An industrial process can run automatically for long stretches. When something unusual happens, the person watching it may be expected to take over with little warning. In 1983, psychologist Lisanne Bainbridge pointed out a problem with that division of labor. The operator may have to use skills the automation has given them less reason to practice.

Archive record · Public domain status
Taking over can mean both stabilizing the process and diagnosing what went wrong. Bainbridge argued that these are demanding jobs, especially when the process has already entered an abnormal state. A person who mostly watches the machine may have fewer chances to build a feel for how it responds.
The paper studied industrial automation, but its question applies wherever an AI system is supposed to call for help. What must the person know at the moment of handoff, and how will they have kept that knowledge current?
