Comment by stephen_g
3 days ago
It’s a pointless exercise though - if one of the pilots wants to crash the plane, there’s almost nothing that can possibly be done. Only if someone can physically restrain them and remove them from the controls.
There’s always going to be many ways they could crash the plane, such a feature wouldn’t help. The pilots are the only people you can’t avoid fully trusting on the plane.
It's only pointless if we assume crashing was the intended result of the pilot. If the switches failed, or the pilot activated the switches by mistake, it's worth considering options for handling the inputs.
There's a balance of accidents to be found, I think. There are likely cases where fuel does need to be cut off to both engines, and preventing that would lead to accidents that might have been recoverable. This case shows that cutting off fuel to both engines during takeoff is likely unrecoverable. There have been cases where fuel is cutoff to the wrong engine, leading to accidents. Status quo might be the right answer, too.
So basically we need software that can 100% autonomously fly a plane. Software that is extremely reliable and trustworthy, basically. Software with multiple fallback options. Multiple AI agents verifying every action this software takes. Plus, ground-based teams monitoring the agents and the autonomous flight software.
Not AI, AI is less trustworthy than normal software almost by definition.
Formally verified traditional algorithms.