There are a number of situations when the landing gear should either be raised later than it usually is or not at all. The pilot commanding the flight must take many things into consideration before deciding to raise the gear in an abnormal situation. Designing an automated system with imbedded decision-making logic to perform such a critical task as deciding when to raise the landing gear may be much more fraught with risk and uncertainty than it first appears.
For the simple and foreseeable situations it may be possible to write some pretty robust software capable of making correct decisions most of the time. But what about situations with complex variables? Who will be able to correctly decide ahead of time what the correct action might be if they don't know what combination of variable factors will be in play in every possible future situation? That's the nature of real-time critical decision-making. Pilots must weigh the consequences of each action and prioritize actions and outcomes accordingly. They don't always get it right but I'd still trust a highly trained and well experienced professional pilot over a software engineer to make the critical aeronautical decisions that affect me any day.
A couple of situations where the decision may be more complex than I'd trust an automated decision-maker to manage:
Tire failure during the takeoff roll at high speed. We're going, but should we raise the gear at positive rate? Maybe, maybe not. A prior risk assessment based upon statistical outcomes could make a case for leaving the gear down to avoid gear damage or possibly being unable to extend it again. However if the tire debris went in an engine, getting the gear up may be critical to achieving required climb performance which is likely worth the risk of damaging the gear or being unable to extend it again. We can live through a belly landing but not a collision with terrain.
What if the tower reported fire appeared to be coming from the right landing gear wheels as we were rotating? Would the computer know what to do? How?
The human capability to assess and prioritize actions, reactions and potential consequences in real-time may be impossible to replace with a computer, AI or not. Let's let computers continue to do what they do best: Crunch numbers and perform repetitive tasks at the behest of the human decision-makers responsible for outcomes. In any case, who is going to trust their life to a software writer who isn't there to suffer the consequences of a wrong decision? Not me...
Maybe a better use of effort would be to design an automated engineer that designs only what it's told to?
[/sarcasm]
westhawk