I could not feel comfortable as a passenger on a commercial airliner without a human on board that could take over and fly the plane manually. There are all kinds of failure scenarios where a computer, AI or not, would get confused. Even just the specter of malware is enough for me to expect a human being, that values their own life, is able to take over.

>Even just the specter of malware is enough for me to expect a human being, that values their own life, is able to take over.

I'd be more concerned abouta human going insane than malware.

I'd imagine they're mostly fly by wire anyway, so a virus could theoretically just disable the human input.

Airbus designs are substantially fly by wire, although supposedly with an isolated control system. Boeing designs are isolated electro-hydraulic.

There's a big difference between isolating a single control link vs an entire control system comprising of, at least, a corpus database and all the supporting code that streams inputs and actuates outputs.

I assume / hope that there still exists manual hydraulic controls for the control surfaces. Maybe not in the cockpit, but somewhere accessible in flight in an emergency.

Modern U.S. submarines are also fly-by-wire, but hydraulic overrides are in the engine room in case of emergency.

I largely agree with you, however auto-pilot has existed for decades. I don't foresee AI taking over the cockpit anytime soon, but a compromise is likely. Perhaps it'll allow for copilots with less overall experience than copilots today.

>There are all kinds of failure scenarios where a computer, AI or not, would get confused.

but there may also be enough kinds of failure—and routine—scenarios where computers do a better job, so on balance they could be much better, just killing a few people in the most absurd situations that would never fool a human. You can't guesstimate reason these things, need statistics.

also, if automated systems have a bright future for us, maybe we have to sacrifice a few people on the QA team in order to get there.

> I could not feel comfortable as a passenger on a commercial airliner without a human on board that could take over and fly the plane manually. There are all kinds of failure scenarios where a computer, AI or not, would get confused.

"Person there just in case to take over in the rare case of emergencies" might not be as well paid as "person who's responsible for the plane at all times". Plus, if you're concerned about malware, you'll probably want a person still involved in vetting the code outputted by some hypothetical AI, so there would still be at least some engineering jobs.

The person to fly in case of an emergency will require the same kind of currency as a person flying all the time. The person flying all the time will be much more capable than the person watching the plane fly, and acquiring experience only in specific training.

> The person to fly in case of an emergency will require the same kind of currency as a person flying all the time.

Almost everyone requires the same "kind of currency", but that doesn't stop salaries from being different.

> The person flying all the time will be much more capable than the person watching the plane fly, and acquiring experience only in specific training.

If the plane can be piloted mostly automatically, that doesn't necessarily matter. What matters only is if the emergency person can handle the emergencies that do occur.

To be clear, I don't have any strong belief that human pilots will disappear any time soon; I just don't think the arguments people are making here are that compelling, because they seem to be assuming that the skill level of flying a plane is much too high for an automatic pilot. As someone who's been skeptical of fully automated driving for a long time now despite popular opinion seeming far more optimistic about it in the short term, it's kind of ironic to see "humans will always be better at flying" presented as axiomatic.

I'm infinitely more comfortable with an AI flown plane than a AI driven vehicle. The issues the plane has to deal with are going to be much more predictable than the issues a vehicle has to deal with.

Certainly things like the ghimli glider are better for having a human at the helm, but those sorts of things shouldn't happen.

Given how good ChatGPT is at coding, I'd bet on software development being radically disrupted before we're boarding planes with zero pilots.

"I apologize for the mistake in my previous response. The altitude is indeed 1,200 feet, not 12,000 feet. The correct action to take is to increase pitc"

I don't think we'll ever be at zero pilots. But one pilot instead of 2(or more) would nearly halve the pilot market, no?

We're facing a really bad pilot shortage right now, even in the fun jobs like flying fighter jets for the Air Force. That change would allow more plains to fly, I doubt it'd hurt the market for pilots - salaries which are already depressed because the airlines have been able to get away with it.

The complaints I have heard from military pilots are that shortages are caused mainly by toxic leadership and unattractive career path options. Combat pilots don't appreciate taking orders from careerists or "shoe clerks". And they don't want to get stuck in staff or management assignments for years just because the service needs a warm body to fill a slot. Plus the day-to-day administrative workload is high even when they're not flying. So, a lot of them in the O-3 to O-5 range just get burnt out and quit. Air Force leadership could fix those retention problems if they actually wanted to, and it wouldn't even be very expensive.

Or a room of remote pilots that are signaled to take control in an emergency?

If a plane gets struck by lightning, the remote connection may cease functioning. So, too, might the onboard computer.

Regardless of what naive optimization might suggest, I doubt the airlines are going to risk the headline "217 Dead In Pilot-less Plane Crash" any time soon.

Flying the airplane is the easy part. The hard bit is responding to equipment failures and other emergencies. There is no way to predict all of the possible failure modes, nor do we have AI that can figure out how to manage unexpected problems in real time. That technology is likely at least several decades away.

There is also a sensing issue. If aircraft sensors fail then they might feed the AI faulty data. And sensor redundancy or fault detection logic can't necessarily cope with that. Whereas experienced human pilots have a pretty good record of using their organic senses to handle such failures safely.

The economics won't make sense any time in the foreseeable future. Pilots are a small fraction of the total cost of a flight, making the upfront cost to automate the extremely edge-case-laden final 10% of safety-critical operations they oversee a non-starter for now.

We'll have terrible Roomba boxes replacing flight attendants long before anything replaces pilots.

>We'll have terrible Roomba boxes replacing flight attendants long before anything replaces pilots.

For American-run airlines (and many European ones too), terrible Roomba-like boxes would be a big improvement in service.

Not for Asian airlines, though.

Not sure about the replacements for flight attendants. I understand they have important tasks during emergencies.

they're a pretty big fraction for shorter range flights/smaller planes (somewhere around 10%?), and more importantly they are a cause of cascading delays. often major delay problems happen because the flight before was delayed requiring the pilots to stop working and you now need to reschedule all your pilots leading to more problems. pilots are more of a problem than flight attendants because there are more attendants, so you can more easily have a few in reserve.

Pilots are more of a problem than flight attendants because not only are there fewer pilots, they need to be certified for the plane you have available. If your reserve pilot is certified on a CRJ-900 and you need to fly a 737, those folks aren't going anywhere.