‹ Culture 3.14 · What we ask ourselves
Question · Decisions
When should a system recommend, decide, or do nothing but present evidence?
The three stances use the same model and distribute responsibility in three different ways.
Why it matters
A system that decides takes on a responsibility that has almost never been formally granted to it. One that only presents evidence hands the whole thing back to someone who was already saturated with information.
Between the two sits the recommendation, which is comfortable for the design and ambiguous for the organisation: it looks as though somebody is deciding, but nobody signs for it.
What we know so far
The cost of the error orders the choice fairly well. When an error is cheap and reversible, deciding automatically frees attention for what matters; when it is expensive or hard to reverse, well-presented evidence is worth more than a verdict.
Declaring the stance explicitly in the interface has reduced more misunderstandings than improving the model’s accuracy.
What remains open
We do not know how the stance should change over time. A system that starts out recommending and gets it right for months generates a confidence no metric has authorised.
Nor is it clear who should be able to move that boundary inside an organisation, or by what procedure.