AI
The movement uses AI heavily, and it is worth being blunt about the shape of that, because "an AI scores politicians" is a fair thing to be suspicious of.
The AI is a verifier, not an oracle. Its job is to check whether reasoning holds, to read enormous amounts of material consistently, and to apply rules that were written down in public before it ran. It does not decide what is true, and it does not decide what you should think.
What it actually does
- Checks a chain of reasoning from first principles. Not an authority said so — does this actually follow? This is the machinery behind Pillar 4: a citizen with 95% of an evidence chain can have the logic checked rather than being told to wait for the missing 5% that was withheld on purpose.
- Reads long-form material against a published rubric. Hundreds of hours of interviews, assessed against criteria written down before anyone was assessed. Consistency is the whole value: a rubric applied identically to everybody is harder to lobby than a panel.
- Applies your value system, not ours. Pillar 3 exists so a citizen can write down their own rules for deciding this versus that and have those applied at scale — to their standard, not the movement's.
- Summarises a month, advisory only. The monthly karma pass writes a narrative, notes standouts and integrity flags, and nominates history-book entries. It never awards a point.
What it never does
It never approves anything, publishes anything, sets an award, or overrides a human. Every path from an AI output to something the public reads passes through a person.
The arithmetic is not the AI
Worth separating clearly: the awards are published equations over recorded acts. You can reproduce them with a calculator. No model is in that loop. Where AI appears, it is reading material and checking reasoning, and its outputs are inputs to a human decision.
Pages in this area
- What the AI does — four jobs: checking reasoning, reading long-form material against rubrics, applying your values, and summarising a month.
- What the AI never decides — the boundary where a model's output stops and a human decision starts.
- Our hard rules — rules the codebase enforces on itself, including one that fails the build.
Where to go next
- Open Data — the area before this one.
- Repos — the area after this one.
- Take Action — what to actually do with this.