Cockpit design is still BS in my opinion. Look where they place the engine power indications in the year 2025. Should be right above the PFD.

Just taking this one comment, I see several major approaches:

1. Better detection of GPS spoofing. Current aviation GPS receivers appear to have been programmed with essentially no thought of a deliberately wrong GPS signal, only loss of signal or corrupted signals.
   1. I don't think you can spoof a signal so that a three-or-more-antenna differential receiver could see the appropriate distance and angles between its different receive antennas. A signal that implies all three antennas are colocated or in a straight line is clearly false if they're located on e.g. a roughly 20m triangle above cockpit and engines.
2. Cross-checking GPS signal against other navigation and measurement sources - baro vs GPS vs radio altimeters, onboard time against GPS time, NDBs/IRSs/GPS positions.
   1. There is a limited boil-the-frog hazard if the spoofing is targeted at one particular aircraft, and the aircraft is using GPS to discipline/update the IRSs, but that's pretty marginal and you should see NDB/other navaids being wildly out of position. You should also be able to determine that e.g. it's odd that the IRS error is always right of track...
3. Reject sudden changes in position/time/etc. Once a GPS lock is received on aircraft powerup at the stand (in a known location), I see no good reason why it would be lost and certainly no reason for the time base to shift more than a millisecond.
4. Terrain radar and terrain-following radar/navigation. I don't think anyone's built low-observability mountains yet.

Obviously you would want to do as many of the above as possible.

It wouldn't so much be ignoring the EGPWS (apart from perhaps #4) as rejecting the false location signal being fed to EGPWS in the first place.

Ah, reminds me of a great film - 2001: A Space Odyssey - When Dave tells the Aircraft to open the pod bay doors, HAL responds “I'm afraid I can't do that, Dave.” Just shift a few words and you get the idea lol ;-)

On paper, in theory, all wonderful.

An aircraft with hundreds of other people’s loved ones is a tough place to automate reliably enough to remove the human - as fallible as we are.

Yes, a lot of engineering, review, testing, more engineering, more review, more testing will be needed. I don't think anyone's disputing that.

But the day will also come eventually; the question is when not if.

Regarding the above opinion on GPS; that's almost all stuff that can & should be implemented on a fully piloted aircraft anyway.

As for whether automation to remove the human in life safety situations can be acceptable... 500-pax trains have been driverless in some places for a few decades, and elevators have been unmanned for around a century. Aviation is different, yes. But not unfathomably so.

[Retire Software Engineer] Both trains and elevators can just be stopped. A plane needs to continue to fly to some destination while avoiding the scenery, etc. A major difference.

Over-automation is harmful, and it was one of the contributing factors in the aforementioned disasters. Autopilot was originally designed to maintain altitude and heading—to simply assist pilots. But we've now gone too far as pilots have very few legal or practical opportunities to hand-fly the aircraft, which is a fundamental skill that separates a pilot from a non-pilot.

Let’s say we eventually make airplanes fully automatic. At that point, human error doesn’t disappear—it just shifts from pilots to software engineers. So why not draw a line and limit further automation? We could introduce a legal requirement for pilots to hand-fly aircraft for, say, 10–15 hours a month to stay current.

Pilots sometimes do not set a very high bar when faced with unanticipated and usual conditions. Air France 447 and Aero Peru 603 are striking examples of how pilots failed to control aircraft that were still controllable.

Perhaps we should lower expectations for a fully automated aircraft. Lowering probability of catastrophic failure from 1e-9 to 1e-7 would make the engineering task far simpler and I doubt it would kill many more passengers than pilots who find themselves outside their zone of competency in aircraft that meet current certification requirements.

So this is exactly my thinking before this thread. It seemed obvious to me that pilots weren't getting enough hands on and the answer would be to increase the hands-on.

But I think lots of things are changing..

At the rate of technology growth, I think it would be easy to postulate fully automatic 100 years from now.

