It would be like if someone asked "what would happen if a C program tried to access uninitiated memory?" And the answer was "Good programmers would never do that".
With that caveat, every autopilot system I've studied turns itself off if the plane is outside its flight regime. So for example, if you pull a plane straight up and do a tail stall and try to turn on the autopilot, the autopilot would refuse to engage as the plane isn't in a "flyable" regime.
What is more, the autopilots I've seen were explicitly programmed to be pilot activated, so even if the plane went from one regime where the autopilot kicked itself off automatically, back to one where the plane could be flown by the autopilot, the autopilot would not automatically re-engage.
If the pilot ejects because they cannot control the plane, neither can the autopilot so it won't be on. If the pilot put the plane on autopilot and then ejected, the ejection would most likely force the autopilot to disengage. If a pilot put a plane on autopilot, walked into the back of the plane and opened a door and parachuted out, the plane would continue flying on autopilot until it ran out of gas and then the autopilot would disengage and the plane would crash.
(by the way Chuck, your important section on your profile is really impressive, I've never seen anything like it but it makes total sense).
So, yeah. I think he means it :-) I was pleasantly surprised when I saw that username pop up here a few years ago.
Auto-GCAS Saves Unconscious F-16 Pilot: https://www.youtube.com/watch?v=WkZGL7RQBVw
From memory it was introduced in the Sukhoi Su-7 in the late 1950s.
https://www.wsj.com/articles/SB10001424052702304791204576402...
Liability as well.
It would be a fair question to wonder what GCAS does when a pilotless plane is heading into the ground. I'm going to guess that the folks who built GCAS have a test to see if there is a pilot in the plane or not. And in the absence of a pilot just augur in, but that is just a guess.
Also, ejection only makes sense in the context of a manned military aircraft. For the question to make sense as posed, the implication is that the pilot had engaged the autopilot and then ejected while the autopilot had meaningful control authority over the aircraft. That's not a scenario the airworthiness certification authorities probably consider as critical. Manned aircraft require deliberate human action to engage the autopilot. What the flight control computers do in general (e.g., limit control inputs to keep aircraft within acceleration/maneuvering limits, automatically enter a stable attitude when stick forces are not present) is not considered the "autopilot".
You could ask the question, "If the pilot ejects, do flight control computers attempt to fly the aircraft in any particular way?" At least for U.S. aircraft, recent ejections (e.g. the Thunderbird ejection near the end of Obama's tenure), suggest the answer is "no". If an automated flight termination system is installed, when safety constraints are violated (e.g., heartbeat signal or veering outside of geographic limits) it usually takes action to put the aircraft into the ground as quickly as possible (e.g., cut power to actuators on an unstable aircraft). This is common with UAVs. I don't think this is the case with manned aircraft.
Them: What if a shark jumped into a volcano?
Me: That wouldn't happen because sharks live in the ocean and can't jump high enough or far enough to reach land.
Them: OK, but what if?
Me: ... It would get burned and die...
So yes, if a pilot ejects and the plane had auto-pilot on and the auto-pilot was programmed to try to land after a pilot ejection and the plane was still safe to fly, the plane would in fact attempt to land.
https://en.wikipedia.org/wiki/Autopilot#Autopilot_for_ILS_la...
https://en.wikipedia.org/wiki/Instrument_landing_system#ILS_...
> Everything within roughly a mile of the park is leveled, and a firestorm engulfs the surrounding city. The baseball diamond is now a sizable crater, centered a few hundred feet behind the former location of the backstop.
> A careful reading of official Major League Baseball Rule 6.08(b) suggests that in this situation, the batter would be considered "hit by pitch", and would be eligible to advance to first base.
Ejected vs. not-ejected is very likely not even an input variable to the autopilot. So the answer would be:
Autopilot will attempt to do exactly what it was doing before the pilot ejected.
Also, the plane would be lighter after the ejection.
And perhaps—I'm just guessing—the cockpit window or ceiling might now be gone? In which case there'd suddenly be a lot of drag from air swirling down into the cockpit.
So no, I don't expect the autopilot would continue to do exactly what it was doing before. It'd have to do a bunch of corrections, and then it would return to what it was doing before :)
(Or those corrections might be enough to make it decide "the pilot" should take over. What happens when there's no pilot and no autopilot? I'm guessing that's down to the way avionics hardware reads out when nobody's touching it.)
The cockpit can be broken or sensors may be malfunctioning as well, so "cockpit missing" is not a great signal.
It might not generalize to any other compiler—and that's the reason you don't try to build anything predicated on any particular assumption about the behavior—but you can still know it for a particular context.
And, in fact, often there are conventions in compiler-writing that mean that every compiler basically does the same thing in response to certain UB code. (Like optimizing away always-false branches, for example.)
So, by analogy: there are a limited number of autopilot AI programs in existence. Of those concrete implementations, is there anything they would be predicted to do in response to this event they're not coded for? What would it look like to them—what branch would they take without an explicit branch for "ejected"—and how would they react to being in that branch?