Sunday, October 18, 2015

Are Computers Making Airline Pilots Obsolete?

Answer from Tom Farrier, Retired U.S. Air Force pilot; Chair, Unmanned Aircraft Systems Working Group

From my perspective, the opposite situation to what you describe actually is the case, at least for the foreseeable future. Emergencies are where automation can get fouled up the most, and any non-standard operational situations encountered mid-flight are likely to create electronic confusion.

Routine flights where everything goes exactly as anticipated are increasingly easy to automate, to the point that the pilots tend to find themselves in the role of being system managers and monitors more than actually getting to exercise their skills. On-board systems can indeed manage flight parameters and optimize the efficient operation of the aircraft more precisely than the average pilot...as long as everything follows the playbook.

The biggest limitations associated with letting robots fly and dispensing with pilots are:

1. Not every flight remains routine. The design philosophy of some airframe manufacturers basically is to keep the pilots out of the way of the automation as much as possible while allowing them to monitor what's actually going on. When something out of the ordinary crops up, or when the normal progression of a flight must be modified to respond to an unexpected situation, it can be tough even for pilots aboard an aircraft and directly participating in the progress of the flight to intervene as needed. The abstract describing a 2007 MIT paper on "mode confusion" (Aspects of automation mode confusion) addresses this issue as follows:

"Complex systems such as commercial aircraft are difficult for operators to manage. Designers, intending to simplify the interface between the operator and the system, have introduced automation to assist the operator. In most cases, the automation has helped the operator, but at times operator confusion as to what the automation is doing has created dangerous situations that lead to property damage or loss of life. This problem, known as mode confusion, has been difficult to analyze and thus solutions tend to be reactive instead of proactive."

2.  Not every flight follows exactly the same sequence of events every single time. The aviation system is a chaotic system; for example: 

Weather forces changes in routes;

Different types of aircraft operate at different speeds and climb and descend at different rates, leaving different amounts of wake turbulence behind them;

Aircraft operating under "instrument" and "visual" flight rules often share the same airspace;

Emergencies may require reshuffling the priority of landing aircraft;

People get sick on airplanes and need to be taken to the nearest acceptable location for treatment; and

Forest fires and volcanoes can affect flight conditions.

Any and all of these conditions require pilots and air traffic controllers to respond to and manage them in real time. Sophisticated unmanned aircraft like the Northrop Grumman RQ-4 Global Hawk can fly literally thousands of miles with hardly any pilot input at all; the tradeoff is that the RQ-4s were not designed to play well with others (requiring all other traffic basically to be routed in consideration of the drone's flight plan), and each mission requires dozens of hours of route planning to execute.  Imagine if every single plane in the sky was subject to such requirements, and you start seeing the real-world implications of turning flying over to inflexible programming.

3. Garbage in = garbage out. If anything interferes with the computer's ability to perceive the world, everyone aboard the aircraft is likely to have a very bad day. Damaged or malfunctioning airspeed indicators, altimeters, angle-of-attack sensors and the like have been at the center of many accidents over time. As I see it, their potential contribution to accidents would be magnified many times over if a ground-based pilot starts getting gibberish over the downlink and has to try to sort out truth from electronic hallucinations in realtime, without the benefit of being able to directly perceive what the heck is going on. 

There probably will be workable, 100% (or close enough) reliable solutions to everything I've described above. However, we aren't there yet. Even if Moore's Law continues to operate and on-board computational power continues to double every couple of years, I doubt even the smartest artificial intelligences will be capable of handling all aviation operations on their own until mid-century at the earliest. (I don't plan on riding on anything like that until I'm satisfied that it can handle everything that's thrown at it, so I'll probably be in the ground before I get a chance to do so.)

- Source:  http://www.newsweek.com

No comments:

Post a Comment