Just before midnight on July 7 last year Air Canada Flight 759, an Airbus A320 carrying 135 passengers and five crew, was on final approach to San Francisco International Airport on a flight from Toronto. There were no weather problems. The pilot was directed by air traffic control to land on Runway 28R. The airport’s second main runway, 28L, had been closed for maintenance work at 10 p.m.
Running parallel to 28R, a little less than 500 feet from it, was a taxiway where four airplanes were held in a line waiting to takeoff while the Canadian jet landed.
Three of the four waiting airplanes were United Airlines flights and the fourth was a Philippine Airlines flight, which was second in the line. With the Air Canada jet less than a mile from touchdown the Philippine pilots realized that it was not heading for Runway 28R but for the taxiway—straight at them. They turned on their landing lights—powerful glaring white lights—as a warning to the Canadian pilots.
At the same moment the pilot of the United flight ahead of them, the first in line, called out urgently to the controllers: “Where is this guy going? He’s on the taxiway!”
During the following few seconds the Canadian pilots realized that something did not look right, and applied power to abort the landing. Fractions of a second counted. Jet engines require four or five seconds to respond to a command for more thrust, and the Airbus was configured for a landing with its landing gear down and the wing flaps fully extended, slowing the speed to around 140 mph.
Before the Airbus began to climb it was less than 60 feet from the ground. Two of the airplanes on the taxiway were Boeing 787 Dreamliners, with a tail height of 55 feet 10 inches. If the Canadian pilots had waited just five more seconds before reacting they would have hit the third airplane in the line, United Airlines Flight 863.
With the four airplanes on the taxiway fully fueled for their flights and more than 1,000 passengers in all five airplanes an impact would have caused a fireball large enough to consume all of them. It would have been the worst aviation disaster in history—far exceeding the previous greatest catastrophe when two Boeing 747s collided on a runway in Tenerife, in the Canary Islands, in 1977, killing 583 people.
A few seconds above the taxiway in San Francisco made the difference between 2017 being by far the safest year in modern air travel rather than the deadliest.
Not one passenger died on a commercial jet anywhere in the world. The only fatalities were 44 involving cargo jets and small commuter propeller-powered airplanes. In the U.S. the numbers are even more reassuring: No major U.S. airline has had a fatal crash since 2001; the last fatal crash that involved a commuter airline of any size was in 2009 when 49 people died on a Colgan Air flight in Buffalo.
However you look at it, these are remarkable achievements. There were nearly 37 million scheduled airline flights in 2017. They were operated in every kind of weather. They were flown by a wide range of airplanes, from the latest models to some that were relics of the analog age, as much as 30 years old. They flew in and out of some countries that have minimal standards of safety. They were flown by pilots who, no matter how well-trained, had variable levels of proficiency. And all this occurred when terrorists are targeting commercial aviation as the ultimate trophy attack.
The airline industry is having a vigorous debate about how this level of safety has been achieved. The Air Line Pilots Association, ALPA, cited regulations issued in 2013 that changed the number of hours of flying experience needed to qualify as an airline pilot from 250 hours to 1,500 hours.
In contrast, the president of the Aviation Safety Network that tracks all accidents credited the “persistent decline” since 1997 to “efforts by international organizations.” (Those organizations, the International Commercial Aviation Organization, and the International Air Transport Organizations, actually have an abysmal record of responding to safety issues.)
It is surely true that, whatever the bureaucratic hurdles, there has been a continuous and successful drive by industry innovators to eliminate the most frequent causes of crashes, largely by introducing more advanced technologies to both the airplanes and the ground systems like air-traffic control and weather forecasting.
But it is also true that serious lapses like the event at San Francisco show that it is sometimes only luck that produces such bullish statistics. Indeed, the real takeaway from that event is that the airline industry is approaching the moment when it will have to make the toughest call in its century-old history: Do robots make safer pilots than humans?
Of all the debates raging about artificial intelligence, AI, its future application in commercial aviation will be the most contentious. We are already being assured that self-driving cars are more or less a given by 2020, as fantastical as that may seem. Aviation technology experts propose that by 2040 we will be just as ready to accept pilotless airplanes. In my view, they just don’t understand how elemental and irrational fear of flying really is.
And yet, looking at recent history suggests that—in time—AI is more or less certain to produce the safest way of piloting an airliner. If anything is obvious about the way crashes have been reduced over the past three decades it is that the scope for humans on the flight deck to screw up has been steadily reduced.
For example, the most striking improvement in the cockpit has been the Ground Proximity Warning System, GPWS. Until the early 1970s many crashes were caused by pilots flying blindly into hills or mountains, euphemistically categorized as “Controlled Flight Into Terrain.” The FAA made GPWS mandatory in 1974. That kind of crash began to decline in numbers but in an upgrade to the system in the 1990s was much more effective and has virtually eliminated the risk.
