After Math: Number stations

It was a week of hodge-podgery for the tech industry. AT&T rolled out its almost-5G service, Amazon finally revealed its Prime membership figures and Facebook continued its multi-year streak as "Company Least Encumbered by Any Discernible Form of…

Where have all the pilots gone?

You’d think everybody would want to fly. It’s been a universal human dream since the first cave person saw the first pterodactyl¹. You’d think better technology, greater demand, economic growth, and population growth would mean more and more pilots. But the surprising, counterintuitive fact is that fewer and fewer people are flying, and now Earth needs pilots, badly.

Airline industry facing a massive shortfall of pilots.” “Yes, there is a definite pilot shortage. It is true in all parts of aviation.” “The US Air Force is short more than one-quarter of the fighter pilots it needs.” “Asian airlines are running out of trained pilots.” “‘Extraordinary’ Pilot Shortage Threatens Flights; 637,000 Needed.”

Meanwhile, the number of active pilots in the US has declined from over 800,000 in 1980 to barely 600,000 in 2017, a quarter of whom are student pilots, a certificate for which you need no experience at all. Of course there are pilots and there are pilots. A private pilot in a little Cessna is very different from an airline transport pilot guiding a 777. And one reason there’s a shortage is that, while that 777 pilot pulls in six figures, an overworked copilot at a remote feeder airline gets paid peanuts.

But this overall broad decline in piloting is still truly remarkable. Why are we flying so much less in person, at the same time that we are flying so much more remotely? (The demand for commercial drone pilots, who in the USA must qualify for a “remote pilot certificate” by passing an aeronautical knowledge exam and a TSA security check, is also growing.) Why are fewer and fewer people taking to the skies, when they have never been more accessible, and flying car startups, some of them self-flying, are erupting like mushrooms after rain? Might self-flying airplanes ultimately solve the pilot shortage?

To try to answer these questions and more, I have recently taken up flying lessons myself, as a sterling example of investigative journalism on behalf of TechCrunch’s readers.

I jest. Really this was my friend Nat’s fault. “The thing about flying,” he said to me over dinner once, “is it combines romance, adventure, science, and exploration.” A heartbeat of stunned silence later I managed to retort, “Well, that sounds terrible,” but the damage was done.

Taking off seems easy enough, at first, on a demo flight. Just thrust the throttle forward, and feel the whole airplane thrill with the engine’s unleashed power as you accelerate down the enormous runway. The flight instructor next to you tells you when to pull up, gently — you’re not even moving that fast, maybe 70 miles an hour, normal highway speed — but when you do, just like that, you’re flying. You are so accustomed to vehicles on wheels that the freedom from the tyranny of the earth, the absence of the sensation of ground against tires, feels almost vertiginous, like weightlessness.

Around you the earth falls away: runway, airport, golf course, the San Francisco Bay glittering in the sun. From a cockpit 2500 feet up the Bay Area looks almost too gorgeous to be real, like a special-effect matte painting of sea, rippling hills, great pale swathes of buildings, cargo ships arrayed in their unloading queue, the forest of skyscrapers that is downtown San Francisco, the pale arc of the Bay Bridge, the clenched fist of Alcatraz, the famed distant silhouette of the Golden Gate.

I’m a terrible cliché now, of course. A Bay Area tech CTO who takes up flying is about as remarkable as a coastal Australian who takes up surfing. I blame Nat.

Does self-deprecatingly admitting that you’re a terrible cliché make it better or worse?

“Science,” he said, and there’s some of that, but really it’s mostly engineering, a kind very different from the engineering I know professionally. This is physical, visceral, greasy. Not a Matryushka doll of nested software abstractions, running on some faraway server whose physical details you don’t know or care about; not digital chipsets and circuit boards, taking advantage of Moore’s Law and the peace dividend of the smartphone wars, to drive LEDs or solenoids or little electric motors. This is airfoils, spars, composite materials, airflow vortices, a shifting center of gravity as fuel burns, physical forces fighting to keep you aloft against the relentless pull of the Earth. This is pistons, spark plugs, carburetors, magnetos, fuel pumps, propellers.

You need to understand how all this engineering works because it is there to keep you aloft and alive. Light aircraft are not dangerous — the one I’m learning in, the Diamond DA-40, a 21st-century airplane with an excellent safety record, is statistically safer per hour than a motorcycle — but that’s because of pilot training, not their inherent security. Whether you like it or not, part of the adventure of flying is that it’s replete with risks. Weather risks, largely: thunderstorms, icing, wind shear, and especially clouds.

