Single-pilot airliners and AI (split from "Envoy captain stricken after takeoff and passes")

RIP captain.
This article came yesterday -- Airlines want you to get comfortable with flying without a co-pilot.
Will the remote-initiated auto landing be the answer for an emergency like this?

https://www.theregister.com/2022/11/21/pilot_single/

https://fortune.com/2022/11/21/airlines-pushing-one-pilot-in-cockpit-passenger-jets-instead-of-two/

 

RIP to the Captain, and condolences to his family during what must be a devastating time. I can’t even imagine.

Having said that, and with all due respect, this incident probably pushed back the “single pilot airliner” back a couple of decades.

 

I agree. With the amount of pax in the back of an airliner, I can’t see 121 going to single for a long time if ever. Part 135 on the other hand, better hope the pilot is healthy. Personally I think FAA should require the same medical (class 1) and retirement (65) for 135, as 121.

https://data.ntsb.gov/carol-repgen/api/Aviation/ReportMain/GenerateNewestReport/106209/pdf

 

Most likely scenario - within 10-20 years AI will take over with a single pilot being a backup for AI.

It is just a matter of accumulating/setting up proper aviation training data ( which is a hugely difficult task - Elon Musk with his Tesla autopilot has been at it for a decade accumulating training data from his ever growing fleet of customer vehicles)

Neural networks are just universal and almost infinitely complex function evaluators - in other words there is no fancy algorithm, nothing. The overall quality of AI response depends on how much properly formatted and scored data you feed thru the network and that’s the problem. We need properly formatted and appropriately scored data from hundreds of thousands of flights but the payoff is AI that basically consist of aggregated experience of all these thousand of pilots who participated in the program.
No single human will be able to compete with that once it is up and running.

PS This need for terabytes of training data is why this is no longer matter of a brilliant individual who comes up with some ingenious algorithm - companies that lead here are data aggregators like Google and Facebook who have access to unlimited amounts of training data provided to them by their users ( images, text etc ) - try Google Lens to see what their AI is capable of these days …

 

Re: the remark about Tesla’s Full Self Driving and the R&D involved….

it still has a very long way to go. I’ve tried it a few times in my small town and do not consider it safe.

Navigate on Autopilot and long freeway drives works well. But the self drive is very much like the drunken 15 year old discussed in the other thread.

 

Their latest is pretty remarkable … their software is progressing at very rapid pace.
Keep in mind this is fully autonomous mode - no GPS assist here ( except in terms of where you gonna go as in locations ) , just purely AI and sensors.

 

A plane is not a Tesla. My jet doesn't even have VNAV let alone LPV yet. I don't think you realize either the complexity of the aviation system nor the pace at which it evolves.

 

A plane operating in a highly controlled environment is actually an easier scenario for AI than a car operating in packed , close quarters and highly chaotic urban environment - that’s we have had autopilots for decades now while autonomous cars are only now becoming a thing …

 

You just proved my point. An autopilot is absolutely nothing like an autonomous car.

 

Autopilots are nothing like autonomous cars because almost all of them are algorithmically driven - precisely because of relatively low complexity that can be expressed by an algorithm.
My point was that while aviation is complicated and requires strict adherence to rules and overall high level of professionalism and structured approach , that very trait makes it more suitable for AI. Aviation is safe not because of insane skills of pilots but because of highly structured and controlled environment and overall high level ofmadherence to rules.

 

The way I see it, I think its closer to 50+ years based on a number of things. Maybe 121 cargo ops will see it a little sooner but not pax ops. Besides we’re there now as modern aircraft can basically fly and land themselves with current technology. Adding auto-take-off to the mix is just a matter of certification as Airbus has already been flying such technology.

The main reason for a 2 crew cockpit is the complexity of the aircraft especially during a failure event, i.e., the need for 2 heads and 4 hands to fly it. So until you can teach the AI what to do during a system failure, to include a failure of its single pilot, one will never get it certified for pax ops in my opinion. In other words, any major failure needs to end up like Sully in the Hudson vs Ethiopia 801.

But the main push behind single pilot ops is more economically driven by various operators than any other reason. Air carrier growth projections have continuously outstripped pilot growth for a number of years. And with covids affect on those numbers, there’s pressure being applied politically to the state regulators who in turn have recently pushed on ICAO to pursue this. Without ICAO writing the guidance there’s no way forward at all.

Regardless, based on my experience with airborne technologies I will never get on a single pilot/AI airliner. I’ll always take my chances on those 2 carbon computers up front and hope they’ve been trained properly. Besides, just thinking of the single pilot/AI combination gives me quasi-flashbacks….

