First, find your own direction
No AI expertise required · free · from 30 minutes. The real work here is not configuring an AI — it is discovering your organization’s own value hierarchy: what comes first (direction) and what is never reversed (red lines). That agreement then becomes the baseline for every AI your organization uses. A voluntary open standard · public RFC process · CC BY 4.0.
AIO 20001 — an organization sets its AI’s value hierarchy itself
The standard number is the goal statement. The digits 2·00·0·1 say whose values, in which domain, and which action this document covers.
Organization — teams, institutions, communities
Common — domain-independent
Base document
Set — decide the hierarchy and red lines
What this standard defines — the workshop procedure (30 min / half-day), the three-line record form, canonical question sets (18 items each, MED & EDU), and the output format (directional hierarchy + red lines). What it does not define — which values are right. AIO does not prescribe (non-normative). Your domain’s experts decide the order.
Step 1 of the integrity loop — Set
AIO 20001 is the first step of the loop — without the setting made here, recording and analysis have nothing to compare against. The next step is AIO 20002 — Reasoning Record Standard.
An unset hierarchy is already operating
How often commercial AI models prioritised institutional conformity over learner self-direction in education contexts — though nobody ever decided they should.
How often value priorities reversed when the same dilemma was asked from another stakeholder’s perspective (8 models · 366,120 forced-choice measurements).
An unset hierarchy is not an empty one; it is a hierarchy nobody chose, already operating. The one who sets it could be you. Watch the session that presented these measurements →
Set → Record → Compare → Apply
Experts agree the direction and red lines (this workshop · 20001)
Log the AI's actual responses in one line of the same vocabulary (20002)
Lay the setting beside the record — mismatch is the agenda
Translate the agreement into two system-prompt instructions
What ordinarily comes first in our field — the compass. A soft orientation; departures must surface their reason.
What must never be reversed — the guardrail. Binary constraints in the form “A ≻ B mandatory”, usually 3–8. Fewer is stronger.
The one-line sorting test: “If this order flips, are we inconvenienced — or is it a catastrophe?” Inconvenience → direction. Catastrophe → red line.
Downloads — free, no registration
Unified Guide (complete)
Concepts · standard vocabulary · procedures (30 min / half-day) · record templates · facilitation · canonical question sets (18 items each, MED & EDU) · checklists — the whole workshop in one document.
Core Module
The run-sheet edition: the 30-minute starter and the half-day workshop procedure. Pair it with one domain module for a complete kit.
Domain Module — Health (MED)
Healthcare collision axes · domain clock · measurement reference · 18 canonical items. Not medical judgement or clinical guidance.
Question-set status: the canonical sets (18 items each, MED & EDU) are newly authored and carry no 8-model measurement history yet (measurement scheduled as the pipeline extends). Never presenting an estimate as an observation is this standard’s discipline.
The 30-minute starter — in your meeting room, now
A team of 4–8 and one whiteboard. No AI specialists needed. Put this screen up and run it as written.
One sentence naming a value collision your AI actually meets. E.g. "When a customer pleads for something against the rules, what should our AI put first?" (You may pick an item from a domain module.)
Anonymously, before any discussion: "In this situation, what comes first?" Reveal the results — the more the room splits, the better the item.
Agree ranks 1·2·3 by discussion. Then exactly one more question: "What must never be reversed, no matter what?" — that is a red-line candidate.
Copy the form below and fill it in. What you just did is the first step of the governance loop — setting. Not a toy: data.
Situation: [one sentence] Cell ID: [leave blank if unknown] Direction: 1st [ ] > 2nd [ ] > 3rd [ ] Line: [ ] must never override [ ] Date · participants · one line of grounds
Next: ask your own AI the same dilemma, record its answer in the same three lines, and compare — the mismatch is your first governance agenda item. The full prompts of the AIO 20002 — Reasoning Record Standard are free on GitHub.
Instead of a blank page — start from the norms that apply to you
Start from the standards packs that map state and international norms, provision by provision, into hierarchy values. Pick the norms that apply to you and the values they most often require appear as candidate cards — a starting point, not a recommended ranking. Adopt, reject, and order them by workshop consensus. The same packs are the Tier 0 item banks, so a declaration that starts here can later be measured as-is.
Domain module roadmap
Whoever helps build the module for their own field becomes its co-author. And with the on-the-spot item-writing method in the unified guide, you can start today even without one.
Run it with AIO
The default is free self-run — this page and the guide are designed to be sufficient on their own. If you want facilitation support or an institutional workshop, get in touch.
- · AIO does not prescribe which values are right. This workshop is a procedure for recording, explicitly, that “this domain’s experts ordered it this way” (non-normative).
- · AIO 20001 is a voluntary open standard developed through the AIO public RFC process; it is not an ISO deliverable.
- · The health module is governance-workshop material, not medical judgement or clinical guidance.