In reality, with machine learning and AI, entirely different systems and processes that are far more efficient and have less failure opportunities would most likely be available far earlier than that.

As others have pointed out, there's lots of ground and and supporting structure that would have to be entirely updated. But that would be the point of total automation... Pretty much replacing everything that is currently available.

Add due to my technology background, my guess is we'll probably see a relatively organic and rapid change when the time comes. It's easy to imagine small freight opportunities where an entirely parallel system might be economically viable.

And once that work is done and once the ROI and safety is proven, the rest of the system would surely follow.

Of course in the short term, additional air time certainly could be useful.

The formulation of the problem is often more essential than the solution.
 

'Thinking that automation is the answer'… but what is the question … what is the problem issue.

The accidents cited and other events are history. Assuming that appropriate safety interventions are applied (not always a good assumption) then these situations and / or outcomes should not be repeated.

An important question is how best to maintain safety, avoid future accidents, particularly in a safe industry where the reasons for this safety level might not be understood or the actions repeatable.

Taking a wide, long term view:
https://accidentstats.airbus.com/wp-...2025-links.pdf
… over recent years the industry's level of 'safety' has been flat-lining, perhaps indicating this is 'as good as it gets' using existing methods.
This may not be just the 'generation' of aircraft, not attributing improvement to technology alone (nor automation per se), but wider aspects of how these technologies are being taught, the education, understanding how to use them, and to operate these improvements in fast evolving operation situations benefiting from greatly improved ATC, comms, information, and infrastructure.

"If I were you, then I would not start from here."

Also see https://www.pprune.org/accidents-clo...l#post11864502

"The formulation of the problem is often more essential than the solution". A Einstein

The counter argument is that pilots continue to make the same mistakes again and again while automation can eliminate errors as it matures. Safety critical software has a reliability that human operators and mechanical systems can only dream of.

A difficulty with this use of machine learning requires adverse outcomes to learn from; whereas the safety objective remains the avoidance of adverse outcomes (error). We might improve safety by being more unsafe!

Machine learning might help by assessing everyday operations - the overwhelming number of successful outcomes.
Unfortunately it is very difficult to identify the process of success because much of the human involvement involves tacit knowledge and intuitive expertise - based on prior experience.

The issue (wicked problem) is how to provide pilots with experience of adverse events when in a safe industry there are fewer events.
Simulation; but this is only a model of reality and similar to 'AI' has no knowledge of the unknown, unforeseeable, surprising events. You might be surprised in a simulator - "I didn't expect that at this time", but not the extreme fundamentally surprising event beyond comprehension - "I don't believe it, this cannot be happening".

An emerging use of technology is 'IA' Intelligent Assistance particularly in aiding situation awareness.
Intelligent from the users perspective, and Assistance according to the situation.
An excellent example of IA in current operations is EGPWS/TAWS. Consider expanding that concept to other operations with machine learning - being able to identify how close to the edge of a safety boundary normal operations are; … then automate the recovery.

However :
"If a machine is expected to be infallible, it cannot be intelligent" Alan Turing

You're absolutely right.

I never meant to imply that AI automation would improve itself. I meant that non AI automation would be re-engineered to prevent it repeating incorrect behaviour.

If full automation is achieved in our lifetimes it won't be by giving a clever computer a camera to see out of the window and teaching it about ATC, EGPWS, TAWS and orphanages.

It'll be by coming up with reliable, 'machine-friendly' systems for managing navigation, maneuvering, separation, loss of power etc.

Spot on..
This topic is being continuously hammered, and for good reason.
Luckily, I work for an outfit that requires pilots to be equally proficient in all levels of automation, and that includes basic manual= raw data.

And being proficient means regular practice, and regular in my book does not mean once every 6 month during sim sessions. That does not equal proficiency, at least not for me. Obviously, a professional pilot needs to be fully in tune with the auto flight system, and intimately understand it. But once that is ticked off, then said pilot should also regularly practice basic flying. Why, some may ask?