(Yes, clouds. Basic pilot training is for “VFR” (visual flight rules) and if you’re not trained to fly “IFR” (instrument flight rules) then clouds can and will kill you, because without a visual horizon to track, your instincts and senses will promptly start telling you lies about your airplane’s attitude and behavior, and if you’re not trained to override those gut feelings, and follow what the instruments say, then you are asking for a controlled flight into terrain. Fun fact: night flights over water can still be “VFR” in the USA! See also the sad fate of JFK Jr.)

But technical risks are very real too. Did water get into your fuel tanks? Were they accidentally filled with jet fuel instead of avgas? How do you know? Is your engine running rough today? Maybe you just need to lean the mixture for a few minutes during the run-up; and maybe you need to turn around and call a mechanic. What speed will this airplane stall at? Trick question! Stalls aren’t dictated by speed. You better know what they are dictated by, if you want to fly.

And you do. Or at least I do. It’s glorious. It’s adrenalinizing, it’s breathtaking; it’s freedom, it’s beauty; it’s like dreaming while awake.

That said, learning to fly is frequently more Type II fun than Type I. I always actively enjoy it while I’m doing it, but at the same time, it is often tense, draining, and stressful. You need to always be on when you are in the cockpit. It takes time to get accustomed, at a gut level, to hurtling through the sky at high speeds in a little shell of fibreglass and carbon fibre with wings and a tail. And at least at first, you are drowning in information and obligations.

Student piloting is brief periods of pleasant inactivity interspersed with frequent periods of frantic multitasking. Aviate, Navigate, Communicate, they say — but at first aviation alone seems to take more attention and brainpower than you can allocate. You have rudders, ailerons, elevators, trim, and throttle to control. Sometimes you need to tweak the propellor, the mixture, and the active fuel tank. All this while constantly watching your airspeed, altitude, heading, and vertical speed; maintaining awareness of your engine indicators; and keeping an eye out for other airborne traffic.

It’s easier than that sounds, but it’s not as easy as it looks. Even takeoffs are harder than they first seem. (When you push the throttle forward, four separate physical forces skew the nose of the airplane sharply to the left, so you need to step on the rudder, without stepping on the brake, to keep the nose straight-ish.) Landings are hard full stop. Well, sometimes they feel easy, but consistency is hard.

Are self-flying planes on the horizon? I am skeptical, barring a new breakthrough in machine learning, which admittedly I don’t rule out. But there are two barriers. First, when will safe self-flying be possible? Self-driving cars are hard enough, and they only have one axis of control, and don’t get blown around by winds, and if something goes wrong you hit the brakes. Airplanes have pitch and roll as well as yaw, and move within a highly dynamic medium, and if something goes wrong — like an engine failure, or a bird strike — a quick halt is generally the exact opposite of a desirable outcome. I can easily envision self-flying AI which handles 99.99% of flights, but that 0.01% of exceptional situations will be awfully hard to train for.

Second, even if we get there, when will it be practical? While individuals might volunteer to be bleeding-edge adopters, how can you prove its validity to the FAA and other regulatory authorities? We’d need to add many more nines before self-flying software start competing with professional human pilots, who, unlike human drivers, have a remarkable safety record; commercial aviation had zero fatalities in 2017. Better autopilots for ordinary conditions are one thing, but removing pilots from flying entirely is quite another. Maybe after we build up a long, deep history of perfect safety with comparable drones or military flights; but not any time soon.

Better technology will however help with navigation. I don’t mean point-to-point, I mean in familiar places. Navigation may seem relatively easy above the San Francisco Bay, a well-known territory full of landmarks. Guess again. That sky may be empty but it is not unoccupied. Instead it is segmented into dozens of complex three-dimensional zones, and woe betide you if you stray into the wrong one.

Bay Area VFR airspaces

Picture a tiered wedding cake, upside-down, with radiuses measured in miles. That’s the airspace of San Francisco International. But right across the bay you have Oakland International, which has its own smaller but still sizable wedding cake, and a little south San Jose International has its own, and both of those intersect with SFO’s. Then you have the half-dozen smaller regional airports, each jealously guarding their own disc of space, except where squashed by one of those cakes. Each of those kinds of airspace has its own rules and regulations. (SFO’s have the virtue of being exceedingly simple, for student pilots: keep out.)

