Treat AI as an Autopilot, Not a Copilot
Microsoft and Salesforce both have used marketing to label AI as a copilot, although this is not the appropriate metaphor from aviation. As a pilot with over 600 hours in the cockpit, AI is an autopilot, not a copilot.
Buckle your seat belt, because we’re about to go on a little ride through why their metaphor doesn’t make sense.
When there are two pilots in the cockpit, one is the “pilot flying” (PF) and the other is the “pilot monitoring” (PM).
The PF’s primary responsibilities are to fly the airplane, which includes doing so with their hands and feet during takeoff and landing, make decisions that automated systems cannot do, and delegate tasks to the PM
The PM’s responsibility is to monitor instruments and navigation, flag deviations, run checklists, and talk on the radios
In the event that the PF becomes incapacitated or ineffective, the PM assumes responsibility for flying the airplane, and they’re carrying the same certificates and required training as the PF.
As a passenger, you don’t care which pilot is the PF or PM.
Pilots rely on autopilots for the mundane tasks to reduce fatigue for the critical phases of flight, which include takeoff, initial climb, final descent, and landing (with a few exceptions, as auto land capabilities do exist).
A core point hammered in during training is to hand fly the airplane and reduce the use of automation (i.e. autopilot) when things don’t make sense.
Also, an autopilot can turn itself off when it exceeds expected ranges, which can occur during severe turbulence or when sensors provide data that is inconsistent.
With that background, let’s land this and return to talking about AI.
AI is not capable of and should not be expected to replace your full set of responsibilities
Most AI does not have the ability to turn itself off, instead it will provide responses that it will not generate or has low statistical confidence
AI requires monitoring to ensure that its performance is value additive
Full situational awareness is not available to AI, which means its outputs are based on a subset of available information
AI is trained and operates with a set of predefined boundaries, and cannot operate effectively outside of those
TLDR: AI is your autopilot. Use it to reduce fatigue, don’t use it during critical events, and always monitor it.
As an aside, the next time you’re flying commercially, here are a few fun things to notice:
During initial climb, you’ll likely hear the engine noise change shortly after takeoff, which is when the throttles are pulled back from takeoff setting to “maximum continuous thrust”
When the gear goes down (wind noise and a slight lurch forward, followed by the engine spooling up), you’re essentially over the “final approach fix”, which is typically 6 miles from the runway; you have approximatley 3 minutes left until touchdown
If you’re seated in the first 5 to 10 rows, you might be able to hear the pilot disconnect the autopilot, which is typically three chimes
