In my previous blog I talked about the desire some airplane designers and manufacturers have to create planes that don’t need pilots. Unsurprisingly, I don’t think it’s a promising idea right now. I thought I’d said all I needed to in that previous post, but a recent discussion with a stranger just added to my thoughts on the subject.
Living in one place and working in another – places that in my case are 2500 miles apart – means you spend more time commuting than most people. That’s how it is for a lot of people in the airline industry, including me. It is both a curse and a blessing too. It takes away a lot of time that could be spent at home with my family, but it occasionally does offer new opportunities to talk to interesting people.
That’s what happened recently on a flight from Seattle to Honolulu, going from home to work. The plane was quite full, but I managed to get one of the few remaining seats in the main cabin of the plane instead of having to sit in the jumpseat. After I settled into a window seat just behind the wing, the gentleman in the aisle seat next to me introduced himself, and we exchanged pleasantries. On the climb out of Seattle he noticed my airline ID and asked about life as a pilot. I mentioned I was commuting, and he nodded saying he did a lot of travelling himself. The gist of his story is that he works as a launch safety officer for a number of commercial and government rocket launch operations and is the guy who pushes the button that destroys the rocket when something goes wrong. He was coming from a launch site on Kodiak Island, and was heading for another in New Zealand.
I’m very much a product of my boyhood era (the 1960s & 70s) and because I grew up during the Apollo program, I’m a die-hard life-long space fanatic. That means I knew about the rockets he works with and was fascinated by the stories he related to me. Or, at least the ones he was able to talk about without breaking government security rules. As part of the discussion, I asked him about the new autonomous safety programming that SpaceX is using in the launch of their rockets. It’s an automated program that replaces the human in the safety loop, using input from all the sensors aboard their Falcon 9 rockets to make the decision whether to self-destruct (or not) when something goes wrong. As expected, he didn’t think much about the idea. And it wasn’t because he was liable to be put out of work, but because he didn’t believe the autonomous system could be counted on to know everything going on with the rocket. As he put it, ‘How are the programmers going to be able to write code for the onboard safety program that would cover every possible glitch, error, mistake, or problem that could possibly be encountered, and still have it make the best decision in a split second?”
That lead to a discussion of the same concept regarding airplanes, and that discussion became the topic for this blog post. There are a number of people who believe we will have large passenger-carrying aircraft flying within a few years with no human pilots onboard. I am not one of those people. It was nearly a decade ago I read some posts on an aviation message board about automated airliners, and the proponents then said we were only a decade away from it becoming commonplace. That decade has now elapsed, and I don’t see anyone getting into remotely-piloted (or computer-piloted) aircraft yet. This lack of progress aside, proponents still say it will happen soon.
I do believe that when artificial intelligence can match the human brain in all aspects, we may very well have automated airliners, and at that point I will have no problem with them. But there’s a colossal abyss between that future and the capabilities of today. Most of the problems I see come down to the same question my seatmate asked about rocket launch safety programs. I’m going to put this into a Q & A format because the proponents seem to have an answer to every question. I just don’t see how it would work in the real world though. (I’m asking the questions, and a notional supporter of automated aircraft is answering). As a side note, this same argument applies to the idea of having just one pilot onboard, with either a computer assistant or a second person remotely helping via communications link.
Q: How do we teach these planes to be able to sense everything a human can, and to be able to analyze something it hasn’t been programmed for?
A: They will be programmed to learn from their experiences and use that learning in the future.
Q: Will they have, for example, a sense of smell to tell when a component is beginning to smolder, but isn’t triggering any other warning sensor? From my experience, that’s one of the first indicators of trouble, coming up long before any warning signal is triggered.
A: That will have to be answered by the certification officials in government.
Q: So you don’t have an answer to that yet?
Q: And you don’t know what types of sensors will be needed to match a human’s capabilities?
Q: Okay. So assuming you do program these airplanes to learn, how do you share that knowledge between airplanes? i.e. if one plane learns how to do something better than before, how do the rest of them do it too?
A: There would be data sharing between all common types of programming.
Q: So there’s a constant updating of the software?
Q: How do we know the updated version works as well or better overall than the old one, with no new glitches because of the update? Remember, when human lives are at stake the government requires rigorous testing for all versions of software.
A: I’m sure we can convince the government to accept the changes since they are only improvements to the original programming. If we didn’t, the slow updates would prevent important safety changes.
Q: These airplanes would fly on constantly changing software, without functional testing between versions?
Q: Okay, what is the computer’s response when something happens that it simply isn’t programmed for?
A: Like what?
Q: Like, I have no idea! That’s why it’s something that isn’t programmed. It’s something unseen before. Like Sully losing both engines beyond gliding range of an airport. Would the computer know how to glide to a river? Is that what would be programmed? Or would it merely look for an open field? How does it choose between those options? Or what about having an engine blow apart and cause the aircraft to lose all hydraulic fluid like the United DC-10 in 1989, causing all flight control surfaces to become useless? Would it know to use differential thrust to steer the plane once all the control surfaces had frozen?
A: We certainly could program the computer to do that.
Q: Yes, you could. But would you have done so before 1989 for the control loss, or before 2009 when Sully ditched in the Hudson? Those were completely unforeseen events that turned out favorably because a human was at the controls and was able to analyze a situation they had never before considered.
This Q & A could go on for pages, but I think you get the idea. It’s the same conflict that goes on all the time in software and hardware labs across the country. But this is not software for laptops, but software to run a machine that hundreds of people are trusting their lives with every time it operates.
This is the heart of my issue with autonomously piloted aircraft. It will only ever know as much as the programmer puts into it. That programming won’t have the years of experience earned by a human actually flying the plane, or a human's ability to improvise something completely unexpected. Now, I’m not against automation. I fly an Airbus A330, a highly automated aircraft. But that automation is there to enhance our abilities and to provide us as pilots the chance to do what we humans do best – to see the big picture of what is going on all around us – and use that knowledge along with the automation to increase safety and comfort for our passengers. Taking that topmost element, the ability of humans to analyze and evaluate a brand new situation based on their prior experience, and replacing it with a pre-programmed computer devoid of a human’s analytical and improvisational capabilities, is far beyond what computers are capable of now. And (I believe), beyond what they will be capable of for a long time to come.
Commercial aviation has become the safest form of transportation ever invented by mankind. It would be a shame to lose that advantage in the name of saving a few dollars by cutting pilots out of the equation. The human brain has no equal now in terms of what it takes to do complex tasks like piloting a plane, and it probably won’t for many years to come.