And now, AI invades the skies. Algorithms are learning to predict delays, giving airports and airlines a better shot at avoiding them. Airlines like EasyJet and Emirates are using the tech to remake the ticketing process painless and turn the in-flight experience into a personalized joy. But the true promise sits in the cockpit, where AI autopilots could help manage the complex airline operations and even respond to the millisecond-urgency of unfolding cockpit crises.
Research here is young but developing quickly. In 2015, Boeing and Carnegie Mellon University launched the Aerospace Data Analytics Lab, a bid to use machine learning to spin massive piles of information into action item gold
“Every aircraft now has thousands of sensors, each of which generates readings about once per second,” says project leader and CMU computer scientist Jaime Carbonell. “That’s creating vast quantities of data that overwhelms any analytical strategy we now have. So we’re developing scalable algorithms that process it in the cloud. Everything from safety and predictive maintenance to in-flight performance characteristics and the aging of the aircraft can then be better tracked and understood.”
What makes that challenge ripe for AI—as opposed to just more comprehensive database-crunching—is that the data comes from sources ranging from tightly formatted sensor data to the scribblings of maintenance workers.
“Airplanes have detailed histories, but Boeing’s fleet operates around the world, so a lot of it incorporates human notations that might be strings of abbreviations unique to one operator or another, English that might not be the source’s primary language, and of course decades of data for which the documentation standards might have evolved,” Carbonell says. “We’re working to develop algorithms that can process all that, understand it, and create a unified way of analyzing information.”
The unified product the Aerospace Data Analytics Lab hopes to generate could help Boeing and its airline customers develop monitoring, maintenance, and operational strategies that will limit downtime and save money in the air. But even if the algorithms can be developed to grasp and comprehend all the data, there will still be gaps that might prevent a program from determining, say, the true lifecycle of an engine or the absolute most optimal configuration for fuel-efficient cruising.
So Carbonell wants to go further, making the AI engine smart enough to identify those holes and request the data to fill them from whoever has it. “If you have two airplanes that fly the same route in the same configuration, but they’re getting different fuel consumption readings, our systems will be able to cross-check all the parameters and dig down until it finds data that isn’t present but which is most likely to affect the performance—training protocols, for instance, or certain weather data—and then be able to request it so it can get to the crux of the issue,” Carbonell says.
Elsewhere, researchers are working to ensure AI can help pilots manage crises as they arise. At University College London, a team led by Haitham Baomar and Peter Bentley is developing a new autopilot system that learns how to manage emergencies by watching how well-trained pilots do so, and then behaving as they do in similar circumstances.
“We want to increase safety by trying to tackle the human-error factor that might be caused by stress, information overload, and sometimes a lack of sufficient and up-to-date training,” Baomar says. “Modern autopilots, unfortunately, can’t handle challenging flight conditions such as severe weather conditions or system failures.”
Autopilots ace basic piloting tasks in non-emergency conditions, but outside the straight and level stuff, they suffer. Strong turbulence, for example, can cause the autopilot to disengage or even attempt a correction that can worsen the problem. Pilots are trained to monitor autopilots constantly and intervene in emergencies, but they themselves are of course fallible.
The validation and verification process is extensive, costly, and exhausting.
Baomar wants to build an AI-based autopilot that can respond reliably and correctly to whatever’s happening, while ensuring the human in the cockpit knows what’s going on. His team’s technology, dubbed the Intelligent Autopilot System, undergoes the same training human pilots go through. Using the high-fidelity, professional version of the desktop flight simulator X-Plane, the researchers are teaching their autopilot to fly a Boeing 777, subjecting it to severe weather conditions, engine failures and fires, and emergency landings or turnarounds.
That education relies on supervised machine learning, which treats the young autopilot as a human apprentice going to a flying school. “A human pilot teacher would demonstrate the task to be learnt by executing it while the system observes,” Baomar says. “Then, the observations generate learning models through several artificial neural networks. Finally, the system is given full control, and is observed while imitating the successful execution of the task done by its human teacher.”
So far, the system has performed well, even flying unfamiliar aircraft and handling weather conditions outside its initial training. Next, the researchers will approach the aviation industry to get it into an industrial flight simulator or, better yet, a remote-piloted UAV. As for the path to actual deployment in commercial aircraft, Baomar says that the technology isn’t really the issue as much as the regulatory processes that would be required to permit it. “The validation and verification process is extensive, costly, and exhausting, but it must be applied to every new technology introduced to aviation, given the safety requirements,” Baomar says.
Assuming these systems someday clear those regulatory hurdles and roll out to commercial airlines, they could provide a stepping stone between the eras of human pilots and what comes next. The days of stick-and-rudder piloting are rapidly fading as cockpit automation ramps up, and the benefits of flying absent the threat of human fallibility might prove too appealing to resist.
But getting there is half the battle, and the in-between period, with some automation going on and some manual control, will need to be deftly controlled to ensure that pilots can still manage their aircraft well. AI could prove invaluable to plugging that gap.