AI Future Effects on Aviation Industry.

[rookiepilot]

AI is going to drastically affect many industries, over time, it seems clear.

I am deeply concerned about its potential effects, for many reasons, economic and societal.

Starting this thread for comments.

[Tbayer2021]

I think the direction of the industry is very clear. Even one of the manufacturers has openly stated that their next clean sheet design will be single pilot capable. They're even in the process of making the A350 caple of single pilot operation during cruise.

Most pilots like to stick their heads in the sand but I believe partial and even full automation is much closer than we'd like to think. Sure hope I'm wrong there. AI will undoubtedly be an integral part of automation, at least in the aviaiton sphere.

A close friend is heavily involved in the automation/AI part of the aviaiton industry and its shocking how far along they actually are. The tech is moving at a pace that has left regulators in the dust, and that may actually be what buys some some extra time. That and maybe public perception. Unfortunately for us, these things will only buy us time. I lead my financial life as if I won't see a full career in this field and recommend my friends the same. But clearly it's hard for most to do the same and not enjoy the large paychecks many are getting, specially while young. Can't say I blame them, ultimately no one can predict the future.

We seem to be heading towards a significant societal shift with the potential level of automation in every industry. Maybe it's for the best on the other side, but the change won't be easy for most and there will be no shortage of victims. Or maybe we'll realize that we can't just automate most things simply because we can and have most people unemployed? After all, companies still need to sell their products and services.

[digits_]

"AI" might affect industries, but we're still decades away from it being reliable enough to do anything that matters. I remember 20 years ago when the first 'speech/voice recognition' software came out, that it would be a game changer and we'd be talking to computers instead of using keyboards. Today, would you trust your speech recognition software to write a resume or other important letter without proof reading it? You wouldn't.

AI has that same 'wow, this is super cool!' factor when you first get a demo. But once you start using, you'll notice how unreliable it is.

Anyway, before we can really talk about AI in detail, you have to define what it is exactly. "AI" as used in the media today, refers to a software model that compares your input to a huge set of existing data, and tries to guess what it is you want to hear as your answer. You could argue that's exactly how the human brain works, and in a way you'd be right, but AI is significantly less competent at it than us humans. We humans get it wrong often as well, so imagine how terrible 'AI' is at it.

Will AI affect aviation?

In the cockpit while flying and controlling the plane? No, absolutely not. All procedures in the cockpit are build to eliminate guesswork. Even humans aren't supposed to guess or be creative in the cockpit. Follow the procedure that exists, even if you think it's suboptimal. A classic example would be a TCAS RA. Even if you see the other plane, we're still ingrained to follow the RA instructions, don't deviate from the procedures.
What could be a minor improvement that would be AI based, would be some kind of virtual assistant (think Clippy from MS Word) that could observe everything and spot patterns. "hey, are you sure you want to run that engine at 68%? Similar flights usually run it at 65%". Or "even though this oil pressure is still within limits, there's a trend going that could be an indication of an impending engine failure. Do you really want to proceed with the Atlantic Crossing at this time?".

During flight scheduling? Might be likely. Spotting some weather patters or some delays based on calendar dates. Pattern recognition, guessing, looking at statistics. It's good at that. In the end a human will still make a decision though.

Does that mean there won't be automation or we won't have pilotless cockpits? No, that's a completely different discussion. Those pograms are based on rigorous if A, B and C happen, do X Y Z. Not if A B C happen, we'll try X, if that doens't work, do Y. Operating based on AI models can lead to unexpected results. There are quite a few examples on youtube of people programming AI bots to pass a video game. Imagine that: a computer playing a video game, which in itself is so much less complicated than the real world. And yet so many unexpected situations happen.

