Then there is the case of UPS Airlines Flight 6 where the final act of the co-pilot was to intentionally crash in a a deserted piece of ground between residential areas.
I haven't seen anything to indicate this was intentional as opposed to a "happy" accident. Furthermore, an important contributing cause of that crash was that the flight deck air conditioning system failed, which allowed smoke and fumes to enter the cockpit and incapacitated the crew. Had that system - which had been reported faulty on a previous occasion - worked, the crew would likely have been able to make a controlled emergency landing. So, once again, this comes down to human error/negligence, although this time on the part of the airline and/or its maintenance crew.
In a lot of other cases the errors reported to the cockpit does not indicate the actual failure experienced, El Al Flight 1862 got multiple engine fires warnings while in actual fact two engines dropped of the aircraft damaging the wing on the aircraft. This lead them to follow the wrong checklist which was responsible for them hitting an apartment complex something an auto pilot will do as well.
Garbage in, garbage out. If you have incorrect information, you're probably going to make incorrect decisions. Had the instruments (sensors) been more useful and indicated the actual error, a human or AI pilot would have been able to make a better decision that could have saved the plane.
You then also have a case where they developed the checklist by winging it ie. United Airlines Flight 232 which was then later instrumental in saving other aircraft like DHL Express Iraq Nov 2003.
DC-10 design meeting:
Engineer #1: We need to make the control systems redundant so that the plane is still flyable should something fail. My proposal is 3 redundant hydraulic systems.
Engineer #2: AND LETS PUT AN ENGINE ON THE TAIL, AND THEN ROUTE ALL THE REDUNDANT HYDRAULIC SYSTEMS NEXT TO THAT ENGINE, AND THEN NOT INSTALL FUSES TO PREVENT FLUID LEAKAGE IF THAT ENGINE GOES BOOM DURING FLIGHT AND PUNCTURES ALL 3 SYSTEMS, THUS RENDERING THE "REDUNDANCY" USELESS
Engineer #1: This sounds like the worst f**king idea in history, people are going to die!
Boss: I agree. Engineer #2, what do you say?
Engineer #2: DOING IT THIS WAY WILL SAVE US 2 CENTS PER AIRCRAFT
Boss: Design approved, build it.
No human or AI pilot could have saved the DC-10 from being a terrible design. Regardless of that, however, any sufficiently competent AI would have come to the same conclusion as Flight 232's flight crew, namely that the only control mechanism remaining was the engines, and would've utilised them accordingly. In fact McDonnell-Douglas's very own F-15 STOL/MTD had an autopilot capable of exactly this in the late 1980s, and ironically the same plane is being used to train the
Intelligent Flight Control System neural network that is intended to be capable of dealing with situations outside the norm of aircraft operation.
Pilot error is a significant chunk of crashes that I agree with but training and proper management can reduce this significantly it just that the airlines ignores this because it costs money. The amount of crashes due to the pilots being to tired or pressed for time is ridiculous. The worst aviation disaster to date happened due to a strict deadline and a radio static error.
Which is exactly what AI pilots avoid. They don't get tired, they don't get distracted, they don't need bathroom breaks. Remove the pilot from the equation and you have one less thing to fail or blame - because software is formally verifiable, it can be tested in an error scenario, and if it does the right thing, it either wasn't the software at fault, or the situation was completely unrecoverable.
