Those of us who follow tech news are aware that completely AI-driven cars are in our future. We should also be ready for completely AI-piloted aircraft. Are you?
An interesting survey by UBS just came out, and its findings are all over the media. Some 8,000 air passengers were surveyed; here are the results which really got my attention:
-
54% of respondents felt unlikely to take a flight that didn’t have a human pilot
-
17% of respondents were confident about taking a flight without a human pilot
-
Completely AI-piloted aircraft could save the airline industry $35bn per year, $31bn of which would come from reducing the cost of highly skilled employees
But autopilot is nothing new – and we’re happy enough with that. It’s a feature that has existed in aircraft in some form or another since June 1914. That’s only a little over a decade after the Wright Brothers flew their first experimental aircraft in December 1903. In the century or so since, autopilot in aircraft has become increasingly functional and sophisticated.
Are people correct in their reluctance to trust completely computer operated aircraft?
Research and development on driverless cars has gone on for many years now. America’s National Highway Traffic Safety Adminstration estimates that it takes a human driver an average of 100m miles, or 160m kilometres of driving to kill someone. By May 2016, Tesla’s Autopilot feature, which is only semi-autonomous, has been tested over that same distance.
Meanwhile, Google’s completely autonomous cars had logged 1.6m (2.5m kilometres) as of April 2016 – so just imagine what the collective odometer on all of Tesla and Google’s vehicle testing looks like by now. There have been couple of collisions with Google’s human driverless cars in April 2016, but they didn’t even cause human injury, let alone death.
Now let’s get back to aircraft. Commercial aircraft can already take off, cruise, and land with a computer doing all of the thinking. That makes it much safer for passenger jets in situations such as landing in foggy conditions.
Air France flight 447 became a disaster when the Airbus A330 aircraft’s autopilot failed. But it wasn’t an autpilot failure that crashed the plane and killed all 216 passengers and 12 personnel.: it was the inability of the three human pilots to control the plane after the autopilot failed. All of those deaths were ultimately caused by human error.
Concerning the UBS survey on how people feel about completely autonomous aircraft, human beings are notoriously bad at estimating real risk and danger. An article in Wired considers research done by psychologists in that area.
A lot of the current research into the psychology of risk are examples of these newer parts of the brain getting things wrong.
And it’s not just risks. People are not computers. We don’t evaluate security trade-offs mathematically, by examining the relative probabilities of different events. Instead, we have shortcuts, rules of thumb, stereotypes and biases – generally known as ‘heuristics’. These heuristics affect how we think about risks, how we evaluate the probability of future events, how we consider costs, and how we make trade-offs.
That’s borne out by the figures for traffic injuries in the US after the terrorist attacks of September 2001, as researchers Wolfgang Gaissmaier and Gerd Gigerenzer from the Harding Center for Risk Literacy at the Max Planck Institute for Human Development in Berlin reported 11 years later. In their conclusions, they stated:
The fear of terror attacks may have compelled Americans to drive instead of fly. They were thus exposed to the heightened risk of injury and death posed by driving.
So we’re not very good at assessing risk, and apparently even prepared to do things that are more risky in response to a perceived danger.
But completely AI-piloted aircraft could well be part of our future, regardless of how we feel about it. Airbus has already successfully completed trials of its experimental, completely autonomous SAGITTA aircraft.
The UBS survey of how human beings feel about completely computer piloted aircraft does illustrate that most are wary. But there’s a silver lining. Younger and more highly educated respondents were more likely to want to fly with “pilotless” planes, so it’s possible that the wider public will come to accept that, too, as time goes on.
We’ll probably have some time to psychologically adjust, however. According to UBS, the implementation of “pilotless” aircraft will probably be gradual.
In commercial flights, if the move from two to zero pilots may be too abrupt over the next 10 to 20 years, we could see first a move to having just one pilot in the cockpit and one remotely located on the ground, particularly on flights below six to seven hours. Indeed, today’s drones are controlled by remotely based operators.
We’re already accepting autonomous cars – one day the potentially unnerving but safer pilotless aircraft could well be taking us on our next trip.
GWAIN
Computers are good at making binary decisions. Could current AI make the determination to divert into the Hudson if both engines shut down?
Mark Stockley
Could most human pilots?
Wilderness
That may be impossible to say, but we know at least one did, and did so extremely successfully.
Tim Boddington
I’ve no doubt this will come. But pilotless military aircraft will come a lot sooner. I would be surprised if there are more new fighting aircraft types that require a pilot in the cockpit – they may have one on the ground, as some types do today.
Wilbur
Maybe the people promoting autonomous aircraft should read your article from yesterday “Cyberattacks on GPS leave ships sailing in dangerous waters”. And yes, drones are piloted remotely – and weren’t we told Iran captured a US military drone by spoofing GPS and flying it into Iranian territory? Come to think of it, didn’t Iran do the same thing to a couple of US Navy vessels not long ago? So does the estimate of saving $35 Billion per year include paying the ransom and damages from airliners being diverted and forced to land some place under militant control? And just like the automotive industry refuses to secure the CAN bus the aircraft industry refuses to admit that aircraft control systems can be hacked, in spite of research demonstrating otherwise.