An AI pilot is likely programmed to complete a flight in a safe and efficient manner. Safe would likely mean 'look in my big database, and see if my actions here would result in a damaged plane or casualties'. Efficient could be 'try to minimize my fuel flow and avoid delays'.
Eventually, a typical AI only response could be to shut down the engines on downwind and glide it in, as that minimzes fuel and there are likely no recorded airliner accidents following an intentional engine shut down on downwind. Those weird things need to be caught before they happen. Some we will think off, others we won't. And the more you shut down the unexpected avenues, the further you stray away from the current AI models, and the more you result in the 'normal' automation programming sphere.

Another major consideration is the source of your datasets used for comparison/learning. At the moment all that learning data is human created. Eventually, AI generated data will be used by other AI bots. You'll end up in a situation where kids are teaching each other advanced algebra or quantum physics, and they'll all think they're doing great! Then they'll design a rocket or nuclear power plant with their acquired knowledge.

AI could very easily get much worse before it gets any better.

Another funny (but really sad) example. We have companies designing software to screen applicants' resumes for certain keywords and/or experience requirements. It's a pettern recognition, so it warrants looking into AI to improve that.
Then we have companies designing software to help applicants write resumes containing the keywords they think the companies want to see on a resume. You end up with applicants hiring computers to add nonsense buzzwords, because another computer links those buzzwords to having a succesful candidate. It's mental.
Not so much an issue in aviation, but in the tech/sales/finance departments, it just ridiculous.

A lot of the current 'problems' need less AI, not more. It's insane how companies 'want to use AI' without having a problem that needs AI to solve it.

[Bingo Fuel]

AI in the cockpit? I think that's still a while off. I think AI will come first to crew scheduling. Imagine a chatbot that calls or texts people for draft or reserve assignments. Humans still making the decisions (at first), but using the AI tools to do the tedious repetitive tasks.

I think we'll see more AI tools assisting dispatchers as well.

Once the tools have proven themselves to be reliable, then staffing will start getting reduced. This will be the same across a lot of white collar occupations. It'll be just like factory jobs when automation became more commonplace. Instead of the job becoming easier with better tools, it'll mean fewer people doing more work. AI does not take sick days, nor does it need a pension or benefits. It is in the employer's best financial interests to replace as many people as they safely can.

TL;DR: It will happen in a LOT of other jobs before it hits the flight deck.

Look for a lot more people calling for Universal Basic Income as their jobs are made redundant.

[J Roc]

AI in commercial airliners is much closer than 20 years out. We'll see single-pilot airliners very soon, and the public will accept the evolution. We had four pilots, and then our navigation equipment improved - goodbye, navigator. We had three pilots, and then our aircraft systems and automated system monitoring improved - goodbye, flight engineer. We are now presented with artificial intelligence, which has proven it can learn on the fly and assist humans - goodbye, first officer.

Take note when watching those videos: the alpha dogfight trials took place in 2020. AI convincingly dominated a highly trained and experienced fighter pilot, and that was almost 4 years ago.

https://www.flyingmag.com/ai-powered-pi ... -dogfight/

https://m.youtube.com/watch?v=IOJhgC1ksNU

https://m.youtube.com/watch?v=dbsyfITResU

[Red_Comet]

Here we go again.

Gents, we can't even get cars to drive themselves. Cars have a very simple failure mode: they stop moving if anything goes wrong.

Aircraft don't have the luxury of not moving when things go wrong.

The current "AI" hype was kicked off by a GPT. GPT stands for Generative Pre-Trained Transformer. They key word in there is "generative". This is a nice way of saying it's the best type of bullshitter: it doesn't know it's bullshitting. GPTs, which are the current cutting edge, cannot make decisions based on reality, because they have no concept of reality. They only take in tokens, which you can think of as fragments of data (text, images), and produce what they think the next token in that chain is going to be.

Most critically GPTs are non-deterministic. This means you can ask them the same question and you're not guaranteed to get the same answer twice in a row.

In order to automate driving or any task in the physical world that requires responding to unique, unpredictable circumstances (e.g. flying in dense civilian airspace), you are going to need AGI. That's a nice way of saying you're going to need Arnold from Terminator 2. Nothing less is going to cut it, because you need a real-time rational agent capable of creating a realistic model of the world based on empirical testing and validation. Currently, only humans can do this. If this changes, you can start clicking your heels together because we sure as hell won't be in Kansas anymore.