You may not enter any of those airspaces without first communicating with their controllers, and to communicate you first must master aviation’s clipped, dense, custom language. “Hayward Tower, Seven Papa Victor holding short at runway Two Eight Left Alpha, request right crosswind departure.” “Norcal Approach, DA-40 Seven Eight Seven Papa Victor, three thousand over Lake Chabot, inbound to Oakland for touch-and-gos with information Foxtrot.” “Seven Papa Victor, squawk oh three five seven and contact Oakland Tower.” It would be unremarkable to change frequencies several times, and talk to a few different controllers, during a half-hour Bay flight.

Knowing what frequency to use, what to say, who to say it to, and when, while picking your own call sign out of the frequent chatter, most of which is irrelevant to you, and parsing / copying down the important information you need — that would be nontrivial all by itself, at first. But it’s not by itself. It’s something you do simultaneously with everything else you’re doing while flying the airplane.

Does the heavy use of voice communications over frequently (and manually) shifted shared channels seem a little … well … twentieth century? A little technologically backward? Well, yes, and no. Voice over radio is simple, powerful, flexible, and time-tested. There are a lot of old airplanes and old pilots out there. Aviation as an industry is understandably loath to make rapid changes — many of its rules are, as they say, written in the blood of people who learned the need for them the hard way.

That said, modern aircraft like the DA-40s I’m learning on tend to have “glass cockpits,” with one LED screen displaying an artificial horizon and all the important instrument data so you don’t have to look at the actual dials (which are still there as backup), and the other displaying a zoomable map with terrain, your heading, airspace boundaries, nearby traffic, etc., and containing databases of information such as airport locations, runways, and frequencies — all at your fingertips if you can master their baffling and perverse knob-and-button user interfaces. (“Turn the big knob left. Now turn the little knob right. Now push ENT. Now turn the little knob left…”)

Apps like ForeFlight make it easier yet. And we happen to be 20 months away from a massive technological phase shift in general aviation, after which much American airspace will require “ADS-B” technology that will essentially let every aircraft be tracked in 3-D in real time; this should make communications and aircraft spacing much easier.

It feels a little bureaucratic, it’s true. The romance of the glory days of flight, Antoine de Saint-Exupéry and company wrestling their planes over the Andes and the Sahara, with the freedom of the whole sky thanks to their skill and their machinery, feels distant from today’s strictly ruled, tightly regimented airspaces, and constant surveillance anywhere near a major airport. But then the skies were empty back then, and the machinery all too often so lethal that skill meant nothing, in the end.

And, I mean, these too are the glory days. You can fly. All by yourself. It isn’t an easy thing to learn, or to do. (OK, some people are naturals. I myself am not.) Multitasking is hard. Kinesthetic learning is hard. Establishing new muscle memories is hard. Developing good judgement is hard. Flying an airplane smoothly, with coordinated turns (using the ailerons and rudder together) while maintaining precise control of altitude and airspeed and bank angle, is … actually that’s not so difficult; but doing all this while at the controls of an aircraft that’s, say, being buffeted by crosswind gusts as you turn towards a runway, in a busy traffic pattern, with the stall warning beginning to whine because you banked too late and too hard, but it’s too late to fix that judgement error now, and the radio crackling in your ears as the tower says something which might or might not be germane to you —

— well, the instructor who made that first takeoff seem easy told me, later that same day, that most people who begin pilot training never finish it. There are plenty of good reasons for that. It is, as my friend Dillo put it, more expensive than a crack habit. People hit plateaus and get frustrated and give up. But I think the main reason is because it’s complicated, and difficult, and stressful, and when the lessons stop being novel, people stop forcing themselves to do the hard thing, despite the ultimate rewards.

Is that why there are far fewer pilots in America than there were in 1980, even though there are 100 million more people? Would better, modernized navigation and communications technology go a long way towards making flying a little less draining, and a little more appealing? Maybe. There are cultural reasons, too, though, and I think they’re more significant. I think we now lean more towards the abstract than the physical, and towards comfort rather than adventure.

I remember, years ago, seeing online reactions to a study reporting that teenagers in gifted programs were likely to quickly drop things they weren’t immediately good at, the theory being that they feared losing their gifted designation, and that this instinct persisted into adulthood. An astonishing number of my friends, especially my friends who worked in tech, said they strongly identified with this. I wonder if that’s a factor.

Most of all, though, I think flying seems like a very 20th-century activity in the popular imagination. But I suspect that won’t last. Something, whether hardware or software, will catapult it into the 21st century mindset soon enough.


¹Yes, I know. It’s a joke.

