Artificial intelligence is efficient. But it’s not at all smart. Which is why a single potentially faulty sensor could have caused two Boeing 737 MAX aircraft to nosedive into the ground.

According to an investigation by The Seattle Times, air safety experts have focused their attention on a tiny fin that extends from the airliner’s nose.

It’s called an Angle of Attack (AoA) sensor.

It’s job is to accurately determine how much the aircraft is pointing up or down.

This one detail has crucial implications for an aircraft to stay in the air.

If the nose is angled too high, and the aircraft moving too slow, the amount of air rushing over and under the wings may not be enough to generate the lift necessary to keep it flying. It’s called a stall.

The artificial intelligence that comprises modern airline autopilots and safety systems takes input from a multitude of sensors spaced around the aircraft. This tells them everything from air speed, engine throttle settings, air temperature and height.

These algorithms then balances all these inputs into an optimal flight profile, designed to get the aircraft where it is going safely and economically.

But an AI’s capacity to deal with anything out of the ordinary is suspect.

STALL WARNING
The Seattle Times, which is based near Boeing’s major US assembly facility, reports investigations began even before the Ethiopian tragedy. They are focusing on the data the Angle of Attack sensors are feeding the MAX8 aircraft.

If an aircraft was detected an angle and speed at risk of producing a stall, this data should have triggered the Boeing’s Manoeuvring Characteristics Augmentation System (MCAS) artificial intelligence.

This fully automated system bypasses the pilot. It takes over the controls and forces the aircraft’s nose down to what it judges to be a ‘safe’ angle.

So what if this AI was being given false information about the aircraft’s actual angle of flight?

And was the Boeing 737 MAX8 making matters worse by being inherently unstable, the result of the design carrying heavier engines than originally intended?

Investigators say the flight characteristics of both the Ethiopian Air and Lion Air MAX8 aircraft before they crashed was consistent with pilots ‘fighting’ the aircraft’s computers.

The pilots were attempting to pull the aircraft’s nose back up to a safe level.

The computer appears to have been repeatedly forcing it back down.

REDUNDANCY REMOVED
There are two Angle of Attack sensors fitted on the Boeing 737 MAX8. This is standard practice for any piece of equipment that could potentially lead to catastrophe if it failed.

But Boeing didn’t assess the AoA sensor’s role as being that critical.

The Seattle Times report says this is why only one AoA sensor was actually attached to the MAX8’s MCAS autopilot.

“But when the consequences are assessed to be more severe, with a “hazardous failure” requirement demanding a more stringent probability of one in 10 million, then a system typically must have at least two separate input channels in case one goes wrong,” the Seattle Times reads.

“Boeing’s System Safety Analysis assessment that the MCAS failure would be “hazardous” troubles former flight controls engineer (Peter) Lemme because the system is triggered by the reading from a single angle-of-attack sensor. ‘A hazardous failure mode depending on a single sensor, I don’t think passes muster,’ said Lemme.”

OVERPOWERED AI
The MCAS autopilot is supposed to be very gentle when it seizes control of the MAX8.

Boeing reported to the US Federal Aviation Authority that the AI would only adjust the rear stabilisers (the horizontal fins on the aircraft’s tail) in increments of no more than 0.6 degrees.

But the system, as actually fitted to the MAX8, can force the tail stabilisers to jump in increments of 2.5 degrees.

The upshot of this is that the AI could quickly put the aircraft into a nosedive, with little opportunity for the human pilots to react and counteract its intervention.

Pilots would have no idea the AI was being fed false information, the report said.

And every time they moved to pull the plane back into a safe angle, the MCAS would quickly force it back down again - not knowing that the pilots were attemping to over-ride it.

Black box data recovered from the Lion Air disaster show this pilot-autopilot struggle appears to have happened 21 times before the aircraft ploughed into the sea.

“The investigation is still ongoing, so neither Boeing or the FAA are able to comment on it directly, but the information so far suggests a situation where inadequate testing and rushed practices led to some real disasters,” the Seattle Times concludes.