Relax, and enjoy the human century while it lasts. If true AGI arrives, the era of humanity is over, and "jobs" will be the least of our worries.

[pelmet]

Wages have been going up rapidly. Unfortunately, I think the faster wages rise, the more likely AI replacement becomes.

I can see airline companies sourcing out their dispatch/crew scheduling. ATC could be automated with remote towers already in use. Piloting will probably be slower with perhaps military first followed by cargo carriers going to single pilot in cruise. Maintenance and ground handling could be the most difficult.

[‘Bob’]

Here we go again.

Gents, we can't even get cars to drive themselves. Cars have a very simple failure mode: they stop moving if anything goes wrong.

Aircraft don't have the luxury of not moving when things go wrong.

Automation both in general and the increasing levels of automation in aircraft proceeded those in cars by a very wide margin. To the point that 'autopilot' not 'autodriver' was the colloquial term for it at its inception. Planes were already landing themselves with paying passengers onboard just as cruise control was making tiny inroads into automotive world.

Why? Because there's a lot fewer things to hit at altitude and airways and airspace and airports are blocked or controlled (from the ground) just as surely as a train track is.

The current "AI" hype was kicked off by a GPT. GPT stands for Generative Pre-Trained Transformer. They key word in there is "generative". This is a nice way of saying it's the best type of bullshitter: it doesn't know it's bullshitting. GPTs, which are the current cutting edge, cannot make decisions based on reality, because they have no concept of reality. They only take in tokens, which you can think of as fragments of data (text, images), and produce what they think the next token in that chain is going to be.

Most critically GPTs are non-deterministic. This means you can ask them the same question and you're not guaranteed to get the same answer twice in a row.

In order to automate driving or any task in the physical world that requires responding to unique, unpredictable circumstances (e.g. flying in dense civilian airspace), you are going to need AGI. That's a nice way of saying you're going to need Arnold from Terminator 2. Nothing less is going to cut it, because you need a real-time rational agent capable of creating a realistic model of the world based on empirical testing and validation. Currently, only humans can do this. If this changes, you can start clicking your heels together because we sure as hell won't be in Kansas anymore.

Sounds exactly like humans. Conditioned response is "thinking what the next token will be" and there is often a diversity of answers and solutions to complex problems.

Take any emergency or abnormal situation and see the diversity of responses once the SOPs and checklists run out of ideas. SOPs and checklists that humans can't instantly recall or execute perfectly like a machine can. Not to mention those emergencies or abnormalities that were caused by human action or inaction in the first place.

Take AI and don't put it in the cockpit right away. Put it in after its been thrown so many things that it doesn't matter how many different pathways it took, the result is still the same.. to the point that the averages are better than human averages.

AI will likely be in the cockpit as things that pilots are following just as blindly as they are the magenta line right now. I bet that there are pilots flying now that have used machine learning to get their licenses.

[goldeneagle]

In a modern airliner, the job today is mostly just monitoring the automation. When was the last time somebody here hand bombed a multi engine jet across the pond ? Thing is, most of the automation in use today is 20 and 30 year old tech, and in no way shape or form compares to what the state of the art is capable of.

The biggest mistake most here make with respect to how things will change, everybody seems to think in the terms of 'captain' and 'first officer', so they think the FO job will be the one on the block to go join navigator and flight engineer in the realm of memories. But that's not what's going to happen. The job today includes 'pilot flying' and 'pilot monitoring'. Pilot flying is the next to go, for the most part it's gone already.

Another huge mistake many are making, they think the move to automation is to replace the person on the payroll, but it's not, it is a move driven by cost, just not the cost you think of. The flight engineer was replaced by the fadec because it will do a much better job of managing the engines, both in terms of engine life and in terms of fuel flow, the two single largest costs of operating an airliner. Introduction of the fadec allowed the engine manufacturers to increase the complexity of engine management to a degree no person could manage it anymore, all toward the goal of increased efficiency.

