AI: Pilot Threat or Bean-Counter Pipe Dream? (Part 3)

Thanks for sticking with me here Network! This is the conclusion to our series about the threat Artificial Intelligence (AI) poses to our jobs as pilots. Having examined the fundamentals behind training an AI in Part 1, and exploring some practical applications in Part 2, we’ll now consider the bigger-picture considerations standing in the way of a robot takeover.

What About Our Customers?

Thus far, we’ve only looked at the technological task of training and AI and how the associated limitations play into an aviation application. What about our customers though?

Individual passengers are skittish. The COVID-19 pandemic has split them into various camps. Some only want to fly on aircraft that require everyone to wear masks and keep middle seats open. Others prefer to roll the dice and want to pay less to ride like sardines no matter what. That’s fine…to each his or her own. However, all airlines are suffering because some passengers are foregoing travel altogether.

If masks and social distancing are causing this many problems, how will the traveling public respond to an aircraft run by AI? When (not if) this time comes, I guarantee that some ULCC types will jump on the chance to ditch pilots while others stubbornly keep pilot butts in seats for as long as possible. They’ll get to charge extra for it, and a significant number of people will pay…just like some subset of people are paying significantly more to fly the same routes in the same aircraft on certain airlines over competitors right now because they’re concerned about Coronavirus.

One common response to this assertion is: “Sure, but boxes don’t care. Cargo airlines will get rid of pilots first.”

On one hand, I agree. Boxes don’t care who’s flying the jet. However, the people who own those boxes do. The companies who manufacture the goods transported on cargo jets won’t want to trust important cargoes to unproven or incapable technology.

Beyond that, you also have to consider insurance companies. How many iPhone 25s fit on a Purple B777 or Brown B747 flying from China to LA? At more than $1000 each, that cargo of phones is worth hundreds of millions of dollars…probably more than the $350M or so the airplane itself costs in the first place.

There’s no way Apple is going to trust a city’s worth of new iPhones to an AI pilot just a few days before release. Even if Apple decides to take the risk, their insurance provider (and the airline’s insurance provider) are going to charge premiums so high that shipping new iPhones on AI-driven airplanes will be cost-prohibitive.

Single-Pilot Ops

The next step in this discussion usually leads anti-pilots to suggest that companies will balance this risk by pairing human pilots with AI copilots and a dog. (The dog being there to bite the human if he or she tries to touch anything.)

This may work for relatively small FedEx ATR 42-300s or C-208s. Unfortunately, I think mental health concerns are the most likely reason this is a no-go for large cargo or passenger operations.

A few years ago, the FO on Germanwings 9525 waited until the Captain had stepped off the flight deck to use the lavatory, then barred the door and crashed the jet into a mountain. Ever since, airline regulations have very strictly required at least two people on the flight deck at all times. If a pilot needs to use the lav, a flight attendant or jumpseater has to keep the other pilot company.

It’s a sad truth, but I think this is the single largest threat to single-pilot operations. Anecdotally, because I’d never do this myself, some pilots have been known to “find” maintenance issues at choice layover locations. It doesn’t take a genius pilot to think of ways to make this happen. It’d be tough now to isolate yourself on a Part 121 flight deck. It’d be relatively trivial to disable an AI copilot. Even when (not if) AI copilots are equipped with anti-tampering protections, a human copilot will still have access to circuit breakers, switches, or other ways of disabling the sensors, power sources, or other equipment that an AI needs to fly.

If I were an insurance provider or shareholder of an airline (oh wait, I am) I would absolutely refuse to allow single-pilot ops for big airplanes. It’d either need to be 2 pilots or no pilots, and we’ve already demonstrated that the technology is nowhere near ready.

False Summit

My favorite 14er in Colorado (so far) is Mount Elbert. It’s a gorgeous hike, but it plays tricks on your mind. Many times throughout the ascent, you see what looks like a summit up ahead. Unfortunately, several times during the hike, you crest that point only to find there’s another, higher point up ahead. By the time you finally reach the summit, it’s almost anticlimactic.

We’ve taken a hard look at the technological obstacles that must be overcome before computers can steal all our jobs. I hope you see why this is a long time away.

That said, I believe that the day will come when pilots are superfluous. Those technological summits will all be reached. However, on the day when technology finally reaches that peak, the jerks who hate pilots will realize that an even more difficult summit lies ahead. That obstacle has a name that will crush the pilot-hater souls just as much as it does for us today. It is none other than: the FAA.