Single Pilot: Let me land the airplane, HAL.
AI Control: I’m sorry Dave, I’m afraid I can’t let you do that.

Not quite. Pilot skill set is the main foundation to safety. Its been proven too many times. Operationally, once the structured/controlled environment crumbles and becomes chaotic the only thing left are those skill sets. Unfortunately not all pilots have the same skill set.

 

You proved my point again. Autopilots are not like autonomous cars because they are not autonomous. You are way overestimating how much work the autopilot does during a flight and way underestimating how much work the human pilots do. The autopilot does the boring and menial tasks so the humans can do the complex ones.

 

This is the third world / airbus theory, and it works very well as long as the failures are ones that are trained for, normally to rote

 

Add to that the technology advancements, extreme altitudes above most birds, other planes and often even weather where airliners spend most of their time, the close controller eye in the B and C and D airspace where airliners transition down, long runways, and so on


As conditioned as many people are, the rules help, but the major advancement in safety didn’t come from the FAA as much as engineers, inventors and marketing and insurance people

 

I guess having a discussion with commercial pilots about how their jobs will eventually be replaced ( or rather heavily supplemented/reduced ) by a AI/computer is somewhat awkward and counterproductive …

I am a computer engineer and fly planes strictly for fun so ultimately I have no dog in this fight …

 

You know nothing of regulatory timelines. Part 23 re-write has been languishing for how long now? Long enough yo turn into MOSAIC.

The lead time for ADS-B compliance was a decade or so.

 

True that - I was thinking in terms of actual technological progress without factoring in regulatory timelines.
As usual, military applications will lead the way …

 

Do you have the comparison in number of midairs prior and post ADSB mandate?

 

I guess having a discussion with a computer programmer that knows nothing about flying commercial aircraft is kinda awkward. I mean any time we say anything that diminishes the AI is not based on facts. Just fear of losing a job.

You are taking an easy way out of the conversation.

 

Ha. Except I'm not a pilot. Only the guy who has to fix the systems you engineer types dream up. The AI/chip part is not the problem its all the peripherals, sensors, transmitters, wires, etc. connected to them. And based on what I've seen and the hours I've spent working with tech support/engineers to fix problems I don't see single pilot/AI aircraft at the 121 level for decades. While it might work on your computer screen, in real world conditions they tend to fall short.

 

Nope. It’s false logic to assume ADS-B would reduce mid-air collisions. Not even remotely a part of the system requirements.

“There is no specific authorization to use ADS-B In for traffic situational awareness.”

https://www.faa.gov/documentLibrary/media/Advisory_Circular/AC_90-114B.pdf#page4

And before you go there, neither CAVS nor the Code relieve see and avoid responsibilities in VMC.

 

I will be long gone from the airlines even if your fantasy becomes a reality in the time frame you stated. This is not an example of people protecting their jobs, at least not in my case. It's an example of the Dunning Kruger effect. You have absolutely no idea what goes on in an airline cockpit, yet you already know how to automate it.

 

You are missing the point … I won’t be automating it - you and thousands others like you will be by having your actions recorded , normalized and used to train the network.
No matter how successful and experienced you are , as an individual you are only exposed to a small subset of the total sum of learning experience aggregate that is being produced out there every day.
The AI will have the benefit of having “experienced” aggregate of every commercial flight in , say , last 10 years - literally millions of hours.
Of course, that’s the ultimate goal and we are far away from it purely on the basis of not even collecting all that necessary data but eventually it will become a reality.

 

It doesn’t work that way

 

We are far away from it, but it's happening within 10 years. Seems like backpedaling to me. My original point, which you missed, is that 10 years is a blink of an eye on aviation. While new models of cars debut every couple years new aicraft are a once in a generation event, or less.

 

Even with all those hours I don’t think it will work. I think a real AI that is self aware will be required to replace people in the cockpit. All that aggregate experience doesn’t matter when the airplane really goes sideways. Independent creative thought and decision making is necessary.

Furthermore it will have to be an AI that is self contained and totally independent of any system on the aircraft.

Anything less will not be adequate. You saying that here in 20 years??

In regards to the military reference that doesn’t transfer all that well. If an autonomous drone fails and smokes a hole it’s not good. There may be mission consequences but there are not 200 people in that hole with the drone.

 

I am thinking military applications within next 10 years or so.