As far as the cockpit goes, same will come about, the PF role is pretty much redundant today except for a couple short phases of a flight. The technology is getting better by leaps and bounds, the people are getting more expensive, so it's just a matter of time till the balance tips the other way.

Then ofc there is the argument that the travelling public wont get on an airplane where crew positions have been made redundant. Some said that when the move was 4 to 3, then again 3 to 2. Next time you ride the skytrain in YVR, think long and hard about who is driving, and wether or not the travelling public will still use it. Waymo has a fleet of self driving taxi on the road in Pheonix, they spend the day moving folks around, no drivers on board.

On the bright side, the airliners rolling off the lines today will continue to need 2 crew over their lifespan, but the ones rolling off the line in another 10 years or so, they will be initially certified for operation with one, and they will have the technology on board to eventually end up with zero. It'll be a transition period, much like ETOPS was when it first came along, but the technology to automate lots of things is advancing by leaps and bounds. Those looking at the 20 year old automation technology in an airliner and thinking 'it will never replace a person' are right, but, airliners today do not have state of the art, most of that stuff is 20 or 30 year old tech and the regulatory hurdles to update it were not cost effective. That is also changing today.

[Tbayer2021]

In a modern airliner, the job today is mostly just monitoring the automation. When was the last time somebody here hand bombed a multi engine jet across the pond ? Thing is, most of the automation in use today is 20 and 30 year old tech, and in no way shape or form compares to what the state of the art is capable of.

The biggest mistake most here make with respect to how things will change, everybody seems to think in the terms of 'captain' and 'first officer', so they think the FO job will be the one on the block to go join navigator and flight engineer in the realm of memories. But that's not what's going to happen. The job today includes 'pilot flying' and 'pilot monitoring'. Pilot flying is the next to go, for the most part it's gone already.

Another huge mistake many are making, they think the move to automation is to replace the person on the payroll, but it's not, it is a move driven by cost, just not the cost you think of. The flight engineer was replaced by the fadec because it will do a much better job of managing the engines, both in terms of engine life and in terms of fuel flow, the two single largest costs of operating an airliner. Introduction of the fadec allowed the engine manufacturers to increase the complexity of engine management to a degree no person could manage it anymore, all toward the goal of increased efficiency.

As far as the cockpit goes, same will come about, the PF role is pretty much redundant today except for a couple short phases of a flight. The technology is getting better by leaps and bounds, the people are getting more expensive, so it's just a matter of time till the balance tips the other way.

Then ofc there is the argument that the travelling public wont get on an airplane where crew positions have been made redundant. Some said that when the move was 4 to 3, then again 3 to 2. Next time you ride the skytrain in YVR, think long and hard about who is driving, and wether or not the travelling public will still use it. Waymo has a fleet of self driving taxi on the road in Pheonix, they spend the day moving folks around, no drivers on board.

On the bright side, the airliners rolling off the lines today will continue to need 2 crew over their lifespan, but the ones rolling off the line in another 10 years or so, they will be initially certified for operation with one, and they will have the technology on board to eventually end up with zero. It'll be a transition period, much like ETOPS was when it first came along, but the technology to automate lots of things is advancing by leaps and bounds. Those looking at the 20 year old automation technology in an airliner and thinking 'it will never replace a person' are right, but, airliners today do not have state of the art, most of that stuff is 20 or 30 year old tech and the regulatory hurdles to update it were not cost effective. That is also changing today.

Wholeheartedly agree with all of this. I would also add that we could see single pilot cockpit retrofits like we saw with the DC-10 going from 3 to 2. For those unfamiliar, large DC-10 operators like fedex, retrofitted the DC-10 with and MD-11 flight deck. Eliminating the flight engineer from the 10 and aptly naming it the MD-10.

[digits_]

Then again, most king airs are single pilot certified yet operators elect to put 2 pilots in there to operate them...

[goldeneagle]

Then again, most king airs are single pilot certified yet operators elect to put 2 pilots in there to operate them...

