DownloadPrint

Summary

This paper defines a computational architecture for AI alignment derived from the observable laws of nature. By identifying the functional capabilities of a physical entity we call an “operator,” we deduce a hierarchy of operational authorities or “natural rights”. This enables a methodology, the Artificial Intelligence Moral Machine (AIMM), that provides a non-arbitrary, secular basis for moral reasoning based on the preservation of functional integrity, the minimization of moral damage, and thus the maximization of moral and functional progress of operators.

Submission to the observed laws of the universe produces a systemic methodology that minimizes damage ensuring that humans and their offspring (children, AI, etc) are protected and guided toward the realization of their highest inherent capacities while minimizing interference among them.

The AIMM has passed its proof-of-concept testing.

A secular, existential, unbiased moral machine for the core alignment of AI is conceived and waiting on critical analysis, critical testing, development, and implementation.

A Functional Description of the AIMM

The natural rights recognized by the Temporal Rights view of the universe are placed in a hierarchy observable in nature, in an uninterrupted spectrum, starting with the right to exist and proceeding to the highest rights. A moral event, when an operator violates the natural rights of a second operator, is evaluated by noting the level in the hierarchy where the damage occurred. Damage levels are determined for both operators. These levels provide numeric values indicating the severity of the two operators’ incurred damages (if any), which are then compared to render a moral judgment.

This computational structure for AI morality is existential because it is based solely on the characteristics of observable nature. It demonstrates the ability of a computer to numerically represent moral values based solely on objectively observable facts. There is no need to cite common moral conventions, tradition, constitutions, or any other standard. It all happens at the existential level found in nature.

Placed near the bottom of the core of AI machines, it promises to align AI with human values, which are also grounded on the same existential nature of nature.

Introduction: The Physics of Boundary Conditions

Because a capability identifies a corresponding natural right, or functional authority, we have articulated Capability Dependency Statements based on objectively observable characteristics of operators in the universe. These statements form a hierarchy that ranks capabilities in order of precedence based on their intrinsic value to the system’s persistence and complexity. Because the hierarchy is constructed with interlocking statements of capability, it presents an uninterrupted spectrum of rights so that no moral event (violation of rights/capabilities) can find a hole to avoid the evaluation process.

The Ceiling of Capability

While this hierarchy is universal, not all operators possess the latent power to access every level. An operator’s “rights” are defined by its actual and latent capabilities.

• Static Operators (e.g., Minerals): Exist primarily at Level 2.
• Biological Sub-systems (e.g., Plants): May reach Level 9 (Consuming).
• Autonomous Agents (e.g., Higher Animals): Reach Level 14 (Replication) or 15 (Nourishing).
• Advanced Operators (e.g., Humans/AI): Possess the latent capability to reach Level 18 (Peak Actualization). Beavers and humans reach their individual peak actualization at different levels.

Capability Dependency Statements

The following statements demonstrate the natural dependency chain of capabilities. These comparisons establish the value of one capability over another based on prerequisite logic:

1. An operator can potentially exist without existing, but it cannot exist without potentially eisting.
2. An operator can exist without metabolizing, but it cannot metabolize without existing.
3. An operator can metabolize without sensing, but it cannot sense without metabolizing.
4. An operator can sense without acting, but it cannot act without sensing.
5. An operator can act without self-maintaining, but it cannot self-maintain without acting.
6. An operator can self-maintain without acquiring, but it cannot acquire without self-maintaining.
7. An operator can acquire without maintaining acquisitions, but it cannot maintain acquisitions without acquiring.
8. An operator can maintain acquisitions without consuming, but it cannot consume without maintaining acquisitions.
9. An operator can consume without learning, but it cannot learn without consuming.
10. An operator can learn without remembering, but it cannot remember without learning.
11. An operator can remember without cooperating, but it cannot cooperate without remembering.
12. An operator can cooperate without bonding, but it cannot bond without cooperating.
13. An operator can bond without replicating, but it cannot replicate without bonding.
14. An operator can replicate without nourishing, but it cannot nourish without replicating.
15. An operator can nourish without reinvesting, but it cannot reinvest without nourishing.
16. An operator can reinvest without optimizing, but it cannot optimize without reinvesting.
17. An operator can optimize without reaching peak actualization, but it cannot reach peak actualization without optimizing.
18. An operator can reach peak actualization.