In case you haven’t noticed, the FAA is ridiculously averse to technological development. Way back in 1986, a pilot and engineer named Klaus Sevier started developing laser ignition and other systems that enabled him to fly a Long EZ at 100 miles per gallon. A few years ago, he flew his plane non-stop from San Diego, CA, to Lakeland, FL. (He sells versions of some of his equipment, if you have an experimental aircraft.)

It wasn’t until decades later that the FAA allowed engine manufacturers to advance past 1930s technology and include electronic ignition systems like this in certified aircraft. There still isn’t a single production aircraft anywhere near as efficient as Sevier’s Long EZ, despite the technology existing for decades, because it would be too costly to meet the FAA’s certification requirements.

The FAA’s new transponder system, ADS-B, took years to develop. They gave the public a decade to implement it, and it still isn’t fully adopted by pilots or ATC. This was for a completely superfluous new transponder system that is still backed-up by the old system. I think this bloated bureaucracy will take a very, very long time to approve any serious AI-based aircraft automation to fly “for compensation or hire.”

Sure, there’s always the Golden Rule – he who has the gold makes the rules. If the airlines decide they can save billions of dollars by implementing pilotless aircraft, the tech companies decide they can make billions of dollars by selling that technology to the airlines, and Boeing and Airbus think they can make billions amidst all that slop, the lobbying efforts in favor of pilotless aircraft will be immense.

However, I’m still a skeptic. Even after we get past all the technological hurdles that make the idea of pilotless aircraft a joke today, the FAA will delay implementation even longer.

Tesla’s Unwitting AI Trainers

I hope by this point you’ll at least agree with me that it will be difficult, expensive, time consuming, and complex to train an AI to safely operate an aircraft in a real-world environment.

Returning to the example of self-driving cars, we now understand how much data companies like Tesla require to train their self-driving AIs. The media has only just realized that Tesla has been using its own customers to conduct that training, in many cases without the customers realizing what’s going on. (Here’s the letter penned by the Chairman of the National Transportation Safety Board, NTSB, on the subject.)

If pressed, I’m sure companies like Tesla would readily admit to this. They cover themselves by insisting that although their features have names like “Autopilot” they still require constant driver attention and frequent intervention. They also admit that their self-driving technology only works under specific driving conditions.

Frankly, it’s the only way any company could possibly afford to get the millions of testing repetitions needed to train an AI for this type of situation. If they had to fund this testing on their own we’d still be decades away from self-driving features on any cars.

Driving takes place in a complex and challenging environment, but aviation is an even tougher scenario. It will take similar or greater amounts of training for self-flying AIs to reach even the accident-prone level of situationally-dependent competence that Tesla’s cars have reached after millions of miles driven. The question is: will the FAA even allow that to happen?

How can aircraft manufacturers hope to attain the repetitions needed to train their AIs without testing them in commercial air service?

In a way, this sort of thing already happens. An aircraft may be built for autoland or RNP approaches, but a given operator has to demonstrate hundreds of successful approaches in VMC conditions, with two pilots on board and ready to intervene, before the FAA will actually allow that operator to conduct these operations in real-world IMC conditions. (The A220 fleet I fly went through this process to get CAT III autoland approved, and is still working on RNP approval.)

Those requirements are for systems using relatively old technology and traditional computing/programming. I can only imagine the FAA will be more stringent with cutting-edge, AI-based applications. (Especially given the utter shame they’ve faced since their wantonly negligent oversight of the B737MAX program.)

This means before the FAA allows an AI to fly a pilotless aircraft, the companies selling that technology will have to demonstrate thousands of safe flight hours for operations that don’t even have a paying customer aboard. Even then, I think the best case for a next step will be AI-equipped aircraft flying under the same crew requirements we have today for at least several years before single-pilot or pilotless operations are approved.

No matter how this training happens, it can’t even start until technology reaches a level that can support it.

Better Options

If all this were happening in a vacuum, I’d be a little more pessimistic about our jobs. However, AI and fixed-wing aviation aren’t the only areas experiencing rapid development right now. I think a far better candidate for pilotless travel and cargo transport can meet those needs sooner and more safely. I’m specifically thinking about suborbital spaceflight.

Don’t laugh. Companies like Virgin Galactic and Blue Origin are already bootstrapping themselves based on the idea of space tourism…short launches that go just high enough to yield a few minutes of freefall and “earn” their passengers astronaut wings. If these companies (or others like SpaceX) decided to point their rockets at an angle instead of straight up, their spacecraft could spend the freefall portion of their trajectory cruising (coasting) toward a destination. Spaceships like Virgin’s SpaceShip2 or Sierra Nevada’s Dream Chaser would then glide to a destination runway while the Blue Origin or SpaceX rockets would land, autonomously, on a landing pad or drone ship at their destination.