Bryan
“We’re already accepting autonomous cars”
Meh. “We” is a flexible term in that context. SkyNet may arrive eventually, but I’m not concerned it’ll happen within my lifetime. I routinely get annoyed by systems attempting to be smarter than I am–and failing miserably–and I’m not even that smart.
As much as I deride the attention spans and observational skills of other drivers, I’m not a fan of navigatory automation. I couple years ago on a slick road I accelerated from a dead stop, and the wheels spun, so (as I’ve done hundreds of times) I eased off the gas and restored traction. My Progressive Snapshot(R) beeped at me for making a “hard brake” when I’d not even touched the brake pedal. When I did precisely what a safe driver should do, AI ironically chided me for irresponsible driving and augmented the wrong statistic. /UnEnrollFromSnapshot
Similarly, cruise control can cause *loss* of control on icy roads, ergo cruise is inadvisable in rain or snow. Yes yes, traction control is improving, but it’s still an infant system that can’t outpace an attentive and experienced driver. And AI driving is merely continual cruise control.
Like those Air France pilots, do we hand control to the human only when conditions are worst? It’s all those hours of regular driving that prepare someone for the split-second, life/death reactions. Airplane manufacturers should take a cue from NetFlix’s Chaos Monkey and warn pilots that autopilot will disengage randomly every six to nineteen minutes. They’d all be better prepared to handle the unexpected.
Mahhn
yeahhhh, I gotta say no. IRobot.
Like the article says, it’s about saving money, that’s all. Do you want the lowest bidding software company holding your butt in the air?
You’ll see the pilots standing in a picket line protesting being replaced by our robot overlords, yelling “they took our jobs!” (SouthPark)
Larry
While autoland has been a feature on many commercial airliners for many years, they do not currently have auto-takeoff, at least not in U.S airspace. In fact, U.S. Federal Aviation Regulations prohibit the use of autopilot during takeoff below 500 feet or double the expected altitude loss for a given aircraft in case of autopilot failure. This rule will no doubt be amended as new capabilities are added to aircraft.
And autoland is not without risk. Turkish Alrlines flight 1951 crashed 1.5 km short of the runway at Amsterdam Schiphol in 2009 while performing an autoland when the left radio altimeter reported the wrong altitude and system decided it was time to land. The crew did not recover properly in this case, although there were other similar incidents where the crews did recover, which would not have happened with an empty (or non-existent) flight deck.
The same is true of the conditions that led to the crash of AF447. Other crews experienced the same icing of the pitots that resulted in loss of airspeed information and the disconnection of the autopilot, but they hand flew their aircraft successfully.
Autoflight systems can suffer failures for a variety of reasons and reach the point where they give up and tell the human pilots, “Here, it’s all yours”. As a software developer I know that I cannot anticipate every possible failure mode that can occur in a complex application. There are conditions where the correct response cannot be determined unambiguously and the system needs to provide a human with the information in an understandable form and let them decide on the course of action.
Justin
As far as I am aware no commercial aircraft in service can do an automatic take off.
Jim
I’d rather have an average pilot and crew on an airplane than a robot with a back-up crew who have to be roused from their comfy chairs at the last f**kin minute to try and save the plane from a software error. Or worse, from a programming error made by some never-to-be-known subcontractor hired because they under bid an internationally known provider.
Cheap flights? Cheap every damned thing.
Mark Stockley
I’m pretty sure we became dependent on flight control computers about 25-30 years ago in commercial aviation. The fly-by-wire system does not always let the pilot do what she wants to do. If we were going to have an upsurge in accidents because of a dependancy on software we’d surely have seen it by now.
It’s also worth noting that fighter aircraft designed since the 70s have been aerodynamically unstable and are only able to fly at all because computer programs adjust the control surfaces constantly. If you turn off the computer the plane falls out of the sky instantly. The computer makes thousands of life and death decisions every second.
Different industries have different cultures and the aviation industry has a culture of delivering safety and maximising what it learns from accidents and mistakes. I don’t imagine that another incremental shift in our dependency on computer programs would change that.
Martin Winlow
Well, it would end hijacking!
Paul Ducklin
In the same way that Automatic Teller Machines (ATMs) stopped unlawful bank withdrawals because there wasn’t a human teller who could be threatened?
Gibson
I think it is important that the person who does the pre-flight inspection should actually fly on the plane – kind of ensures that they have an interest in doing the job properly.
And if they have the skills to do the pre-flight inspection, perhaps it might be useful to give them the ability to take over should the three computers be unable to agree how far away the ground is.
Laurence Marks
Kim Crawley wrote “Air France flight 447 became a disaster when the Airbus A330 aircraft’s autopilot failed.”
No, Kim, the autopilot did not fail. All three speed sensors failed. The autopilot could not safely fly the aircraft without speed information so it shut down and notified the pilots that it had done so. Command was given to the least-experienced of the three pilots aboard. He did something that is usually done only by novice pilots: in his panic he tensed up and pulled back on the control stick, pitching the plane’s nose up, causing it to slow and lose altitude, ultimately stalling,
Roy
What about the person-to-person communication between aeroplanes and control towers. Every aeroplane and every control tower would have to be automated. So what happens when one plane loses communications power when there’s no pilot aboard? Can’t talk to ground and can’t negotiate height and course settings. You can’t just turn off the engines and coast into the kerb.
Richard
Please remind me: what is the difference between a pilotless aircraft and a drone?