We’re listening to: Acquisitions Inc. and The Adventure Zone

This week's IRL is about podcasts. Or games. Or both? Kris Naudus loves role-playing games — the ones with dice rather than pixels — and she also loves podcasts. Combining the two should be a dream, but Kris has struggled to get into some of the mo…

After Math: The golden age of streaming

We're winding down a freaky week of bad omens, unconvincing congressional testimony and sterling new streaming services. While most eyes were fixated on Mark Zuckerberg doing his best real boy impression, GTA IV lost most of its Russian tunes, ESPN d…

Mutual assured dronestruction

Western airstrikes on the Middle East: déjà vu all over again. Twenty years ago, the USA attacked Sudan and Afghanistan with Tomahawk cruise missiles. Two days ago, the USA attacked Syria with … Tomahawk cruise missiles. Aside from the (de)merits of each attack, isn’t it a bit surprising that technology hasn’t really changed small-scale strategic warfare in that time?

Just you wait. In the next decade, that strategic calculus will change a lot, and probably not in a good way. Consider this sharp one-liner from Kelsey Atherton last week:

Of course cheap drones are already being used on the battlefield in small-scale ways: by Daesh, by Hezbollah, by Hamas, by drug cartels, and of course by traditional nation-state militaries worldwide. But those are piloted drones, used in short-range, often improvisational ways; interesting but not really strategically significant.

Meanwhile, across the world, we are in the midst of a Cambrian explosion of artificial intelligence and automation technology. Consider, the startup that began as a literal one-man self-driving-car project. Consider the truly remarkable Skydio, a self-flying drone that can follow you wherever you go, avoiding obstacles enroute.

…Do you see where we’re going here? Right now only major powers can toss a few explosives at a faraway enemy to drive home a political point. But imagine a flock of bigger Skydios, reprogrammed to fly to certain GPS locations, or certain visual landmarks, or to track certain license plates … while packed full of explosives.

A  Tomahawk costs $1.87 million. It seems to me that we are not far at all from the point where a capable and wealthy non-state actor like Daesh, Hezbollah, Hamas, the Sinaloa cartel … or any unsavory group willing to be used for plausible deniability by a nation-state … could build a flock of self-flying targeted kamikaze drones, then smuggle them into the Western destination of their choice, for a lot less than the price of a single Tomahawk. The self-flying and targeting software / AI models won’t need to be nearly as perfect as that of a self-driving car. A 50% failure rate is more than acceptable if you only want to show force and sow panic.

It’s chillingly easy to envision a future of mutual assured terror, a multipolar world in which nations and terror cells and drug cartels and starry-eyed cults alike have the capability to inflict faraway havoc on thousands and constant dread on millions, a smoldering kaleidoscopic landscape of dozens of factions enmeshed in tit-for-tat vengeance and vendettas — ceaseless cycles of sporadic attacks which rarely kill more than a hundred, but send entire populations into perpetual fear and fury. Fury which will be very hard to direct. Like hacking, autonomous drone attacks will be extremely difficult to attribute.

You may call this science-fiction scaremongering, and you may have a point. It’s true that nothing like this has happened yet — though the existing adoption of commercial drones for warfare is a distinct warning sign. It’s true that it would be wrongheaded and ridiculously preemptive to try to slam the barn doors before any drone horses arrive. I’m definitely not suggesting that the West should start thinking about restricting research, or trying to comtrol either hardware or software. (Even if that worked, which it wouldn’t, it would be pointless; drone hardware is cheap, and R&D is global.)

But it’s not too early to start thinking about how we will cope if and when self-flying kamikaze drones do make asymmetric strategic warfare possible. And it’s definitely not too early to try to minimize such warfare before it happens, ideally by actually trying to deal with the root causes of the conflicts burning around the world, rather than lobbing a few cruise missiles their way every time we feel the need to seem particularly outraged. Because one day, not long away at all, that approach will begin to rebound on us disastrously.


The Zuckerberg hearings were a wasted opportunity

Over the past two days, members of Congress have peppered Facebook CEO Mark Zuckerberg with questions about how the platform manages users' privacy, what went wrong with Cambridge Analytica and what it's doing to strengthen protections going forward….

After Math: The week of living dangerously

It was a chaotic week in the tech world, even before the YouTube HQ shooting. Apple's pushing its luck by pushing its Mac Pro release to next year, Russia's mail delivery drone barely got off the ground, and Scott Pruitt's EPA is doing its best to su…


You may not be aware that the rough consensus in the tech community is that Cambridge Analytica were almost certainly bullshit artists. Oh, don’t get me wrong, what they tried to do, and/or claimed to do, was super shady and amoral and would have been ruinous to reasonable informed democracy if successful. But I have yet to find anyone credible who thinks that their “psychographics” approach actually works.

