Could AI leverage Air Traffic Controllers to relieve the shortage?

[Arthur Hanson]

There is currently a shortage of 1200 air traffic controllers, could not AI leverage the talents of existing controllers to handle the workload. AI would only have to relieve about ten percent of the workload to solve the shortage and as time progressed maybe even handle the bulk of the workload? any thoughts or comments appreciated.

 

[Jert]

It might take years to demonstrate and prove such capability is safe and reliable (even though large part of aircraft operations are by computers but under human close control / supervision).

I do not think people today are comfortable enough to let invisible computers to decide whether they live or die on a massive scale.... (look how autonomous driving is going, one step forward and two steps back)

 

[Arthur Hanson]

It might take years to demonstrate and prove such capability is safe and reliable (even though large part of aircraft operations are by computers but under human close control / supervision).

I do not think people today are comfortable enough to let invisible computers to decide whether they live or die on a massive scale.... (look how autonomous driving is going, one step forward and two steps back)

It wouldn't be the computer making the decisions at first, but giving the controller the best options and this would start out as a simple program and be added to as the changes prove themselves. This will be true as we try to improve literally everything we want to use AI/ML on. At first it would be a limited test as with all new technologies. If we didn't follow this path with jet engines, we would still have prop airliners.

 

[lilo777]

By its nature, AI tools/model are non-deterministic. It's really difficult to prove that they are 100% correct even if they behave great. I do not know enough about the air-traffic controller job to say whether AI-generated advice is useful or not. If the controller still has to analyze every advice that may not be much of a help. In somewhat similar situation, AI shows great results in X-ray picture analysis (for different types of diagnostics) and yet, from what I read, AI results still get verified by a human (or two). I assume that the models are tuned up to minimize false negatives but still, because of the liabilities (and morals), if one can't prove that false negatives are at 0%, every X-ray shot have to be analyzed by a human.

 

[tooLongInEDA]

In principle I see the challenge here as the safe and most efficient management of limited network capacity (flight routes, runway space, etc).

As with autonomous driving (and road pricing when that inevitably comes with it), a fully automated non-human system should be more efficient. Right now we are probably buying some safety margin in these systems at the cost of efficient network/resource usage. So I feel certain there are potential efficiency and cost savings left here that we should at least consider. Given the alternative is building more airports and roads, or facing more congestion and delays. It's not just a US issue.

It should ultimately be possible to do this as safely, if not more so, than current systems. But it may take a long time to do so and persuade governments and the public. I'd be reluctant to take humans out of the system completely. Better to start with AI assistants. Also, I'm sure interaction between the people is a key part of the job.

It certainly feels possible that AI could be used to improve the performance (productivity, reliability, etc) of air traffic controllers (as it can with coding).

But there must already be a huge amount of pre-AI technology and automation in air traffic control.

But what is behind the recruitment challenge in US air traffic control ? It's still heavily unionised in much of Europe - but I think less so in the US.

 

[blueone]

But what is behind the recruitment challenge in US air traffic control ? It's still heavily unionised in much of Europe - but I think less so in the US.

The US system for training and certifying new air traffic controllers is ridiculously inefficient. The training and certification process occurs at a single government facility (the FAA Academy) in Oklahoma City. You must be 30 years old or younger when you start training, and you must be a US citizen. You must have excellent oral English communication skills. You can get the academic training up front at accredited colleges around the country, but there's no way around moving to OKC for 3-5 months. At the OKC Academy you'll also be assigned an air traffic control specialty, which is based on FAA needs at the time. And then there's the challenge that the air traffic control system will assign new graduates to an airport in whatever city they choose, without prior notice, and the controllers will probably get moved around through their careers to other cities as they get more work experience, which allows them to be controllers at progressively more challenging airports. (The FAA pays some relocation costs, but from what I've read it looks quite meager compared to corporate relocations.) So if the controllers have families their lives can be just as challenging with unplanned relocations as some military personnel experience.

And after all of that, there's a mandatory retirement at age 56. So your career is probably limited to about 30 years, though you can apparently become an instructor at the ATC-accredited colleges.

Some of the current controller shortage was caused by the shutdown of the FAA Academy during the COVID pandemic. Also, federal government shutdowns that are caused by congressional funding fights cause the Academy to shutdown, which lengthens the training period and stalls the pipeline. I've read there are approximately 2600 students at the OKC academy at any given time.

My spouse has a friend whose husband has been trying to become a controller, so she hears some of the frustrations first hand. You can't help but shake your head in disbelief when you hear of the details.

 

[Arthur Hanson]

My father was on a congressional set up team to increase efficiency in the military bases of all types. They had a good record of return on investment, but not in working with politics. Even though they saved substantial net money, Congress shut down the team to save money. I guess not much has changed from years ago.

 

[tooLongInEDA]

The US system for training and certifying new air traffic controllers is ridiculously inefficient. The training and certification process occurs at a single government facility (the FAA Academy) in Oklahoma City. You must be 30 years old or younger when you start training, and you must be a US citizen. You must have excellent oral English communication skills. You can get the academic training up front at accredited colleges around the country, but there's no way around moving to OKC for 3-5 months. At the OKC Academy you'll also be assigned an air traffic control specialty, which is based on FAA needs at the time. And then there's the challenge that the air traffic control system will assign new graduates to an airport in whatever city they choose, without prior notice, and the controllers will probably get moved around through their careers to other cities as they get more work experience, which allows them to be controllers at progressively more challenging airports. (The FAA pays some relocation costs, but from what I've read it looks quite meager compared to corporate relocations.) So if the controllers have families their lives can be just as challenging with unplanned relocations as some military personnel experience.

And after all of that, there's a mandatory retirement at age 56. So your career is probably limited to about 30 years, though you can apparently become an instructor at the ATC-accredited colleges.

Some of the current controller shortage was cause by the shutdown of the FAA Academy during the COVID pandemic. Also, federal government shutdowns that are caused by congressional funding fights cause the Academy to shutdown, which lengthens the training period and stalls the pipeline. I've read there are approximately 2600 students at the OKC academy at any given time.

My spouse has a friend whose husband has been trying to become a controller, so she hears some of the frustrations first hand. You can't help but shake your head in disbelief when you hear of the details.

That's so different from how it used to be here in the UK. Doesn't sound good at all. We had a neighbour who was in ATC at Heathrow. Very well paid job. Never needed to move. Retired relatively young with a cruiser on the River Thames. Supposedly privatised here now.

 

[Arthur Hanson]

Jensen Huang at the Deal Book interview with Andrew Ross Sorkin conference laid out the pathway that AI will have to take to manage air traffic control systems. It will be interesting to see when this comes to pass and if AI can handle air traffic control, what will it not be able to handle?