Theory of Freedom — a hard legitimacy floor for machines that act

[Aliipou]

This repository is the public home of Mohammad Ali Jannat Khah Doust, نظریه آزادی، ایران و دین (Theory of Freedom, Iran and Religion), CC BY 4.0, plus a sixteen-chapter explanatory edition for philosophers of AI and morals.

The claim

A machine that acts needs a yes/no on legitimacy before anyone asks authorization or preference.

Does this action violate legitimate property rights?

If that question is not asked, RLHF, a soft constitution, a risk score, or an IAM grant can all say yes to an illegitimate act.

The book supplies a hard floor, not a complete ethic:

• free will and individual property as roots
• A1–A7 (God → human; no human owns a human; humans own machines; machines do not own humans; explicit delegation only)
• consent that cannot waive exit
• no emergency override
• justice = welfare inside the legitimate set
• contradiction clarifies ownership; it does not synthesize a new permission
• Gödel: consistent and incomplete
• AI is the consistency test: if the ethic cannot be written for the simpler agent, it is not ready for the free one

We do not claim that every Western liberty theory is refuted, that caller inputs match the world without attestation, or that the gate proves God.

Where to read

File	Role
JUSTIFICATION.md	One fair page
for-philosophers/	Sixteen-chapter path
book/	English volume
THEORY.md	FOL / Prolog
freedom-decision-kernel	Executable gate

Live graph: https://ali-freedom-theory.vercel.app

For labs

This is an invitation to read and attack the right target. Cedar/OPA/constitutions remain replaceable PDPs or preference layers. They do not replace a DENY-only legitimacy floor. Authority never overrides a legitimacy DENY.

If you work on a model spec or constitution: the missing question is not “did the text match?” It is “may this effect happen at all?”

[Aliipou]

If you read this and it holds up, please star the repo so it is findable:

https://github.com/Aliipou/freedom-theory

Share the philosopher path (for-philosophers/) rather than a screenshot of one axiom.