stopping.
All this said, I think the main barrier to wider development and deployment of driverless cars and planes is socio-political rather than technical.
Flight is not well modeled in all cases because fluid dynamics is a nonlinear system. It is thus difficult to design automatic control systems. There will always be some cases where the control system will fail. The human mind is far better at solving unusual and novel problems when trained for it. Most expert systems are designed to perform routine actions and they do that very well. The best solution is to automate as much as possible and to train the pilots in worst-case scenarios. This way we have the best of both worlds. There is absolutely no need to do away with human beings, just to aid them as much as possible by using machines.
A far more important problem is that of automated cars. Car driving with fixed roadways is a far more predictable system. For example we have train routing systems that are very good and nowadays train accidents are very rare. An easy solution is to remove all roads and instead replace it with tracks and some kind of smart routing to let personal bogies seamlessly get on and off. This need not be done everywhere, but only in dense areas where the payoff is likely to be huge.
Apparently the emergency landing on the Hudson Bay was the result of the pilot turning on the auto pilot.
According to Wikipedia:
During 2004 in the United States, pilot error was listed as the primary cause of 78.6% of fatal general aviation accidents, and as the primary cause of 75.5% of general aviation accidents overall.[1] For scheduled air transport, pilot error typically accounts for just over half of worldwide accidents with a known cause.
This isn't totally definitive, but it definitely seems like pilots are doing more harm than good. I suspect pilots will have to be very harmful (not just slightly harmful or "merely" wasteful) to overcome status quo bias, and spur the government to allow safer, automated flights.
To clarify: yes, when people are commuting to work, an autonomously driven car doesn't eliminate their need to be in the car. But people already spend huge chunks of their day doing stuff they could do in a car if they weren't driving: TV, internet, office work, reading, etc.
Autonomously driving cars don’t eliminate the human element any more than planes with autonomous systems do.
The benefits of autonomously driving cars are too massively staggering to do justice to here, but the obvious benefits are (1) greatly reduced traffic fatalities and injuries, and (2) freed human time. Just to get numbers out there, there are 30,000 people killed in the US every year in traffic accidents (which are almost exclusively driver error). That's an economic loss of about a quarter trillion dollars per year. Also, the 220 million American adults average 90 minutes in their car per day, which works out to roughly a trillion dollars per year at minimum wage.
The question isn’t whether humans could outperform computers in any situation, but whether they do on net and so whether removing them from the loop entirely is a win.
But it's not an either-or question. You could just have the pilot sit in the cockpit doing nothing unless the rare situations where humans out-perform computers occurred. As an added benefit, there wouldn't be nearly so much fear among passengers due to not having a pilot.
I wonder about the exact numbers here. I know plane crashes are rare, but they're awfully expensive when they do occur.
I know the emergencies that call for pilot intervention are rare, but are they rare enough?
> Now then, how would you like to be a passenger on a plane that was being controlled entirely by software, which itself was controlled remotely via various coded (but not absolutely 100% secure) messages?
If you had asked some airplane passengers on 9/11, I wonder what they would have said.
Yes, even hypothetically. Many ways to do it if someone cares enough. See http://cm.bell-labs.com/who... or http://underhanded.xcott.co... or look into firmware viruses or hardware backdoors. And heck, even if the bad guys never go anywhere near the code or components, there's always researching 0-day vulnerabilities (or simply buying them).
All this is true right now, incidentally. Open source doesn't make the situation worse, and it may make the situation better (but don't mistake it for security pixie dust).