Is technology now good enough to fully automate the US air traffic control system? I strongly doubt it, but I wanted to give better informed Dopers a place to chime in. I didn’t want this topic to distract from the thread about the plane crash at DCA the other day. (Breaking News) Passenger plane crashes near DC-area’s Reagan Airport - #129 by Bullitt
ATC isn’t just about keeping the dots on the screen from colliding. They need to be real-time problem solvers, language interpreters, and crisis counselors all potentially at the same time.
An AI will struggle to handle a panicked, disoriented, or overwhelmed pilot. Pilots don’t always give clear, accurate, or complete information, especially in an emergency. AI will not recognize if a pilot sounds distressed or if their communications don’t entirely add up. AI works best in highly predictable, repetitive, structured environments with clear rules and procedures.
Automation can eventually be a helpful assistant to ATC, but totally handing over the reins would be an astoundingly bad idea.
Efforts are now afoot all over the planet to increase the amount of automation in both ATC and in piloting.
A lot of current procedures already amount to ATC’s computers suggesting a course of action to the human controller who evaluates and approves it, then relays it to the human pilot who also evaluates and approves it, then commands the airplane’s computers to do it.
Getting those people out of the middle is the goal of governments and private indusry alike. They hate labor costs.
We’re decades from being able to do much of this for real. But the groundwork is being laid now.
My guess is we’re about as close to this as we are to truly self-driving cars that can be safely trusted on all roads in any conditions.
It needs so-called ‘strong general AI’, and ChatGPT notwithstanding, I don’t think we’re close to that yet.
Exactly. What happens to an SDC when you go to an event with parking in the grass where one doesn’t normally drive & there are no lane lines to follow? What happens on a snowy night when there are no lane lines to be seen & followed because everything is white? What happens in a flash flood when the guard rails on the side of the road get lower & lower until they are underwater (I’ve seen that myself, multiple times. I know the tops of those guardrails are 2-3’ high & that my car can’t ford that & I turn around), will the car stop or keep driving until the car starts floating downstream? What happens when the water is only an inch or two on the roadway but up ahead there is no roadway because it washed out?
What happens when the plane that landed in front of you hasn’t cleared the runway quick enough? Or a plane crept over the hold short line? Or a plane that was stopped & then authorized to cross a runway to continue taxiing doesn’t start rolling quickly enough? Or when there’s an in-flight emergency & they need to be directed to circle somewhere while running checklists & diagnosing the problem & figuring out a best course of action & then needing vectors (directions) back to the departure airport because that wasn’t in their original plan?
For routine takeoffs & landings it’ll might work fine today but it’s all those not-necessarily-everyday-but-not-so-uncommon things that are much, much tougher to program for.
The common refrain in the cockpits, and I expect the ATC breakrooms alike, is that weather & traffic congestion are the two things that provide the humans with job security.
I know there have been cases where planes got into dnagerous situations that were then saved by the pilots taking actions that no one had anticipated or planned for. It would be extremely difficult to automated the actions of pilots or traffic controllers to show that same degree of creative problem solving. In that light, it seems like we’re a long way from automating those roles.
But, a fair debate should consider whether automation can prevent those situations which require human intelligence to get out of.
If the flying was automated along with ATC, you wouldn’t have planes straying onto runways for the ATC to worry about. In theory, anyway.
I don’t think we’re ready to automate air traffic control. There will always be weather changes, mechanical breakdowns, and other uncertainties that planes have to deal with. I just think that it’s more complicated than we’ve discussed so far. Automation might not be able to solve the big problems that humans can, but it might be able to recognize and correct small issues to prevent them from becoming big problems.
We could probably do that now, it would just require buckets & buckets of money to put sensors in both the planes & all of the intersections at every airport as I’m not sure GPS can be trusted to be that precisely reliable 100% of the time.
Of course you would say that
I would think automation would be able to handle congestion better as human beings have a long decision making time.
I am all in favour of automating the traffic control for the private planes of the billionaire owners of AI and tech firms and, if you twist my arm, AirForce One and AirForce Two too. I know there may be some collateral damage when they crash into other jets, but sacrifices must be made for the greater good of humankind.
ETA: And a link to The Pit, brand new thread, right on topic:
As said, present AI isn’t up to the task. At least as important is that present AI doesn’t actually understand anything, and since it doesn’t really know what it’s doing or the point of doing so it also won’t know when it’s doing something wrong.
