This essay was originally published at the end of 2017. It’s recently been updated with the assistance of Carl Alviani.

Expecting a human to deal with all the interruptions, stresses and invasions posed by networked machines is a bit like asking a goat to get through a salt-lick the size of the Eiger. To get around this problem, machines need to deal with these interruptions and stresses on our behalf. This is not a trivial project.

For years, we’ve struggled to create rulesets that can be automatic for day-to-day activities, free from common errors and pitfalls, and beneficial enough to actually be used and appreciated by users. But these rulesets were all based on trust in a machine or mediator — trust that we increasingly realize is misplaced. …