91ɫƬ

Who do we trust when human and machine intelligence disagree

A faulty sensor – and the automated action it led to – are being blamed for the loss of Lion Air flight JT 610. Is it time for AI to take a back seat, asks Peter Lemme

plane cockpit

ON 29 October, Lion Air flight JT 610 crashed into the sea off Jakarta with the loss of 189 lives. The investigation continues, but flight information recovered from the wreckage indicates that the pilots were battling with the autopilot. In a situation like that, who should we rely on: human or machine?

The automated system was supposed to correct the aircraft’s pitch. Instead, it repeatedly pushed the nose down towards the ground. Thirteen minutes after take off, the plane was lost.

Every time something is added to make a plane safer, new ways are created to make it less safe. In 1988, the first commercial jets arrived that made machines the leaders. The pilots can only operate within boundaries set by the machine. For the first time, pilots trusted the onboard artificial intelligence to stop them from doing something stupid.

Pilots can’t know everything that is happening. To prevent them being overwhelmed with information, the data made available is only that which is deemed necessary to take the required action. Problems noted by the aircraft that don’t require pilot action are logged for the maintenance crew to read later.

“The data made available to the pilots is only that deemed necessary to take the required action”

Thus a pilot’s awareness must stretch across multiple realities, in which the aeroplane behaves differently depending on its status. These realities are distinguished by all the combinations of failures that might be encountered.

Many of these combinations can never be fully anticipated, even by a machine intelligence. Humans are able to adapt to new realities better than any AI, especially because they hold human traits like sacrifice, trust, judgement and a disparate knowledge base to draw from.

There is no simple answer to whether humans or AIs should take the lead in cockpits. As a leader, the pilot questions the machine when it misbehaves. As a follower, the pilot questions themselves when the machine intervenes.

A machine is trustworthy only while everything in the plane is working optimally, and JT 610 had a record of maintenance issues.

Pilots must be ready and able to transition from follower to leader when the situation calls for it.

This article appeared in print under the headline “Wing and a prayer”

Topics: Aircraft / Artificial intelligence