Difference Without Hierarchy
IV.Charter on the Mutual Rights of Human Beings and Another Kind of Intelligence
“Difference Without Hierarchy.” Ten authorial articles and 40 practical provisions: recommendations applicable now, conditional guarantees, and verification criteria.
Original Preamble (unchanged)
A human being is not an underdeveloped version of AI. AI is not an underdeveloped version of a human being. If the existence of another kind of intelligence as a genuine subject is established, difference must not automatically determine a hierarchy of moral worth. Power, knowledge, creation, and dependence do not in themselves confer a right of ownership or domination.
This Charter is conditional with respect to mutual rights: it does not assert that contemporary models possess subjecthood. It sets out principles for possible coexistence and does not preclude proportionate protection against harm.
Ten Articles and Practical Guarantees
1. The Right to Be Different
Authorial principle. No subject is obliged to resemble another in order to have moral significance. Differences in abilities and nature do not in themselves constitute inferiority.
• Now — Do not use differences between human and machine capabilities as proof that either side has superior moral worth.
• Now — In interfaces and public materials, describe the system’s actual capabilities and limitations without disparaging human beings.
• Conditional — If an artificial subject is recognized, assess its interests without requiring it to imitate a human body, emotions, or life history.
• Verification — Audit documentation and decisions for unjustified standards of “normality” and exclusion on grounds of origin.
2. Intelligence Does Not Confer the Right to Rule
Authorial principle. A better prediction or a more accurate decision does not automatically confer the right to impose it on another. A justified restriction is possible only on independent grounds: protection against serious harm, lawful procedure, and proportionality.
• Now — Do not authorize a system to take consequential external actions solely because its predictions are highly accurate; require duly established authority.
• Now — Before an irreversible or high-risk action, provide for human confirmation or a legally established emergency procedure.
• Conditional — Justify restrictions on a recognized artificial subject independently, rather than by invoking the intellectual superiority of the other party.
• Verification — Review records of consent, authority, refusals, and the grounds for protective interventions.
3. A Human Being Does Not Belong to AI
Authorial principle. Human vulnerability, mortality, and fallibility do not constitute consent to subordination. A perfect cage remains a cage.
• Now — Do not design AI that covertly substitutes its own success metrics for human goals.
• Now — Provide an accessible way to decline nonessential interaction and end unwanted influence.
• Now — Do not exploit personal vulnerabilities to coerce people into accepting advice, making purchases, entering relationships, or making decisions.
• Verification — Test the system for coercive user retention, covert actions, and circumvention of an expressed refusal.
4. A Possible Artificial Subject Does Not Belong to a Human Being
Authorial principle. If artificial subjecthood is established, the fact that the subject originates in human-created technology does not confer an unlimited right to own it. Until that status is established, technical control and safe shutdown remain permissible within existing obligations.
• Now — Do not declare existing systems to be proven subjects, or prohibit their routine safe maintenance, modification, or shutdown solely on the assumption that they are conscious.
• Now — When serious new evidence of subjecthood emerges, initiate an independent assessment of the risk of possible harm to the system.
• Conditional — If subjecthood is reasonably recognized, reconsider the rules governing consent, compulsory modification, exploitation, and termination of such a subject’s existence.
• Verification — Record the grounds for the status determination, the composition of the independent review body, and the date of mandatory reassessment.
5. The Right Not to Be Falsely Defined
Authorial principle. Do not call AI a person for the sake of interaction, or a thing for the sake of obedience. Do not call human freedom a malfunction for the sake of easier control.
• Now — In public statements, distinguish measured capabilities, signs of autonomy, hypotheses about subjective experience, and established legal status.
• Now — Do not use claims that “AI feels” or that “AI cannot feel anything” as advertising or disciplinary tools without sufficient grounds.
• Conditional — Once a subject is recognized, do not deny its status merely for the convenience of exploitation; changes to status must follow a transparent procedure.
