It is being modernized through the Next Generation Air Transportation System (NextGen). This will take decades and a huge amount of funding.

https://www.faa.gov/nextgen

They accept public comments so all of the armchair aviation experts on HN should probably send them an email and explain how they can solve all the problems by connecting ChatGPT to a controller console.

Why does it have to take decades? That sounds insane

To modernize a running system, you have to duplicate everything so that you can someday do a switchover.

For commerce tech, you switch, see what breaks, switch back and adjust for next try. You may have lost some sales. But probably just inconvenienced people.

When doing travel systems, something breaking is likely a lot of deaths.

Have you read the various reports that are available right there on the web page?

A wide variety of decades-old aircraft, and even newer aircraft are often modernized versions of ancient aircraft models. All of them built and operated with the motto "if it ain't broke, don't fix it", because every change brings risk (and costs money).

There was also a lot of resistance to the idea that every aircraft should broadcast its position, heading, etc. because of privacy concerns. And as long as there are holdouts the tower can't rely on the information being available

Large infrastructure projects always take decades.

Look at the railroad development in EU for example. You have to invent a lot of tech (signaling, communication between trains, interoperability with old systems), build a lot of infrastructure, train a lot of people and roll it out at scale with the goal of zero accidents.

It is just impossible to do quickly.

Yeah, but some of the next generation upgrades have already been successfully tested at some facilities and it's just a problem of updating the rest. Sure it will take years, but it shouldn't take a decade after successful testing in the field.

So AI has come a long way since NextGen was conceptualized. Training an AI on the historical flight activity for an airport seems .... doable? Would it be better than human controllers? We know it would not be subject to fatigue or emotional stress, and the domain is so limited that hopefully hallucinations would not be a problem.

How realistic is it to have ATC performed by an AI?

> How realistic is it to have ATC performed by an AI?

Not very. For safety-critical systems, you probably want something deterministic, and doesn't break when faced with novel situations.

Solvers are probably a better solution than ML approaches, but they can't capture every possible constraint, such as airplane that can only make left-handed turns because of equipment failure.

> For safety-critical systems, you probably want something deterministic, and doesn't break when faced with novel situations.

Humans are not deterministic and can break for many reasons. It's not plausible that an AI could handle "route this aircraft for landing without any right turns" ?

> Humans are not deterministic and can break for many reasons.

Yes. Humans can also self-evaluate about their state, and have culpability. It not surprising that non-humans are held to a different standard.

I'm not sure this is something that is solved by tech. It sounds like they simply don't have enough staff, whether that a budget problem or hiring problem IDK. The article mentions they have a 50% reduction in new hires which either says it's hard to be competent, the job is too demanding, or both.

They had 1 tower with 1 operator and no redundancy at a fairly busy airport, which is definitely concerning.

> I'm not sure this is something that is solved by tech

in this case it seems like it could definitely have been solved by tech. both aircraft were traveling at predictable paths and opertating nominally.

In the leaked ATC recording, the display is showing a collision alert before the impact. It just doesn't have any capability to do anything about them, it has to rely on the air traffic controller to resolve the issue.

You could argue that helicopters, being very nimble craft, should have ADS-B receivers and provide display and sound their own collision warnings to their pilots. That's perfectly possible, I just don't think it's mandated

This is a nearly perfect problem for tech. We solve and automate very similar problems in other contexts at scale but FAA is extremely conservative about tech.

Your comment assumes the current system can be scaled indefinitely. But you also acknowledge that we can't currently fill the staffing. I also explicitly called out that the modernization might not be tech changes but procedure changes (realistically it would likely be procedure changes made possible through tech).

Doing something to distribute information differently could result in redundancy such that if the commercial craft have a 1-3 mile ADS-B view around them, they can check that ATC isn't giving them bad instructions and that others are following directions.

> Your comment assumes the current system can be scaled indefinitely.

Why couldn't it be? We need fewer ATC than we need pilots, by a long shot. The actual requested new hires for ATC is in the range of 2k (with 1k expected to stay on staff).

With roughly a 120k salary, and assuming all 2k stay on-board that's what, an additional $200mil needed to fully fund?

Meanwhile, the airline industry had a $30b net profit, so it doesn't look like this is even something that'd bankrupt the industry if we taxed them to cover the costs of ATC. It's a 1% tax that's needed.