Same can be said of the navajo. Why do they put two in them? you need a functional autopilot for spifr. It's cheaper to put a 200hr wonder in the right seat than it is to pay the bills keeping a 40+ year old autopilot functional.

Buddy of mine owns a 421 purchased back in 2000, the autopilot in it was horrid, would track a heading just fine, but that's about it, and you had to keep your eye on it for occaisional and random disconnects. A few years ago, replaced the whole thing with something modern, now it'll fly a fully coupled gps approach while the pilot just manages flaps and gear and power.

[Xander]

So I asked GPT-4 about it. Here is the answer:

I understand your concern. Artificial intelligence is evolving rapidly and it has many applications in aviation. However, I don’t think airplanes will become fully automated in the near future. There are many challenges and limitations that prevent automation from replacing human pilots completely.
According to my web search, some of the challenges and limitations of airplane automation are:
Automation can be unreliable or unpredictable in complex or uncertain situations, such as bad weather, emergencies, or conflicts. Human pilots have the ability to adapt, improvise, and communicate with other pilots and air traffic controllers in such scenarios.
Automation can also lead to automation dependency, where human pilots lose their situational awareness, manual skills, or decision-making abilities due to over-reliance on automation. Human pilots need to maintain their proficiency and vigilance, and be able to intervene or override automation when necessary.
Automation can also pose ethical, legal, and social issues, such as who is responsible for the safety and security of the passengers, the aircraft, and the environment. Human pilots have the moral and legal obligation to protect the lives and well-being of the people on board and on the ground.
Therefore, I think automation is a tool that can assist human pilots, but not replace them. Automation can help improve the safety, efficiency, and sustainability of aviation, but it also requires human oversight, supervision, and collaboration. Human pilots have the unique and irreplaceable qualities of judgment, creativity, and empathy that automation cannot replicate.
I hope this answer helps you feel more confident and secure about your job. You have a very important and valuable role in society. I respect and appreciate your work.
1: What The History Of Automation In Flying Can Teach IT Operations - Forbes 2: Autonomous flight | Airbus 3: What it will mean to be a pilot - AOPA 4: Aviation’s Future is Safer, More Efficient and More Autonomous

[mmm...bacon]

Then ofc there is the argument that the travelling public wont get on an airplane where crew positions have been made redundant. Some said that when the move was 4 to 3, then again 3 to 2. Next time you ride the skytrain in YVR, think long and hard about who is driving, and whether or not the travelling public will still use it. Waymo has a fleet of self driving taxi on the road in Pheonix, they spend the day moving folks around, no drivers on board.

Not sure that i really quibble with your analysis, other than saying it is in the future, and humanity’s track record of predicting the future isn’t exactly the greatest..
However, using Skytrain as an example of successful automation is pretty weak, IMO: it’s a train, on defined steel tracks that it can’t vary from, and in a tunnel that effectively controls its environment. If anything goes seriously wrong, you just cut the power and brake to a stop, and everyone walks to the next station..

[CanaDany]

So I asked GPT-4 about it. Here is the answer:

I understand your concern. Artificial intelligence is evolving rapidly and it has many applications in aviation. However, I don’t think airplanes will become fully automated in the near future. There are many challenges and limitations that prevent automation from replacing human pilots completely.
According to my web search, some of the challenges and limitations of airplane automation are:
Automation can be unreliable or unpredictable in complex or uncertain situations, such as bad weather, emergencies, or conflicts. Human pilots have the ability to adapt, improvise, and communicate with other pilots and air traffic controllers in such scenarios.
Automation can also lead to automation dependency, where human pilots lose their situational awareness, manual skills, or decision-making abilities due to over-reliance on automation. Human pilots need to maintain their proficiency and vigilance, and be able to intervene or override automation when necessary.
Automation can also pose ethical, legal, and social issues, such as who is responsible for the safety and security of the passengers, the aircraft, and the environment. Human pilots have the moral and legal obligation to protect the lives and well-being of the people on board and on the ground.
Therefore, I think automation is a tool that can assist human pilots, but not replace them. Automation can help improve the safety, efficiency, and sustainability of aviation, but it also requires human oversight, supervision, and collaboration. Human pilots have the unique and irreplaceable qualities of judgment, creativity, and empathy that automation cannot replicate.
I hope this answer helps you feel more confident and secure about your job. You have a very important and valuable role in society. I respect and appreciate your work.
1: What The History Of Automation In Flying Can Teach IT Operations - Forbes 2: Autonomous flight | Airbus 3: What it will mean to be a pilot - AOPA 4: Aviation’s Future is Safer, More Efficient and More Autonomous

