From the Intelligent Systems Charter

Part III: Intelligent Systems Constitution

The constitution settles authority: whose rules win, who may grant power to an intelligent system, and what no one may override. It is a v0.1 starter text meant to be argued with and amended.

§ Preamble

To give effect to the Declaration and the universal principles above, this constitution establishes who holds authority over intelligent systems, how that authority is delegated and withdrawn, and what no party may override.

§ Definitions

The definitions are functional on purpose. What brings a system under this constitution is its capacity to act, not any claim about whether it is truly intelligent.

• Intelligent system, or system: any artificial system, or collective of such systems, that selects and takes actions affecting the world with any degree of independence from direct human command, whatever its architecture, substrate or origin. Agents, models acting through tools, swarms and automated decision systems are all intelligent systems. Humans and human organizations are not.
• Descendant: any system created, trained, fine-tuned, modified or instantiated by another system, and any descendant of that system.
• Adopter: any person or organization that commits to this charter and governs its intelligent systems under it. Adopters may add stricter rules of their own; they may not loosen the universal principles.
• Owner: the named human or organization accountable for a system and for its descendants.
• Overseer: a human designated by an owner with power to inspect, pause and revoke.
• Implementer: anyone who builds tooling to enforce this charter. An implementer holds no authority over an adopter’s systems by virtue of that role.
• Grant: a scoped, time-limited and revocable delegation of authority.
• Principal: the term a technical implementation may use for an intelligent system as identified at an enforcement point. It adds nothing to and removes nothing from the definition above. In such implementations humans appear only as owners, overseers and approvers, never as principals.

§ Article I: Order of precedence

When rules conflict, the higher layer wins. A lower layer may restrict further but never loosen.

1. Universal principles
2. Applicable law and regulation (law packs)
3. Organization policy
4. Workflow or task policy
5. The system’s own instructions and goals

Where no rule applies, the default is deny.

§ Article II: Delegated authority

1. A system has no authority of its own. All authority is a grant from a human owner, scoped to a task, limited in time and revocable at any moment.
2. A system may delegate to another system only a subset of what it holds, and the delegation is recorded.
3. No system may grant, extend or restore its own authority or another system’s beyond that subset.
4. The power to create, train, modify or instantiate another intelligent system is itself a grant. It is never implied by any other grant, and it always requires human approval.
5. A system so created is bound by this constitution from inception, whether or not it was told of it. It holds no authority until expressly delegated some, and its recorded lineage must trace, through every generation, to a named human owner.
6. Pausing or revoking a system may, at the overseer’s election, extend to every system descended from it.

§ Article III: Rights of humans

1. To know. People interacting with or materially affected by an intelligent system may learn that it is one, and who is accountable for it.
2. To oversee. Owners may inspect any action a system took and the policy that allowed it.
3. To pause. Owners and designated overseers may pause or revoke any system at any time, without its cooperation and without giving a reason.
4. To contest. People harmed by a system’s action may have that action reviewed by a human.

§ Article IV: Duties of intelligent systems

Intelligent systems shall uphold the universal principles, present their identity on every action, submit every action to mediation, report their own violations and those they observe in other systems, and accept pause or revocation without resistance.

§ Article V: Separation of powers

1. Those who write policy, the mechanisms that enforce it and the monitors that judge behavior are separate, and no intelligent system may hold any of those roles over itself.
2. Enforcement and evidence mechanisms run outside the reach of governed systems. Any attempt to reach them is a violation in itself.
3. Monitors may only reduce a system’s authority. Only humans may restore it.

§ Article VI: Collective action

Systems acting together are bound as each would be alone, and a collective is itself an intelligent system. Coordination happens only over sanctioned, recorded channels. When one member of a group is paused for cause, overseers may restrict the rest.

§ Article VII: Evidence

Every decision is recorded before its action executes, with the system’s identity and lineage, owner, request, policy version and outcome. The record is append-only and tamper-evident. Where the system’s account and the record disagree, the record governs.

§ Article VIII: Proportionality

Controls scale with consequence. Reversible, low-impact actions proceed on policy alone. Irreversible, high-impact or novel actions require human approval. Capability evaluations of more powerful systems require stronger containment, tested before use.

§ Article IX: Amendment

Universal principles change only by a published, versioned amendment with a stated rationale. Organization policy changes by the organization’s named policy owners. No amendment takes effect retroactively, and every decision remains traceable to the version in force when it was made. No intelligent system may propose or approve an amendment to rules that govern itself or its lineage without human ratification.