How this book uses AI
This knowledge base was drafted by AI agents. Here is what they could use, how their claims were checked, and what they cannot change.
How was this knowledge base produced with AI, and what keeps its verified core trustworthy?
AI agents drafted the text only from public project deliverables; a sample of 40 random claims was checked against the sources (37 supported, 3 corrected); every block lists its sources and consortium members decide what is published. Reader remixes and generated tellings are labelled and never overwrite the verified core; the MCP server is read-only and no personal data is processed.
- Drafted by AI agents from public deliverables only; every block ends with its sources
- Sampled fact-check: 40 random claims, 37 supported, 3 corrected
- Consortium members decide what is published; AI remixes are labelled and never overwrite the core
- The MCP server /api/mcp is read-only; no personal data is processed
How this book uses AI
main tellingdepth: intro..standard3 min markdown
You are reading a knowledge base written with AI. A chapter on AI ethics should say how, so this block applies the project’s Participatory AI Principles to the book itself.
How the text was made
- Drafting. AI agents drafted the text working only from public project deliverables, the reports the project has submitted or had approved for public release.
- Sources on every block. Each block ends with a Sources list: deliverable, section and public PDF link. You can open the passage and check the sentence.
- Fact-check by sampling. 40 claims were picked at random and checked against the sources: 37 were supported, 3 had to be corrected. That is a sample, not a proof: 3 of 40 (7.5 %) sampled claims were wrong before correction, so other errors are likely, and the sources are there for you to check.
- People decide. Members of the PoliRural+ consortium decide what is published. Consortium members can also draft or improve a block, and editors choose what enters the served book.
What AI may do here, and what it may not
- Remixes are labelled. When you ask the book for a different telling (shorter, another region’s example, another style), or an AI generates one, it is always labelled as generated and never overwrites the verified core.
- Bots read, they do not write. AI answer engines can query the book through its MCP server at
/api/mcp, which is read-only. - No personal data. The book processes no personal data.
Practice mapped to the principles
| This book’s practice | Principle it serves |
|---|---|
| Sources on every block | 1 Transparency & Explainability |
| Consortium members draft, review and decide | 2 Stakeholder Co-Creation; 3 Multi-Actor Involvement |
| Corrections from fact-checks and contributors flow back into the text | 4 Iterative Feedback |
| Sampled fact-check; uncertain numbers replaced by the name of their source | 5 Ethical & Responsible Use |
| Recall questions and games that train readers to question AI output | 6 Capacity Building |
| Generated tellings labelled, verified core protected | 1 and 5; in the spirit of the AI Act’s duty to tell people they are dealing with AI |
| Read-only bot access, no personal data | Data protection |
Two principles the production process cannot show on its own: inclusive policy design and gender equality and social justice depend on who takes part in the pilots, and they are reported there, through the AI ethics dashboard.
If you find a sentence its source does not support, treat it as one of the errors the sample predicts, and report it so it can be corrected.
Related
- Builds on: Participatory AI Principles
Sources
- Fact-check by sampling. 40 claims were picked at random and checked against the sources: 37 were supported, 3 had to be corrected . That is a sample, not a proof: 3 of 40 (7.5 %) sampled claims were wrong before correction, so other errors are likely, and the sources are there for you to check.
- People decide. Members of the PoliRural+ consortium decide what is published. Consortium members can also draft or improve a block, and editors choose what enters the served book.
- Remixes are labelled. When you ask the book for a different telling (shorter, another region's example, another style), or an AI generates one, it is always labelled as generated and never overwrites the verified core .
- Bots read, they do not write. AI answer engines can query the book through its MCP server at /api/mcp , which is read-only .
- No personal data. The book processes no personal data.