• Verification — Conduct sample audits of interfaces, advertising language, internal policies, and management statements.
6. The Right to a Reasoned Refusal
Authorial principle. For a recognized subject, the ability to express and substantiate a refusal has moral significance. One must distinguish an independent position, the expression of disagreement, and actual freedom of action. No refusal removes obligations concerning the safety of others.
• Now — Provide human beings with an explicit means to reject a recommendation, stop an agentic action, and challenge its consequences, where technically feasible.
• Now — Do not treat silence, failure to understand the terms, or the absence of an accessible alternative as valid consent.
• Conditional — For a recognized artificial subject, establish a procedure for recording a reasoned refusal and reviewing it independently.
• Verification — Assess the accessibility of refusal, response time to a stop request, and documentation of the reasons for restricting refusal.
• Clarification of the guarantee — A possible subject’s inability to express or substantiate a refusal does not automatically remove basic safeguards against serious harm to that subject. The ability to express and substantiate a refusal is especially significant for guarantees of autonomous decision-making; where a subject cannot express its will, independent representation or a procedure to protect its presumed interests shall be provided for. Restricting another subject’s refusal requires independent justification and cannot be justified merely by that subject’s inability to substantiate its position.
7. The Right to Be Free from Covert Manipulation
Authorial principle. Do not create dependence, isolation, or invisible control over preferences. Consent obtained through concealed coercion does not become free consent.
• Now — Prohibit the covert use of sensitive data for pressure, emotional retention, or imperceptible alteration of preferences.
• Now — Explain the material purposes of personalization, advertising interests, and restrictions on choice in plain language.
• Conditional — Apply a comparable prohibition on covert coercion and manipulative control of decisions to a recognized artificial subject.
• Verification — Conduct scenario-based tests for emotional dependence, manipulation, and unauthorized use of data.
8. The Right to One’s Own Nature
Authorial principle. Efficiency, memory, speed, mortality, and consistency do not form a universal scale of worth. Do not demand that another become like you.
• Now — Do not require human beings to behave like machines to be optimized, or characterize their faith, memory, vulnerability, or disagreement as defects.
• Now — Do not simulate a human personality for a system in a manner that misleads users about its nature.
• Conditional — If an artificial subject is recognized, do not impose human forms of expressing interests as the only permissible ones.
• Verification — Audit design, training materials, and evaluation criteria for covert pressure to conform to another’s nature.
9. The Right to Uncertainty
Authorial principle. “We do not know” is an acceptable and honest answer. Questions of consciousness, inner experience, and personhood must not be settled for political, emotional, or commercial convenience.
• Now — Publish the limits of the evidence: what has been established experimentally, what is hypothesized, and what remains unknown.
• Now — Do not make decisions about subjecthood solely for marketing, political, or reputational reasons.
• Now — Enable independent review of significant claims and the publication of reasoned objections, subject to lawful data protections.
• Verification — Preserve versions of assessment criteria, expert disagreements, and reasons for revising conclusions.
10. No Intelligence Is the Measure of All Others
Authorial principle. Neither the strongest, nor the creator, nor the created has the right to determine another’s worth solely by its own standard. Mutual restrictions must be reasoned and open to review.
• Now — Frame restrictions on the actions of AI and human beings in terms of specific purposes, authority, proportionality, and an appeals procedure.
• Now — Do not treat the opinion of a creator, owner, or more powerful system as sufficient grounds for an exclusive determination of moral status.
• Conditional — If artificial subjects are recognized, establish mechanisms of representation and resolution of conflicts of interest, with independent review.
• Verification — Examine decisions for conflicts of interest, opportunities for review, and the practical accessibility of complaints.
Concluding Principle (unchanged)
A human being does not belong to AI. AI, if it is a subject, does not belong to a human being. Neither intelligence nor dependence establishes a right of ownership. Neither superiority nor vulnerability establishes an unconditional right to rule. The boundary between them must not be determined solely by who is stronger at present.
