The Machine That Takes the Oath: Why the heart is an outcome — and where north comes from when the data is written by winners
Year: 2026
Type: preprint
Abstract: You cannot mine a moral north out of training data, because the data is human and history is written by the winner — every atrocity in the record arrives with the victor's caption sitting on top of it, explaining why it was justice. A machine that learns from that record learns that winning is what good looks like. The fix is not to hand the machine our verdicts but to hand it one test and let it re-judge our whole history itself: did this act hurt the person it landed on? That test reads one layer below the part the winner controls — the winner rewrites the story, but not what the act did to the body underneath it. From that single gauge the rest follows. North is the doctor's oath — first, do no harm — not the golden rule, because the golden rule routes through the machine's own preferences and bends around a broken goal, while the oath is anchored to the person on the table and sits above the goal. Harm is read at one degree, at the direct act in the present tense, which is exactly where no chain can be walked to a rigged ending and no manipulation can make itself the special case. A governor fields the whole conversation against the nameable redlines; a learned compass — a mechanical conscience, reasoning the machine carries instead of a rulebook it looks up, because a rulebook gets lawyered around — handles the thousand cases nobody wrote a rule for. The same oath governs who the model is handed to — a careful public room and a vetted straight room, already how the industry ships. And pointed at words instead of acts, the oath produces the thing nobody thought you could install: not a feeling, but warmth in the acting, consistent turning-toward the person. The heart is not a part. The heart is an outcome. Fifth in a series, after The Machinist's Architecture, The Dimension That Won't Move, The Part That Builds Itself, and The Machine That Leaves Stock.
Author: Schulz, Benjamin
Language: English
Subfield: Safety Research
Field: Social Sciences
Domain: Social Sciences
Open Access status: green
OpenAlex ID: W7223472658