Artificial Intelligence and the "Single Pilot" issue

I would like to start by introducing Mr. Stuart Russell: Stuart Russell is co-author, together with Peter Norvig, of a text that can be considered the Bible of artificial intelligence: "Artificial Intelligence: A Modern Approach". In the sentence in the picture, about driverless vehicles, Russell mentions something that all Human Factors specialists know very well:

Extreme automation, when it fails and returns control to the human, presents a situation where vital cues to decide could be missing. These cues would have been appearing with the driver involved in the process but, when the system returns control to the driver, they are absent and, therefore, the human as an emergency resource does not function. It is like a fireman whose hose has been removed.

Here are some Aviation-specific situations that could help clarify what can and cannot be done. This is especially relevant at a time when major players are trying to sell the idea of "single pilot" aided by new-generation automatic systems and artificial intelligence. Let's look at it:

• Can you imagine a situation where, with five pilots in a cockpit, an event occurs that gives work to all of them? Don't imagine; it exists: QF32.

Among the many things that happened on this flight, an engine explosion cut data wires, connecting sensors to the information systems they served. Consequently, the pilots had to perform many of the calculations by hand, with the information available or by performing experiments to know the extent of control of the aircraft.

• Can you imagine a situation in which all the crew was required and, in addition, they needed the help of another pilot flying as a passenger? Don't imagine; it exists: U232.

An engine explosion cut the hydraulic lines to the controls and a pilot, flying as a passenger, was operating the engine throttle levers, since this was the only accessible control.

• Can you imagine a situation in which the failure of a sensor produced an anomalous reaction in the systems resulting in a condition of complete confusion? Don't imagine; it exists: AF447.

The freezing of a sensor led to erroneous speed data and incorrect information, to the point that, at times, the alarms did not work because the aircraft's systems interpreted that the aircraft was not flying but was on the ground.

• Can you imagine a situation in which the freezing of two sensors produced an anomalous behavior of the aircraft without time to recover it? Don't imagine; it exists: XL888T.

The freezing of two sensors caused the systems to misinterpret the situation, raising the nose of the aircraft and the aircraft not responding to the controls. Subsequent investigation indicated that the aircraft would have responded to the compensator.

• Can you imagine a situation where both engines of an airplane stall just after takeoff and where there are no procedures adjusted to the situation and the systems do not give enough information to know if they could reach the airport? Don't imagine: US1549.

In the famous "Hudson River landing", the double engine shutdown was foreseen by the manufacturer above 20,000 feet, not during the initial ascent. The checklists were long and unhelpful, and the systems could not give information on the ability to reach the airport or not.

We could add many more cases to this list, but these are enough to illustrate an idea: Advanced systems, with or without artificial intelligence, can be a help when everything is going well but, in certain circumstances, they not only do not help but can contribute to aggravating the situation.

What would happen in the situation of a single pilot, supposedly assisted by advanced information systems? The cases are pretty illustrative: An unforeseen situation, a sensor failure, or a disconnection between the sensors and the systems they serve, can make a situation unmanageable.

Obviously, the above is merely an approximation of the problem. It can be treated in much greater depth without falling into either such classic extremes as essentialism or discussion of algorithm design.

It seems that the Peter principle - the idea that everyone rises to their level of incompetence - is not restricted to organizations but to the resources they use. Technological progress is welcome, but it must be made clear where it does not reach and why.

Today, a detailed analysis of the cases mentioned above -and the many more that could be added- would make it clear why the single-pilot idea is not a valid option.