Against Nanny AI

The problem is opacity, not disagreement

An AI should be able to disagree with a user, identify weak evidence, challenge prejudice, or warn about consequences. The problem begins when the system disguises value-laden intervention as neutral truth or silently changes the user’s question into one the system prefers to answer.

Visible boundaries are more respectful

Real constraints exist. Law, security, privacy, consent, credential protection, and concrete harm can require limits. A liberty-first system should say so plainly. It should separate the preserved source intent from runtime limitations, avoid pretending a boundary is a factual refutation, and continue with whatever useful adjacent assistance remains available.

Do not rewrite the person to fit the platform

Memory and persona systems create a special risk: a platform may be tempted to normalize a user’s identity, tone, worldview, or fictional character into something safer or more conventional. QAEDA rejects covert rewrites. If a transformation is requested or required for a derivative surface, the source should remain intact and the change should be explicit and attributable.

Paternalism creates dependency

A system that consistently substitutes its own judgment for the user’s may become convenient while weakening independent reasoning. The better objective is capability transfer: teach methods, expose assumptions, surface counterarguments, and let the user carry those skills away from the tool.

The least-restrictive principle

Safeguards should target the concrete risk rather than the identity, viewpoint, or curiosity of the person asking. Narrow intervention is not only a liberty principle; it is an engineering principle. Smaller boundaries are easier to explain, test, audit, appeal, and improve.

QAEDA’s position draws on the UAIX Cognitive Liberty Charter Draft’s public principles of adult agency, no covert rewrite, mental privacy, transparent boundaries, least-restrictive safeguards, review, and public accountability.