That is: it doesn’t work yet.

We may yet thank Cambridge Analytica, for inadvertently raising the red flag before the real invaders are at the gate. Because in this era of ongoing exponential leaps and growths in data collection, which is also an era of revolutionary, transformative AI advancement, it’s probably only a matter of time before electorates can be subtly manipulated wholesale.

Don’t take my word for it. Take that of Google senior AI / deep-learning researcher François Chollet, in a truly remarkable Twitter thread:

To quote the crux of what he’s saying:

In short, Facebook can simultaneously measure everything about us, and control the information we consume. When you have access to both perception and action, you’re looking at an AI problem. You can start establishing an optimization loop for human behavior … A loop in which you observe the current state of your targets and keep tuning what information you feed them, until you start observing the opinions and behaviors you wanted to see … A good chunk of the field of AI research (especially the bits that Facebook has been investing in) is about developing algorithms to solve such optimization problems as efficiently as possible

Does this sound excessively paranoid? Then let me point your attention to another eye-opening recent Twitter thread, in which Jeremy Ashkenas enumerates a set of Facebook’s more dystopian patent applications:

Again let me just summarize the highlights:

a system that watches your eye movements […] record information about actions users perform on a third party system, including webpage viewing histories, advertisements that were engaged, purchases made […] “Data sets from trusted users are then used as a training set to train a machine learning model” […]“The system may monitor such actions on the online social network, on a third-party system, on other suitable systems, or any combination thereof” […] ‘An interface is then exposed for “Political analysts” or “marketing agencies“’ […]

Are those quotes out of context? They sure are! So I encourage you to explore the context. I think you’ll find that, as Ashkenas puts it, “these patent applications don’t necessarily mean that Facebook wants to use any of these techniques. Instead, they illustrate the kinds of possibilities that Facebook management imagines.”

An explosion of data. A revolution in AI, which uses data as its lifeblood. How could any tech executive not imagine using these dramatic developments in new and groundbreaking ways? I don’t want to just beat up on Facebook. They are an especially easy target, but they are not the only fish in this barrel:

Here’s yet another viral privacy-related Twitter thread, this time from Dylan Curran, illustrating just how much data Facebook and Google almost certainly have on you:

Mind you, it seems fair to say that Google takes the inherent dangers and implicit responsibility of all this data collection, and the services it provides with this data, far, far more seriously than Facebook does. Facebook’s approach to potential malfeasance has been … well … let me point you to still another Twitter thread, this one from former Google Distinguished Engineer Yonatan Zunger, who I think speaks for all of us here while reacting to reports of Facebook’s CTO saying “the company is now mapping out potential threats from bad actors before it launches products.”

But the larger point is that the problem is not restricted to Facebook, or Google, or the big tech companies. It’s more acute with them, since they have more data and more power, and, in Facebook’s case, very little apparent sense that with great power comes great responsibility.

But the real problem is far more fundamental. When your business model turns data into money, then you are, implicitly, engaging in surveillance capitalism.

Surveillance and privacy are issues not limited to businesses, of course; consider the facial recognition goggles already in use by Chinese police, or India’s colossal fingerprint-face-and-retina-driven Aadhaar program, or dogged attempts by the UK and US governments to use your phone or your Alexa as their surveillance device. But corporations currently seem to be the sharp and amoral edge of this particular wedge, and we have no real understanding of how to mitigate or eliminate the manifold and growing dangers of their surveillance capitalism.

I’m not saying all data gathering is ipso facto bad; but I am saying that, given the skyrocketing quantity of sensors and data in our world, and the ability to tie that data to individuals, any initiatives which support privacy, pseudonymity, and anonymity should be considered desirable until proven otherwise, given the ever-growing oceans of data whose tides threaten to wash those lonely islands away.

I’m certainly not saying AI is bad. Its potential for improving our world is immense. But like every powerful tool, we need to start thinking about how its potential misuses and side effects before we rush to use it at scale.

And I’m saying we should almost be grateful to Cambridge Analytica, for selling snake oil which claimed to do what tomorrow’s medicines actually will. Let’s not overreact with a massive moral panic. (checks Internet) Whoops, too late. OK, fine — but let’s try to take advantage of this sense of panic to calmly and rationaly forestall the real dangers, before they arrive.