Skip to content
MNSTRY.org
Menu Explore

Topic · 5 articles

Machine judgment

Automation is usually argued as a question of capability, as though the only thing to settle were how much a machine can do. These articles argue the other question, which is what it may close. A rule that splits, a scanner that cannot tell a mention from a use, a claim format that will not let its author approve it, an evidence ledger honest enough to grade its own fix as unknown, and a layout broken by two components that each decided correctly.

Start reading The five-minute version

The through-line: Discernment and Cathedral of the mind and Structural, not behavioral , the long reads these articles expand on.

  1. 1 Start with the split that decides everything downstream, between the half of a rule a machine can settle from the artifact and the half that needs somebody to read the situation. The unambiguous half
  2. 2 Before a check can rule well, it has to read correctly, and the commonest misreading is mistaking an explanation for the thing being explained. Prose is not behavior
  3. 3 Move from what a machine may rule on to what it may ratify, where the cleanest control is a format in which the self-approving sentence cannot be written. The proposer cannot ratify
  4. 4 Ratification aside, machines and people both file findings, and the honest ones grade their own recommendation lower than the observations that led to it. The ledger says unknown
  5. 5 End where deciding goes wrong without anyone being wrong, in a system that let two parts settle the same fact and had nowhere to notice they disagreed. Two classifiers, two truths

You have walked What a machine may decide end to end: the split inside a rule, the misreading underneath it, the line at ratification, the honesty of an unknown, and the cost of two authorities for one fact. None of these turned on how capable the machine was. Every one turned on whether somebody had decided, in advance and in writing, which questions it was allowed to close.

Practice Take one automated check you rely on, a linter rule, a metric alert, a form validator, and write its rule out in full. Draw a line through it separating the half a machine can settle from the artifact alone and the half that is judgment being faked. Then demote the faked half to a note addressed to a person, with the reason it cannot be automated written into the check itself, so the next reader inherits the finding rather than the noise.

Follow this topic: RSS · agent version