Both of the latter companies have demonstrated safe and successful launches and landings. SpaceX even managed a formation approach and landing when they shot Musk’s Tesla Roadster to Mars. It’s tough enough to execute a nice formation landing with a pair of T-6s. To accomplish this with a pair of rocket ships is simply outstanding.

These companies have designed their spacecraft from scratch to operate without pilots. Public sentiment and FAA approvals already accept that fact. Sure, costs are high right now, but who do you think has the ability to actually accomplish all the innovation necessary for these advancements?

On one hand you have Boeing, the company so shortsightedly addicted to quarterly earnings reports that instead of taking advantage of an opportunity to lead its industry by applying updated B787 technology to a B757 replacement, it gave us the cop-out that is the B737MAX. They’d have to put forth what we’ve already identified as an unbelievable amount of effort to develop and train AIs at their own cost, and then get the FAA to approve them. Although I think highly of Boeing’s aircraft and heritage, I don’t see their board having the vision or courage to undertake a task of this magnitude anytime soon, if ever.

On the other hand, you have companies like SpaceX who have done more in a couple decades to advance space travel than all the governments of the world have accomplished since the 1960s.

Suborbital transportation won’t be for everyone. However, customers pay thousands of dollars for a nice first class seat on a transatlantic flight, tens of thousands for a suite on a carrier like Emirates, and millions every year to travel on private jets. I see a day when suborbital rocket travel will reach a price point these customers will be happy to pay to get from New York City to Beijing in just a couple hours.

If rockets can beat airliners to the punch with pilotless craft carrying those passengers, will it even be worth the cost and effort to develop AI-operated airliners? At that point, you might as well just keep flying big jets with two pilots. Sure, you won’t make quite as much money without those highest-paying customers. However, there’s a lot of room between the price of an Etihad suite and the ULCC 787 model that just won’t die. Traditional airlines will be able to remain profitable for a long time to come flying with two pilots even after Musk, Bezos, and friends implement rocket-based travel.

The Long Game

Let’s remember: I’m not making this argument to say that we’ll never have pilotless aircraft. Eventually, yes, pilots like us will need to worry about job security. AI technology will improve, hardware will improve, and regulations will bend to economic pressure…as they always have. There will be planes, trains, and automobiles (and more) transporting human beings and expensive cargo without any pilots or drivers onboard.

The important thing to realize is that we are not the pilots that technology will put out of work.

Technologically, the AIs at work in our world today are less capable than your average 4-year-old, and only marginally more capable than Gru’s minions. The biggest things holding these AIs back are the logistical and economic difficulty of gathering datasets needed to train them. (Though there are plenty of other technology-based reasons they aren’t read for prime-time.)

Even when, many years from now, we learn better ways to train our AIs, they’re going to face tough regulatory hurdles. Whether it’s potential customers or government flunkies, we’ll remember all the times that AIs did things like steering and accelerating toward a concrete barrier shortly before killing their passenger. Certifying those aircraft will not be easy. Convincing passengers, goods producers, and insurance companies to trust them will be even harder.

I’m 41 years old and have no fear of losing my job to AI in the next 24 years before I hit the FAA’s mandatory retirement age. TPN’s youngest correspondent, Tristin Stevens, can essentially reverse those numbers. I believe she’ll be able to enjoy a full flying career herself.

My oldest kid is 11. If she decides to become a pilot, I might be a little more concerned. However, I still think that suborbital rockets will provide better pilotless travel options than airliners. Even if rockets eventually lose out in that competition, it will take many decades to get there.

If you’re a pilot right now, I say rest easy and enjoy your career. If you’re an aspiring aviator wondering whether you should start down this career path, I say: “YES!” You have plenty of time to enjoy a lot of great flying, long before AI boots you out of your seat.

In the meantime, be sure to fill up your treasure bath while you can so you won’t be reliant on that job anyway. If you’re really worried, make sure you stay sufficiently educated that you can easily transition to something else if I’m completely wrong.

In the meantime, comfort yourself by asking your phone some questions with follow ups while you enjoy your next deployment or layover. I assert that you don’t need to even start worrying until Siri, Alexa, or Cortana can explain why she thinks it isn’t a cat and what it is instead.

< Part 1 | Part 2 | Part 3 >

Did you enjoy this article? TPN has hired a full staff of Correspondents who publish great material almost every day. If you’d like to read more great aviation content, we’d like to invite you to join our subscription-only Community Website.

This exclusive social network is a place where serious aviators from among The Pilot Network’s 25,000+ members can find great content, receive quality mentoring, and network with other aviation professionals. We’ll see you there!