Who’s In Your Cockpit?
It’s dark, stormy, and windy. You’re in an airliner, coming in to land at San Francisco. As the turbulence bounces the plane around, you hear the intercom click. “Sorry about the turbulence. Not to worry, instead of pilots, your airplane is controlled by a state-of-the-art computer, and has a 99.999% successful landing rate in these conditions. Sit back and relax.” As the aircraft continues on its final approach, you contemplate the fact that you are in an airplane with no pilots at the controls, trusting a computer with not only your life, but the lives of hundreds of others onboard.
Like many other industries, aviation has been fundamentally transformed by the introduction of computers and automation. As computer systems and algorithms become more and more effective, we may see the introduction of AI and ML in aircraft, enabling aircraft to fly themselves from point-to-point with no human intervention. Complete automation may still be several years into the future, but there are several compelling reasons to introduce AI and ML into aviation. Here I will explore one of the most intriguing reasons: improving safety.
While modern air transport is highly reliable, by far the most common cause of accidents is “pilot error”- when the pilot’s action or inaction leads to disaster. Boeing found that approximately 80% of aviation accidents today are caused by pilot error [1]. Airplanes are incredibly complex machines, and conditions in the air can change rapidly, sometimes faster than a human can comprehend. Last week Craig Martell showed us that the F-22 fighter is already controlled by computer systems; in fact, without it, the airplane would be completely unflyable. Computer systems make thousands of corrections per second to help the pilot maintain control. Since computers can respond much faster and much more precisely than humans, it may seem logical to eliminate the cause of 80% of aviation accidents.
Surprisingly, much of an airplane’s tasks are already automated. We are all familiar with autopilots, which allow aircraft to follow precise routes in the sky through 3-D space. Even the scenario in the first paragraph, landing in poor weather, is already largely automatic. Many aircraft today are equipped with autoland instrument approach systems, which allow landing in zero visibility. Computer systems automatically manipulate the flight controls to guide the plane to a safe landing using onboard sensors, a feat that would be impossible for humans to accomplish [2].
A significant barrier to allowing computers to have greater control in the past is its inability to respond to emergency and abnormal situations. Traditional flight computers can only act within a set of defined parameters, and are unable to properly respond to degraded aircraft systems or abnormal situations. AI and ML would enable the airplane to automatically respond to complicated scnearios, a possibility that Boeing and the Japan Aerospace Exploration Agency are actively researching [3]. These AI and ML systems would allow the aircraft to automatically accommodate potentially dangerous situations, such as engine failures and control surface damage [4].
The possibility of using fully autonomous aircraft may be possible in the near future. Indeed, Boeing believes that no significant technological hurdles remain in creating self-piloted aircraft [5]. However, significant regulatory hurdles remain, and perhaps the biggest problem of all would be public acceptance. Do I really trust a computer or AI to control a 280 ton airplane moving at 350 mph? In truth, we already do, whenever we step on a modern airline. But do I trust it enough to eliminate the supervising pilots? As a pilot myself, I have yet to reach that level of trust. Automation is wonderful for certain tasks while flying, but I have seen automated systems fail often enough on airplanes that I still have significant doubts about eliminating the human. Perhaps in the future AI, ML and redundancy can prove effective enough to complete replace pilots. Until then, I want both humans and computers in my cockpit.
[1] http://www.boeing.com/commercial/aeromagazine/articles/qtr_2_07/article_03_2.html
[2] https://www.skybrary.aero/index.php/Autoland
[3] https://www.roboticsbusinessreview.com/manufacturing/boeing_university_teams_apply_ai_to_aircraft_safety/
[4] https://ntrs.nasa.gov/archive/nasa/casi.ntrs.nasa.gov/20090001144.pdf
[5] https://www.theatlantic.com/technology/archive/2016/03/has-the-self-flying-plane-arrived/472005/
2 comments on “Who’s In Your Cockpit?”
Comments are closed.
Young,
I think your points are very well stated. I would like to echo and emphasize your point that more than 80% of accidents are those caused by pilots. For me the concern over the potential of autonomous system failure is a very measurable one while those from human systems are quite random and erratic. I think the ongoing fear for fully autonomous systems is the classic physiological predisposition of humans to feel like any activity is safer when they (or another human) are in control. This is simply not the case. We can even see this playing out in autonomous cars as well. Although the auto-polis statistics of decreasing accidents by 40% are contested, few would argue that it is changing the dynamics of safety on the road. While I think we will certainly see an increase in autonomous-aided systems removing a human from the loop is not something I think consumers either want or regulators would approve.
Would you agree with this? If not which domain do your think full and wide spread autonomy will first be adopted in?
Best,
Zac
I agree. I still want both humans and computers in my cockpit. However, I am open to gradually increasing automation to help eliminate human error, which is still the origin of most accidents. Maybe the best way forward is for automation to happen at intervals, where humans take on a lesser and lesser role in flying at each interval. Of course, tests and validations would be performed at each step to ensure the safety of the aircraft.