First of all, for the satisfaction and joy of fully being in control of the aircraft using your flying skills, and honing them on a regular basis in real life.
That instills confidence with the pilot to not hesitate to reduce the level of automation, should a situation arise, which requires the pilot to take manual control.

If thoroughly briefed, so the PM is stepping up the monitoring, then I see no additional major risk with raw data flying, when conditions are favourable. And it is no real issue to switch on FD / AP/ AT etc should the PF start to feel a bit overwhelmed.

And for reference, being a long time TRI/TRE, watching trainees in the sim, I can clearly see the difference in performance between pilots who regularly hand fly, and those who do not. And that difference in performance applies not only to them flying manually, but also how they handle the automatics.

I make it a point, during LIFUS with new cadets, to demonstrate that an A320 is in the end "just" an airplane, and the laws of physics applies equally to the bus as they do to a C-172, a Diamond or whatever.
So I do the approach announcement to the cabin at around 10000 ft (shorthaul), then I disconnect everything (including the bird) and raw data the A320 all the way. And it often really open the trainees eyes to the fact that this sophisticated machine can fly just fine, without all the fancy whistles and bells (which for the record are great and have their place).

I strongly believe that the greatest safety feature of any airliner are well trained pilots, who have the competence, and CONFIDENCE, to control the aircraft without any automation available, and that takes regular practice, period.
And should I ever end up in ELEC.EMER.CONFIG, the last thing I would want to worry about is controlling the aircraft. That should be second nature, so that I can focus on solving the issue at hand.
Rant over, Rgds.

I'll let you in on a secret, but you have to promise not to tell anyone. Shepards Freedom 7, Gemini and Apollo spaceships were too fully automated. Space Logic is simple - go up or go down.

They were automated, but not 'fully automated' - the crew could manually override at any time (which happened several times).

As one lunar astronaut was asked by a TV watcher on the way to moon, “who is driving now”? “Issac Newton”, was the reply. Jim Lovell did the trajectory corrections approaching Earth on Apollo 13 manually with the Mk. 1 eyeball.

This is a very good summary of where we are now. And there does seem to be near complete consensus among professionals that they currently need more hands on time as opposed to less. So your entire post makes complete sense.

However, the point of this thread was to talk about something different. The bottom line question is "at what point will risk and cost be reduced to the point that total automation is significantly better than a human, and the human's only roll is take over if the automation completely fails."

Obviously it isn't now. So restating the question, 'When is total for automation likely and what needs to happen before it is possible?

I think a lot of that has been covered, but still interested in case anyone has other ideas.

SpaceX are doing well in the business sector they are in. Without knowing quite what they are measuring reliability seems to be a tad over 99% per flight. But this seems a long way from the reliability of airliners in the last few decades.

https://accidentstats.airbus.com/hull-losses/
In 2024, all aircraft generations have a combined fatal accident rate of 0.31 hull losses per million flight cycles.

Yes.
And also the scale of the PFD itself is not really satisfactory. It should be, in my opinion, more zoomed in, or the PFD bigger.
I even chatted with a test pilot about varying scale. For precise maneuvers, zoom it. For large pitch changes, unzoom it. However it could cause disorientation. In some cases it would help precision. A bad choice overall but it tries to tackle a real problem.

Also, the sensitivity of thrust levers is often too much. The entire range of the throttle/thrust levers is a few centimeters.
On airbus, the notches (CLB to TOGA) take up half the space whereas they do not increase thrust so much (and do not increase it at all above a certain FL)
And the space between idle and climb, so between descent and something not allowing the approach to continue, is only 5 cm.

Same goes for the sidestick. There is not enough precision right in the middle.

On boeing airplane you don't have the inertial part of the variometer (I think) so it's much less useful than airbus.
And the out of trim forces are super heavy.

Everything is debatable but I think they do not do as much as they could to improve the ease of manual flying.
Airbus design is terrible and played a huge role in AF447.
​​​​​​​
Also the second part of your message was just not relevant at all.