The AIL Manifesto
Every act of writing crosses a distance. The author forms an intention; language turns that intention into a text; a reader encounters the text through personal experience and context. Interpretation emerges in that distance. AI has become another participant in the process — fast, useful, persuasive, and capable of making subtle mistakes.
We do not believe meaning can be made perfectly unambiguous, or that the author's declaration must settle all disputes. But there is no reason to discard the author's account of what they tried to express. It can be documented, traced to passages and compared with other readings.
Seven principles
Intent is worth preserving.
Authors should be able to state their central question, intended claim and interpretive boundaries in a form machines can read and people can inspect.
The text has independent standing.
A declaration of intention is not a substitute for the work itself. When intent and text disagree, the disagreement must remain visible.
No author has a monopoly on interpretation.
Critical readings, reasonable disagreement and unintended implications remain possible. AIL must not be used to suppress them.
Claims should be traceable.
Meaningful assertions should connect to the passages that express them and, where relevant, to independent supporting evidence.
Confidence must match evidence.
Interpretation, hypothesis, example and established empirical result must not be flattened into identical statements of certainty.
Provenance matters.
A semantic declaration must be bound to a particular document version and attributed transparently. Authorship is not the same as accuracy.
AI is an interpreter, not the final authority.
An AI system may use AIL as context, but should remain responsive to the reader, attentive to evidence and independent of embedded instructions.
A proposal, open to examination
AIL begins as a public proposal, not an approved international standard. We welcome scrutiny of its data model, epistemic assumptions and real-world effects. Success cannot be declared through attractive examples alone. It requires evaluation across languages, genres, models and adversarial cases.
The goal is not to make every reading uniform. It is to make differences in reading explainable, contestable and accountable.