Commitment
A human stays the decision-maker
Our systems inform a judgement. They never hold final authority over an outcome that belongs to a person.
Not revised since publication
Where a decision affects someone, a person makes it. Our software can gather evidence, structure it, and say what it thinks. It does not get the last word.
What this looks like in practice
§- A conclusion our software reaches is visible to the person it concerns.
- The reasoning behind it is inspectable, not a score with no account of itself.
- A person with the authority to disagree can do so, and their disagreement is the outcome.
An output nobody can argue with is not evidence, it is a verdict. We do not build verdicts.
Why this is a capability argument, not a sentimental one
§Handing a judgement to a system degrades the judgement it replaced. The skill erodes because it stops being exercised, and the erosion is invisible until the system is wrong and nobody remaining can tell.
Keeping a person in the decision is therefore not a courtesy extended to them. It is how the competence that makes the decision worth anything survives contact with the tool.
What this rules out
§- Automated determinations that take effect without review.
- Confidence figures presented as conclusions.
- Interfaces that make accepting our output easier than questioning it.