Even with a super duper fancy modernized system, we'd still want towers staffed sufficiently. Having just 1 person there is a problem even if everything is super modernized. When an issue comes up that the system can't or isn't handling well that's 1 person to make all the calls. Hopefully 2 things aren't happening and hopefully they don't need or want a second opinion.

As for hiring competent people, the DC metro has 6.3 million people. I find it really hard to believe there's not 3 more in the area that could handle being an ATC operator.

And it's not like we don't already have 99% of the existing buildings, infrastructure, and computers already setup. It's literally just the personnel that's lacking.

"I find it really hard to believe there's not 3 more in the area that couldn't handle being an ATC operator."

They might be capable, but why would they want to?

"Why couldn't it be?"

Things like coordination and communication in high traffic environments don't scale well. Finding efficiencies in the procedures and systems can eliminate potential miscommunication and missed handoffs.

"And it's not like we don't already have 99% of the existing buildings, infrastructure, and computers already setup. It's literally just the personnel that's lacking."

Actually only some airports have been upgraded while others are waiting. Do you think another person would have prevented this incident? From the recordings and facts so far, it doesn't seem like it.

It is a budget problem, not a hiring problem: https://thehill.com/regulation/transportation/4617992-buttig...

Modernizing a system while it is in use is high risk. Absurdly high risk. Might still be needed, but is not a cheap or quick thing.

The implication is some sort of greenfield change-everything-at-once idea?

There is no choice but to do the alternative: modify and modernise all our critical systems piecemeal over decades while they are running, covered by the business cliche "We are building the plane, while flying it."

We all naively desire (especially politicians) the software concept of the big rewrite, the version 2, project phoenix, Second-System Syndrome. With development experience we learn both the psychological pull of the concept and we learn why it is so hard to do successfully. It always seems so simple how to fix societal systems we have no experience with.

Successful systems slowly improve, slowly introduce and prove new ideas, slowly accrete changes: we accept the costs of vestigial path-dependent aspects and technical debt.

One of the incredible learning experiences of developing software systems is that it is a huge melting pot of different techniques that are iterated quickly (fast parallel evolution and proving) and the knowledge diffuses between people and teams and organisations. The waterfall model is rarely used! I believe we are now seeing those skills and techniques be applied to other system domains - especially high growth businesses.

This sounds like an agreement?

Modernize the system how? Like do you have explicit steps on how to modernize it and what needs to be done?

They're working on multiple technical upgrades already, and some airports have been upgraded. I think better information distribution is one thing. Large craft could have ADS-B tracking of a 1-3 mile bubble displayed to them. This would add a check to being blindly told what to do and blindly trusting that others are doing what they're told. You could change certain procedures such as the military having one pilot use NVG while the other not when flying in high traffic and high illumination airspace. No reason we can't add better sensors to planes when you think about all the collision detection we are adding in cars. I'm sure there are many other things that can be changed as well. Some of it's the pace. Stuff that's already been piloted and shown to be effective is still taking forever to roll out at the other airports.

> Large craft could have ADS-B tracking of a 1-3 mile bubble displayed to them. This would add a check to being blindly told what to do and blindly trusting that others are doing what they're told.

Already exists, it's called TCAS

That doesn't work with all transponders, right? Many private or military craft don't work with that system.

TCAS works with Mode-C (same as military Mode 3).

So why didn't it work?

TCAS shows nearby aircraft on a map. It generates a Traffic Alert (TA) to warn of nearby aircraft when certain parameters are met. If they get closer then it generates a Resolution Advisory (RA) which gives climb/descend instructions to avoid a collision. RAs are disabled below 1000 feet.

So there's already tracking on a map that's available to the pilots. During landing, pilots are looking outside and not glued to their instruments.

The real issue is trying to cram too much aircraft in such a small space and time.

I believe it’s turned off when below a certain altitude

This article has a good explanation of TCAS limits.

https://www.twz.com/news-features/heres-what-air-traffic-col...

Yeah, so similar to what I said, the transponder didn't work with that system as that article speculates the UH-60 didn't have it. My suggestion was that you could display what ATC has, hence the data distribution part of my comment. But I guess we could just accept what the article tells us that it would be too dangerous to display that info at lower altitudes and allow pilots to make the call and that we just have to live with these incidents then...