Similarly, another invisible killer has been largely beaten: wind shear. Many fatal crashes of both commercial and private jets were caused during the critical phase of final approach when, without warning, there would be a sudden and violent shift of wind direction powerful enough to destabilize an airplane, causing pilots to lose control. Now wind shear alerts using detectors on the ground and in the cockpit give pilots enough time to abort a landing and, if needed, go to an alternate runway or airport.
However, always underlying the impact of the digitalization of instruments and automation on the flight deck is the progress of the amazing machine that gave us the Jet Age in the 1950s: the jet engine.
In the propeller age engine failure remained a problem nobody seemed able to fix. The large piston engines required to give enough power to fly 100 or so passengers across the Atlantic at 300 mph were at the limits of what that technology could achieve, and the stresses often led to engines failing, leaving an airplane crippled and hard to fly.
Today’s jet engines provide such reliable power that an airplane like the Boeing 777, with only two engines, is certified to fly over oceans carrying as many as 400 passengers at around 600 mph and be as much as five hours’ flying time from the nearest runway on the assumption that in an emergency it can reach that runway on one engine. And those engines are continually monitoring themselves during a flight, sending data to ground bases so that problems can be detected before they become critical and if spare parts are needed they will be waiting when the airplane lands.
With engines as dependable and as efficient as that and with the latest navigational aids the workload of pilots has significantly decreased and with it some of the stress. This was accelerated when the European airplane maker Airbus took stress reduction a large step further in the 1980s. They introduced all-digital flight controls, named fly-by-wire, and combined them with the concept of “envelope protection” in which pilots could not endanger the flight by over-reacting to an emergency by exceeding the safe limits of the hands-on flight controls.
There are some differences between Boeing and Airbus about how this technology is applied (Airbus uses sidestick controls, Boeing retains the central yoke, once known as the joystick) but both systems have greatly diminished the work load and opportunity to compound an emergency by making the wrong response to it. The other two principal planemakers in the world, Embraer in Brazil and Bombardier in Canada, produce jets with equally sophisticated cockpits.
And so there is now a crunch point approaching in the balance between how far pilots should be left in control of flying.
For at least two decades, there have been vocal concerns that pilots are already left with so little work to do on regular flights, particularly on longer routes over the oceans, that they lose alertness and vigilance. More seriously, that when faced with a sudden emergency air crews have not been sufficiently trained to deal with it—as was the case of Air France 447 in 2009 when 228 people died when an Airbus A330 crashed into the South Atlantic. (Steps have since been taken to fix the problem of crews who are suddenly required to handle “upset recovery” so that they can avoid a similar disaster.)
There may in the future be a technology that embraces AI and is sufficiently mature to safely handle all the demands made of pilots all the time in every kind of condition. Futurology prophets are sure that will happen. Until it does, here is the real dilemma: Sometimes pilots get into the kind of mess that apparently happened with Air Canada Flight 759. At other times pilot action is indispensable in saving the airplane, no matter how advanced the cockpit is.
On Aug. 3, 2016, an Emirates Boeing 777 carrying 282 passengers and 18 crew on a flight from India was making its final approach to Dubai International Airport. As it was about to touchdown there were several sudden shifts in wind direction and speed and, as a result, the airplane landed “long”—too far down the runway. The pilots decided to abort and applied power to go around for another attempt. It was too late. Before the engines could reach full power the jet hit the runway and skidded to a stop.
Nobody on the airplane died. Twenty-two passengers and three crew were injured but everyone was safely evacuated before fire consumed much of the airplane—and an explosion killed a firefighter.
The decision to go around was taken too late, but crash investigators always stress that they are not in the blame game. The aim of an investigation is to identify cause and, if that cause can be eliminated, to recommend how. In this case the crew had not been warned of dangerous wind conditions. So human actions, in the control tower or in the cockpit, created the failure.
On Sept. 13, 2015, involving a similar Boeing 777, a British Airways flight from Las Vegas to London was seconds from lifting off the runway when one engine suffered a catastrophic failure. The airplane was loaded with 20,000 gallons of fuel for an 11-hour flight. Nonetheless the captain successfully aborted the flight and all the passengers and crew safely evacuated as smoke and flames consumed the failed engine.
How would robots have handled each of these emergencies? At Dubai could a robot have detected the threat of the wind changes and powered up in time to make a successful go-round? At Las Vegas could the acuity of the robot beat the acuity of the human pilot and, in very few seconds, assess the crisis and abort with the speed that the British Airways pilots did?
Nobody can yet really answer those questions with any confidence.
As for the close call at San Francisco, a former chairman of the National Transportation Safety Board, Jim Hall, said the Air Canada pilots had been involved in “the most significant near-miss we’ve had in this decade.” The NTSB final report on the event has yet to be published, but everything suggests that but for the quick reflexes of the Philippine Airlines pilots in turning on their landing lights the Canadian pilots would not have been alerted to their error soon enough. So humans were able both to create a crisis and to avoid disastrous consequences. And that will be true for some time to come.