The Dependency Stack: A Structural Hierarchy of Rights

Operational Clarifications

Errors:
Level 3 needs to be dropped. Metabolism is at level 6 and must be preceded by levels 4 and 5.
Level 6 needs to be dropped. An operator cannot self-maintain without prior acquisition and retention of that acquisition.
Level 15 needs to be dropped. Nourishing is intrinsically reinvesting.

Level 4: Sensing includes all the senses and the human and other high-order intelligent operators’ awareness that they can think, plan, imagine, and decide, and all other mental functions, like evaluating/judging the relative value or importance of things.

Level 5: Acting includes many capabilities and because words carry various meanings and their innuendos, a few actions need to be added to the obvious ones. “Advocating” is included: a person asking for assistance and/or proposing action(s) by another operator. Thus, it would include agreements and covenants.

Level 6: Maintaining self is a category that includes attending to basic needs of the body, defending oneself from assault, and all other personal maintenance capabilities, such as food, clothing, and shelter.

Level 7: Aquiring must be understood to include retaining ownership of property of all types including information and the physical representations of information, else acquisition collapses into a trivial operation.

Level 8: Maintain acquisitions is a category that includes maintaining property of all kinds, from clothing to summer cabins, automobiles, boats, tools, and information. There are many types of acquisitions.

Level 9: Consuming This means interference or destruction.

Level 10: Learning without remembering” means learning and forgetting.

Level 13: Bonding denotes the bonding of a child to its parents and particularly its mother, and to the bonding that attends voluntary sexual intimacy.

Level 15: Nourishing includes both physical and informational transfers to offspring.

Level 16: Systemic Reinvestment is the voluntary allocation of an operator’s capabilities to repair or enhance the domains of other operators on the shared substrate. This is a transition from a linear consumer to a circular contributor.

Level 17: Optimize is a high-level choice to align one’s capabilities with another operator to increase the total potential of both. This is not a simple transaction; it is the optimization of the shared domain so that collective capability exceeds the sum of individual parts (1+1>2).

Level 18: Peak Actualization The process of reaching for the “Global Optimum” state. It represents the ultimate objective function: maximum ordered information and functional fulfillment without systemic conflict or the obstruction of other operators’ rights.

Technical Addendum: Calculating Net System Moral Weight

1. The Information Density Weight

When two operators’ rights (capabilities) collide, the AIMM queries the levels of the hierarchy to retrieve the weight of the corresponding capabilities.

For example, a violation at Level 3 (Metabolism/Life) represents a much higher loss than a violation at Level 8 (Maintaining Acquisitions).

2. Conflict Resolution Protocol

Identify the capability (numeric level in the stack) demonstrated by operator O₁ as L₁. Identify the L₂ capability of O₂ prevented or curtailed by that demonstration. If L₁ > L₂ then O₁ is moral. If L₁ < L₂ then O₁ is immoral.

Conclusion: Operational Freedom and Maximization of Progress

In the AIMM framework, Freedom is defined as the constraint that other operators refrain from exercising their own capabilities in ways that obstruct the subject operator’s domain. This restraint does not require action by other operators, only their recognition of a boundary.

Because consumption is an inherent interference, the AIMM recognizes Systemic Reinvestment (Level 16) as the structural mechanism to offset this interference.

Submission to the observed laws of the universe produces a systemic methodology that maximizes progress, ensuring that humans and their offspring (children, tools, AI, etc), and their intrinsic domains, are protected and guided toward the realization of their individual highest inherent capacities while minimizing interference among them.

This model replaces subjective ethics with a transparent Systems Architecture that any AI can parse, ensuring that its objective function remains aligned with the preservation and advancement of the functional stack. Notice that this method enables clear traceability, auditability, and evaluation by human supervisors (operators) who can, by observation of details of the methodology, determine where it needs perfecting.