If the rules it’s following accidentally tell it to slam planes into the ground or each other it’ll do so, because it doesn’t understand that preventing that is the point. And it won’t tell a human something is wrong and ask for help because it won’t know that anything is wrong.
And then there’s the problem that the people who want to get rid of all the air traffic controllers are motivated by greed, so you can fully expect an automated system to be as cheap and shoddily made as they can get away with. While being trusted with tasks they know it can’t handle, because they don’t care in the slightest if thousands die as long as they can make an extra few bucks. A hidden advantage of human aircraft controllers (and pilots, and so on) is they are almost certainly much less psychopathic than their bosses and will actually try to keep the people in their care alive.
I gather you have never seen automated guidance systems at work?
As well as assisting the ATC, it would be good as a redundant system(s). So the automated system can flag if it thinks it has found an issue with what the human is doing and has alternative plans set up that the human can choose to switch to.
…which I would WAG is probably the case already. There’s no doubt plenty of intelligent systems helping the ATC, part of how these kinds of accident are so rare (I would speculate about some of the differences with the recent incident but I think it would be a hijack).
What would a fully automated air traffic control system have done on 9/11?
Once properly ordered by the folks in charge, it would have implemented the existing plan to ground aviation in an emergency.
For darn sure it would have been a confusing shambles. Just like the human-executed effort on 9/11/2001 was.
Believe me, IMO any idea that we’re ready for automated ATC is simply laughable. But the idea that 9/11 is some sort of useful example of “It couldn’t possibly do that” falls IMO far short of convincing.
I believe that AI-controlled cars are—or soon will be—more reliable, safer, and more accurate than human drivers under normal driving conditions. However, they may not yet outperform humans in abnormal conditions, such as driving in snow or on grass. Therefore, the best approach is to let AI control the car while keeping an alert human in the driver’s seat, ready to take over whenever abnormal conditions arise.
The challenge is ensuring that drivers stay alert and engaged when the AI is functioning well. Some folks might drift off if they have little to do. One possible solution is to use pressure sensors on the steering wheel that stop the car if neither hand is detected on the wheel. Another is to use iris tracking that triggers an alarm if the driver’s eyes aren’t looking forward for more than a couple of seconds. These safeguards would help ensure a human is always ready to step in during unpredictable conditions.
I’m not familiar with air traffic control, but perhaps a similar principle applies: AI may be better than humans under normal conditions, but less reliable in a crisis. Thus, the ideal setup might be to let AI handle routine tasks while a human controller remains on alert—again, with something like iris tracking to ensure they’re actively monitoring the systems.
Only when (and if) general AI consistently and reliably outperforms humans in both normal and abnormal conditions should we hand over full control.
Personally, I think AI has its uses, but I also think that it is extremely foolish to entrust something that is safety critical to a device that you don’t understand how it works and can’t guarantee that it will respond in a deterministic way.
AI can be trained to do some very interesting things, but the I an AI isn’t actually very intelligent yet. Current AI systems have no actual understanding of what they are doing, and because of that you end up with occasional weird things like an AI that can’t accurately count the number of "R"s in the word “strawberry”. That’s not the type of system that you want to trust human lives to. You never know when it’s going to do something weird.
And if it does do something weird, how do you fix it? You don’t understand how it works. You can try to train it better, but again, the strawberry example is a good reference. Even when the AI was told that it was wrong, it continued to give the wrong answer. It took a fair amount of training to get the AI to properly count the number of "R"s in strawberry. If you find a bug in your AI air traffic control system, how do you insure that the bug gets fixed if you don’t understand how the system works?
A lot of folks are assuming that automation means AI. There are other ways to automate things. You can program deterministic systems to automate the flow of air traffic without using AI. Modern aircraft use quite a bit of automation just to control the plane, but it’s not AI doing the controlling.
I think a system to fully automate air traffic control would be possible, but it would be extremely complex and would require a LOT of testing to get all of the bugs out of it.
Why can’t a display in an airplane cockpit show the same display as in the air traffic control center? [more precisely a truncated version of it–within X miles of the aircraft…]
It would seem that this kind of display would have prevented the Reagan crash.
My non-pilot guesses: data lag to the cockpit, and pilots have too much info coming in already