I find myself disagreeing with Lemmy more and more these days, but ATC seems like one of those things that could really benefit from AI. I don’t like generative AI being pushed into everything these days either, but a well designed AI can take in all of the things an air traffic controller has to manage and identify things a controller might miss.
Of course this is a money making operation which isn’t ideal because capitalism, but I’m fairly certain this will either reduce or maintain existing incident rates while making it more efficient
The issue is YOU are talking about REAL AI - statistical analysis and modeling leading to decision making with increasingly better fits.
The Transportation Secretary is most CERTAINLY vaguely referring to “that Claude thing that can do anything”, and it’s a series of plane crashes waiting to happen.
Couldn’t agree with both of you more. There are opportunities to use analysis and modeling to make travel safer than ever, but I highly doubt the powers at be are interested.
Yeah, ATC seems like a great problem for automation but a horrible problem for LLMs.
If plane is entering airspace, give them a path that doesn’t intersect any other current path or holding pattern if no such path is available (simple version). Or add in velocity and time for the more complex version if the airspace is too busy, so that any paths given do not intersect within x minutes of each other.
Plue takeoff and landing scheduling, which is just the same problem but with some of the path on the ground.
And then add monitoring to ensure planes follow their assigned path within some error margin and a loop that attempts to make contact to correct the course as well as redirect nearby planes farther away from it until it starts moving predictably again.
And weather monitoring that informs parameters, like how close the paths are allowed to be, climbing and turning attack angles, airspeed/groundspeed, etc.
Also things like airspace restrictions (which should be dynamic to handle things like major weather events, accidents, military training, detected drone activity, military action (friendly or enemy), etc).
And then, most importantly, talk to actual ATC people (at high and low levels) from start to end of development, to make sure the system works and is helpful, maybe starting by just simulating how it would direct the current traffic by a separate system to see if any accidents would result.
Or they could just hire more people so they aren’t perpetually understaffed and over worked. This is a safe proven method that has worked for decades, no unproven tech from questionable ai businesses needed.
I find myself disagreeing with Lemmy more and more these days, but ATC seems like one of those things that could really benefit from AI. I don’t like generative AI being pushed into everything these days either, but a well designed AI can take in all of the things an air traffic controller has to manage and identify things a controller might miss.
Of course this is a money making operation which isn’t ideal because capitalism, but I’m fairly certain this will either reduce or maintain existing incident rates while making it more efficient
The issue is YOU are talking about REAL AI - statistical analysis and modeling leading to decision making with increasingly better fits.
The Transportation Secretary is most CERTAINLY vaguely referring to “that Claude thing that can do anything”, and it’s a series of plane crashes waiting to happen.
Couldn’t agree with both of you more. There are opportunities to use analysis and modeling to make travel safer than ever, but I highly doubt the powers at be are interested.
Yeah, ATC seems like a great problem for automation but a horrible problem for LLMs.
If plane is entering airspace, give them a path that doesn’t intersect any other current path or holding pattern if no such path is available (simple version). Or add in velocity and time for the more complex version if the airspace is too busy, so that any paths given do not intersect within x minutes of each other.
Plue takeoff and landing scheduling, which is just the same problem but with some of the path on the ground.
And then add monitoring to ensure planes follow their assigned path within some error margin and a loop that attempts to make contact to correct the course as well as redirect nearby planes farther away from it until it starts moving predictably again.
And weather monitoring that informs parameters, like how close the paths are allowed to be, climbing and turning attack angles, airspeed/groundspeed, etc.
Also things like airspace restrictions (which should be dynamic to handle things like major weather events, accidents, military training, detected drone activity, military action (friendly or enemy), etc).
And then, most importantly, talk to actual ATC people (at high and low levels) from start to end of development, to make sure the system works and is helpful, maybe starting by just simulating how it would direct the current traffic by a separate system to see if any accidents would result.
Or they could just hire more people so they aren’t perpetually understaffed and over worked. This is a safe proven method that has worked for decades, no unproven tech from questionable ai businesses needed.