I could override it, but I'd have to justify why.
Decisions drift to the AI
Nobody formally gave the AI decision authority. But its recommendation gradually became the default, while the human reviewer became the person who approves it. Overriding the recommendation now takes more effort than accepting it.
- The AI's recommendation is accepted nearly every time.
- Nobody can say who owns the AI's settings or accuracy.
- Overriding it is rare and has to be explained. Agreeing never does.
Who: whoever acts on the AI's recommendations. Look at: a month of recommendations. Count: how often people overrode them, and whether those overrides turned out right. You're probably looking at this pattern if overrides are almost nonexistent and nobody owns the AI's accuracy. Probably not if people override when they should and someone tracks whether the AI is getting better or worse.
Reviews that go through the motions. That pattern is about a human review step. This pattern is about who is really making the call.
For the next month, write down each time you overrule the AI and why. At the end of the month, check which of those decisions turned out to be right.
Set one expectation out loud. If you are asked to do something you believe is wrong, by a person or by the AI, you are expected to say so.
People tend to rely on automated recommendations and miss their errors, especially when they are busy or overloaded (Parasuraman and Manzey, 2010). Established.