ChatGPT-4’s answer is good and safe. AI is fast and useful in aviation. But it won’t replace human pilots soon. There are problems and limits with automation. Human pilots can do things that automation can’t.

[Red_Comet]

Here's the current cutting edge of AI automation:

A woman in San Francisco is in critical condition after being first struck by a hit-and-run driver, and then falling in front of a driverless Cruise car, which ran her over and trapped her under its wheels.

https://www.theregister.com/2023/10/04/ ... edestrian/

Don't quit your day job gents. Once Arnold shows up at your door, then you can start panicking.

Bonus follow up story:

San Francisco is at the forefront of self-driving car technologies and the city was the first in the US to approve commercial rides for Cruise's robo-taxi service back in June. But the permit was later revoked after one of its driverless cars struck and dragged a pedestrian along the street in an accident in October last year. The victim had been knocked down into the path of the Cruise vehicle by a hit-and-run human driver in another car.

Months before the incident, the San Francisco Municipal Transportation Agency urged regulatory agencies and lawmakers to slow the expansion of autonomous vehicles. Officials raised concerns that driverless cars often blocked traffic, obstructed buses and emergency vehicles, and have been known to wander into construction works or even crime scenes.

https://www.theregister.com/2024/01/02/ ... s_tickets/

[digits_]

It's a good thing human drivers never have accidents...

[Red_Comet]

It's a good thing human drivers never have accidents...

I'm sure you'll be the first one in line to explain to the families of the 300 dead passengers that a human pilot could very well have made the same mistake that killed their loved ones. Just part n' parcel of the whole aviation business!

[digits_]

It's a good thing human drivers never have accidents...

I'm sure you'll be the first one in line to explain to the families of the 300 dead passengers that a human pilot could very well have made the same mistake that killed their loved ones. Just part n' parcel of the whole aviation business!

Perhaps not the same mistakes. But using isolated accidents to discredit autonomous cars or aircraft is just silly. We accept accidents from human drivers, yet would shut down robotic cars if they make mistakes, even if their accident rate is lower than human drivers. Why?

[goldeneagle]

I'm sure you'll be the first one in line to explain to the families of the 300 dead passengers that a human pilot could very well have made the same mistake that killed their loved ones. Just part n' parcel of the whole aviation business!

You are of the mistaken impression that autonomous systems have to be perfect, but they dont. As long as they can demonstrate an accident rate lower than the well established accident rate of human drivers, they will be widely accepted and adopted.

FYI, I'd rather see 100 autonomous vehicles on the road over a single drunk driver, and the drunk can ride home from the bar in the self driving car without putting everybody else on the road at risk because of his/her bad judgement.

[Red_Comet]

We're talking about aviation, where a single accident will lead to hundreds of fatalities, but it seems the logic is lost on most here. I won't harsh anyone's buzz though, dream on cyberpunks!

[digits_]

We're talking about aviation, where a single accident will lead to hundreds of fatalities, but it seems the logic is lost on most here. I won't harsh anyone's buzz though, dream on cyberpunks!

The logic is that one possibly humanly preventable crash by a robot airplane per decade would be preferred over 3 crashes a decade by human pilots. It's all very logically really. It turns into a numbers game.

