In multilingual governance environments such as the European Union, policy discussions often evolve through successive interpretations across languages and institutional perspectives.
A simplified example may look like this:
French proposal
↓
German interpretation
↓
Italian concern
↓
Spanish amendment
During such discussions, the meaning of the original proposal may gradually drift.
The issue is not simply translation accuracy, but differences in assumptions, intentions, and risk perceptions among stakeholders.
In practice, participants may agree on the wording of a policy while still holding different underlying interpretations.
One possible approach is to document stakeholders’ reasoning and intentions in natural language during the discussion process.
A framework such as the C-I-B-R structure (Concept, Intent, Boundary, Rationale) allows these perspectives to be captured transparently.
Generative AI can then assist by analysing these natural-language records and extracting meta-level concepts that appear across the different positions.
This helps participants see where their assumptions diverge and move beyond formal agreement toward a personally grounded understanding of the underlying concepts.
Importantly, such an approach does not aim to control AI systems.
Rather, it supports human understanding, helping stakeholders maintain a shared conceptual reference point during complex discussions.
This may contribute to greater transparency and accountability in AI-related decision processes, particularly in multilingual governance environments such as the European Union where shared meaning across languages and institutions is essential.
https://github.com/Shiraki5995/ai-governance-sandbox-2-multilingual
- Značky
- ai regulation Member States
- Prihlásiť sa na účely uverejňovania komentárov