[CarbonCycles]

Adding a bit more context to this good post, the control algorithms are designed and developed around assumptions and an operational envelope deemed to be "normal".

An airplane is much more complicated from a systems perspective (but yet can be more easily automated from an operational POV). To make an airplane fly, there are dozens of control surfaces, very complex mechanical systems, and not to mention a greater degree of danger (i.e., explosions) and loss of life should these systems fail. Yet, when everything "works" within their engineered/designed parameters, we can construct control algorithms around these operational envelopes.

Also, at 10k feet in the air, there are rules that govern flight paths (overseen by controllers); radar and beacons that help keep planes on track; and most importantly much less traffic/congestion (as opposed to a bus or car).

This provides the pilots (and also sailors) the ability to automate their flight at steady-state. This is where it gets interesting. If a plane loses a control surface or a critical system fails, these controls systems will be pushed outside of their operational envelopes, and this is where a human takes over.

It's not a question of "if" but "when" things will fail....

[LeifCarrotson]

The low-hanging fruit of straight and level flight is, I'd argue, appropriate for automation, but they need to get out of the way when the plane goes outside the normal operating envelope.

Whether these tools are just analog trim tabs that allow the stick to be adjusted so that a stable aircraft basically holds itself in straight-and-level flight, or complex fly-by-wire autopilot algorithms that keep an unstable aircraft in straight-and-level flight, taking the mundane, repetitive, mind-numbing workload of closing the loop between the six-pack of instruments and the sticks for hours on end. A pilot who has been watching the VSI and heading for three hours on a cross-country flight, tediously keeping the plane within 300 feet of its assigned flight level and ensuring that the passengers don't even perceive tiny changes in heading and attitude, or a trucker who has been driving their tractor-trailer cross-country for 60 hours in 7 days, holding an 8.5-foot-wide vehicle between a pair of 12-foot-wide lane lines, and adjusting the throttle up and down hills to keep their speed within a couple MPH of the speed limit, is going to be fatigued and when it comes time to deal with an emergency on landing or a sudden, unexpected brake check from a small car pulling in front of the truck, that fatigue has a cost.

There's a point where cruise control and autopilots can make these tasks safer, allowing pilots to keep ahead of the needs of aircraft, and allowing drivers to keep their eyes scanning the other vehicles, their mirrors, and the road ahead, rather than going back-and-forth between the speedometer and the lane lines. I absolutely acknowledge that there's also a point where this automation gets in the way or causes operators to tune out and be unprepared when asked to take over when the envelope is exceeded. Right-sizing that level of automation is critical!

[CarbonCycles]

This topic reminds me of an observation I made a while ago. I have found many folks (esp. non-technical) confuse automation with intelligence (more-so Artificial Intelligence).

Automation ==/== AI. Automation consists of a series of systems that are designed (i.e., programmed using control algorithms, logic rules, machine/deep-learning) to operate under a series of pre-set boundaries. The boundaries are typically defined under "normal" operating conditions, and safe-guards are put into place once the automated systems deviate from those "normal" operating conditions (I see a lot of practitioners esp. in the ML world fail to account for abnormal conditions...may be this is what happened w/ the Uber driver?).

The only comment I will say about AI is that I'm glad I am seeing RL and the like come more into play. I also contend this feels like a supervisory problem.