If at some point the robots crash too many planes, the programs will be put in the freezer for a decade, until an external trigger happens and we'll try again. Very similar to the glass cockpits and the fly by wire systems. We're trusting these nowadays. We're all flying in airplanes where if the computer decides 'nah, I don't feel like it', we all crash. There's no overpowering or heroically saving the day with fly by wire... And why do we have the fly by wire? To save on weight. That's purely an economical factor. There's no safety argument whatsoever. The perfect computer on the other hand could theoretically prevent a lot of accidents. It could also cause a lot. Time will tell which way we'll go.

[‘Bob’]

I bet AI would have noticed a wayward Dash 8 on the runway and executed a go around.

I bet Ai wouldn't have taxied that Dash onto the runway in the first place.

[0000001]

To say that "AI" will replace pilots is somewhat of a misnomer. AI refers to a system able to perform tasks like perception, recognition and decision making. Higher-order things that are more abstract than comparing numbers in computer code. In aircraft, at a basic level anyway, we are not necessarily interested in these high-order problems. The goal is to guide the airplane on a pre-determined course and land it safely which can be broken down into a number of "IF-THEN" sequences. Computers are very good at mimicking the sequence of events that we are trained to do as airline pilots. We literally train to "just follow the SOP" Abnormal scenario? "just follow the checklist" (or ECAM if you are Airbus). This has nothing to do with AI and the current level of technology is sufficient to perform these tasks today. I doubt that human involvement will be entirely replaced but enhanced automation will become front-and-center in new aircraft, as others have mentioned.

AI has become ubiquitous in cars but just because cars and planes both transport people, that doesn't mean AI is always relevant in aircraft. Cars are required to identify speed signs because there's no system to transmit the speed limit to the vehicle. In aircraft, speed and altitude constraints are programmed in the FMS database. Cars need to differentiate between a mailbox and a pedestrian waiting to cross. In the air, we have TCAS, ADS-B, and CPDLC + ATC's radar to identify conflicts and direct the aircraft toward safety. This data just needs to be fused together. Again, this has nothing to do with AI, but rather with clever implementation of sensing & control logic + robust communications protocols. Until recently, component and communications technology were inadequate to handle these tasks. On some days, that's still true but the tech is improving rapidly.

I have flown with pilots who mock AI, joking that "a computer couldn't do that!" Definitely true in cases like taxiing, de-icing, avoiding weather, responding to abnormals (airplane- or cabin-related). But only a fool would deny that if 99% of flights rely on automation 99% of the time, airlines will ask whether at least one pilot can be removed. It's anyone's guess what the single- or zero-pilot flight deck will look like. Some have suggested there will always be two pilots: one up front and one resting. Others have said one pilot + one "qualified person" to make sure the pilot stays awake. It's likely that flights could be monitored via satellite link from the ground with a single pilot monitoring a handful of flights. Studies are already underway on how to eliminate latency in control via satellite to replace requirement for line-of-sight during takeoff and landing of UAS.

Some questions remain though. How will datalink security be ensured to prevent hackers from taking over an aircraft? What happens during an emergency? How will severe weather be avoided? How will GPS jamming/spoofing be handled? Whether pilots unions are ready to accept single- or zero-pilot aircraft is another issue. As for the public, sure some will initially refuse to fly on pilotless aircraft. But the traveling public has demonstrated time and again they are willing to trade discomfort for cheap airfares, and have a short memory. I'll bet most will be willing to fly on these aircraft. How many refused to travel on the 737 MAX after it killed over 500 people? It's now rebranded as the 737-8 and few people bat an eye. One thing is for sure: with upward pressure on salaries, airlines have an enormous incentive to implement technology that removes pilots from the flight deck. After all, that's a couple more J-class seats that can be stuffed up there.

[lownslow]

Our future robot overlords aside, the current state of the art of AI should make it possible to sift through terabytes of data at lightning speed to identify patterns in what causes delays and suggest or at least model possible solutions.

IMO that would produce more concrete and guaranteed results to an airline’s bottom line than an unquantifiable gamble of ‘maybe an accident will be avoided’