I don’t have any direct knowledge but I believe there is already an intense race underway to produce a pilotless military aircraft - whoever gets there first will have the advantage of being able to utilize flying contraptions that can routinely pull 12+ Gs plus host of other benefits.

 

So you'll be fine when AI goes and flies your plane strictly for fun because computers are better then people? Maybe it can just bring you back that $100 hamburger and some GoPro footage.
Some of us who fly for a living do it because we have passion for it. I'm sure as hell not doing it for the paycheck.

Do we even know the age of the pilot? Lots of assumptions about bad heart= old.

 

And I thought this whole discussion has been about airliners. More backpedaling.

 

No, of course not - I was talking about commercial applications - people still ride their bikes purely for fun and will keep flying for the same reason.

Btw. I am not working on any of that stuff, I just know people who do and generally try to keep myself up to date.

… and finally, eventually my job will also be on the line …
https://techmonitor.ai/technology/ai-and-automation/deepmind-alphacode-ai-software-developer

 

I’m not saying you are wrong. In fact I agree with you but 10-20 years is a pipe dream.

 

Your judgement is derivative of your actions - it is combination of judgement you have inherited from actions of others ( your initial learning process ) and the continuous feedback loop where you learn from your own actions what works and what does not - also known as acquiring experience.

AI works the same way - it needs both the actual action and some kind of normalized score for that action.

Anyway, I am not trying to sell you anything or change your opinion on viability of AI so I will end here - it will surely be interesting to watch.

 

AI doesn't work the same way.

https://www.nature.com/articles/d41586-019-03013-5

Excerpt:
A self-driving car approaches a stop sign, but instead of slowing down, it accelerates into the busy intersection. An accident report later reveals that four small rectangles had been stuck to the face of the sign. These fooled the car’s onboard artificial intelligence (AI) into misreading the word ‘stop’ as ‘speed limit 45’.

Such an event hasn’t actually happened, but the potential for sabotaging AI is very real. Researchers have already demonstrated how to fool an AI system into misreading a stop sign, by carefully positioning stickers on it1. They have deceived facial-recognition systems by sticking a printed pattern on glasses or hats. And they have tricked speech-recognition systems into hearing phantom phrases by inserting patterns of white noise in the audio.

These are just some examples of how easy it is to break the leading pattern-recognition technology in AI, known as deep neural networks (DNNs). These have proved incredibly successful at correctly classifying all kinds of input, including images, speech and data on consumer preferences. They are part of daily life, running everything from automated telephone systems to user recommendations on the streaming service Netflix. Yet making alterations to inputs — in the form of tiny changes that are typically imperceptible to humans — can flummox the best neural networks around.

These problems are more concerning than idiosyncratic quirks in a not-quite-perfect technology, says Dan Hendrycks, a PhD student in computer science at the University of California, Berkeley. Like many scientists, he has come to see them as the most striking illustration that DNNs are fundamentally brittle: brilliant at what they do until, taken into unfamiliar territory, they break in unpredictable ways.

 

A car has a steering wheel, brake, and gas. So much more complex than an airplane! Amazing humans can even drive one.

 

Public acceptance might also be considered. As a non-pilot member of the flying public, I can't say for sure how many years of service it would take for me to consider flying in an AI controlled plane, but it would probably be measured in decades. I doubt I'm alone in that.

 

Well it looks like your timeframe just got bumped back a bit. Latest guidance from the FAA recommends more manual flying and less use of automation in commercial ops. Seems they place equal importance on individual pilot skills as they do technology.
https://www.faa.gov/documentLibrary/media/Advisory_Circular/AC_120-123.pdf

 

Outside of a FAR no one really cares what the jokers at the FAA think

However at work I have flown some single pilot planes and we have tossed another pilot in because the pax have the fetish fantasy about a pilot dying, if they want to pay for a pointless second pilot who are we to stop them.

I don’t think the facts matter too much with the masses, and I don’t think a single pilot airliner is marketable, a no pilot airliner concept is DOA

 

135 requires a first class in some cases.

 

Technology, ......ology! AI? Phoey! Let's talk today and real world.
Has anybody had an actual incident of inflight incapability? Thats where this thread should go. I'll toss in an actual experience to begin.
Part 135, two pilot crew. My co-pilot passed a gall stone while over water and one hour from land. He had a recent Cl 2 medical.
Next?

 

There’s also the human side of the equation…how is this one pilot whose job it is to keep an unwavering watch on the AI pilot going to acquire the knowledge and skills to adequately monitor and/or fly should it become necessary?