I feel like ATC is one of those pattern-recognition constraint-satisfaction problem jobs where a (non-generative!) algorithm can probably do a pretty good job.
Everything they are doing has known algorithms. Some are np problems though, but ai isn’t any better for those than the existing algorithyms.
but ai is hype today so everyone needs to do it. In a few years this will die off and ai will be used for where it is useful. Just like every other ai that has been a fad for a while since 1950.
Why do you think this is going to replace air traffic control work? It’s picking which gate to park the plane at. This was done by airline and airport operations teams, not ATC. Imagine if you could automatically pick gates to reduce the time a plane spends taxiing and/or minimize time passengers spend walking. That’s 100% a useful application for computer optimization algorithms. Humans aren’t going to do that better and it’s not a function of safety that tower or ground control needs to do.
Computer algorithms solve problems all over the world for companies already. I bet airlines already have teams of people using computer algorithms to figure crew management, flight routing, cost optimization, etc.
The fact that they’re exploring quantum computers and non-classical algorithms just suggests that gate allocation is NP-Hard. Sure things go wrong when computers fail already, Look at Southwest or Delta’s recent meltdown, but to act like this a bad thing is just nonsense. This should be looked at as a good thing that airlines are working on.
LLMs and generative AI aren’t going to be any good on this problem. The article is using the older, non-buzzword computer science meaning, which includes algorithms for this exact problem, such as the ones used for a category of difficult problems known as constraint satisfaction problems. These problems were artificial intelligence problems before the term “AI” was turned into a marketing buzzword.
Allocating gates is one problem that traditional computers and algorithms struggle to do quickly, with calculation times increasing disproportionately to the size of the problem.
But, Dr Doetsch is confident that approaches using quantum computing will crush the problem.
“Quantum algorithms will allow optimally assigning gates, and other resources, even in large airports and travel networks. These algorithms will be able to respond to changing external factors with updated optimal solutions in real time,” he says.
This stuff is cool, and has nothing to do with generative AI.
AI doing air traffic control work?
Jesus fucking Christ, what could possibly go wrong?
I feel like ATC is one of those pattern-recognition constraint-satisfaction problem jobs where a (non-generative!) algorithm can probably do a pretty good job.
Everything they are doing has known algorithms. Some are np problems though, but ai isn’t any better for those than the existing algorithyms. but ai is hype today so everyone needs to do it. In a few years this will die off and ai will be used for where it is useful. Just like every other ai that has been a fad for a while since 1950.
Why do you think this is going to replace air traffic control work? It’s picking which gate to park the plane at. This was done by airline and airport operations teams, not ATC. Imagine if you could automatically pick gates to reduce the time a plane spends taxiing and/or minimize time passengers spend walking. That’s 100% a useful application for computer optimization algorithms. Humans aren’t going to do that better and it’s not a function of safety that tower or ground control needs to do.
all that’s fine and good, but one just needs to see your username to fathom what could potentially go wrong.
Computer algorithms solve problems all over the world for companies already. I bet airlines already have teams of people using computer algorithms to figure crew management, flight routing, cost optimization, etc.
The fact that they’re exploring quantum computers and non-classical algorithms just suggests that gate allocation is NP-Hard. Sure things go wrong when computers fail already, Look at Southwest or Delta’s recent meltdown, but to act like this a bad thing is just nonsense. This should be looked at as a good thing that airlines are working on.
LLMs and generative AI aren’t going to be any good on this problem. The article is using the older, non-buzzword computer science meaning, which includes algorithms for this exact problem, such as the ones used for a category of difficult problems known as constraint satisfaction problems. These problems were artificial intelligence problems before the term “AI” was turned into a marketing buzzword.
This stuff is cool, and has nothing to do with generative AI.