Under AI Conditions 1.2.2-4

Why AI Is a Direct Load Case for Encoded Objective Law

The load case named in (1.1.1) — AI-mediated coordination — describes systems that produce consequential decisions about persons at speeds that outpace human oversight. Encoded objective law is directly exposed to this pressure because AI systems are increasingly the operational layer through which institutional protocols are executed. When an algorithm denies a loan, flags a traveler, moderates speech, or allocates medical resources, it is performing protocol enforcement — applying rules to persons, with consequences. The question is whether the protocol's moral constraints survive the transition from human execution to automated execution.

This sidebar maps the AI-specific failure modes that encoded objective law must address.

Four AI-Era Failure Modes

Enforcement Without Comprehension

AI systems can execute protocol rules at scale without understanding the moral purpose those rules serve. A content moderation algorithm applies the letter of a policy without grasping the principle — suppressing both hate speech and the documentation of hate speech under the same rule, because both contain the same flagged terms. A lending algorithm enforces risk thresholds without understanding that a threshold calibrated to one population produces discriminatory outcomes in another. The enforcement is technically faithful. The moral constraint the enforcement was supposed to carry has been lost in translation.

The failure mode is precise: the protocol's hard constraints are executed correctly, but the soft constraints — proportionality, context-sensitivity, judgment — have been automated as though they were hard constraints. The enforce-in-code / enforce-in-culture distinction collapses. What should have been a judgment call has been encoded as a binary rule, and the binary rule is applied at scale with no mechanism for the system to recognize when its execution violates the constraint's purpose.

Design response: Protocols that incorporate AI execution must explicitly tag which constraints are safe for automated enforcement (genuinely binary, universal, violation-intolerant) and which require human judgment at the point of application. The tagging is a structural requirement, not a suggestion. Constraints that require contextual assessment must include escalation paths to human deliberation — and the escalation must be triggered by the system's own design, not by the affected person's ability to navigate an appeal process.

Opacity as Default

AI systems frequently operate as black boxes — producing consequential outputs from inputs that no human reviewer can trace or audit in real time. The opacity is structural: the model's decision logic is distributed across millions of parameters that do not correspond to human-readable rules. When an AI system denies, approves, flags, or recommends, the institution may not be able to explain why in terms that satisfy either epistemic integrity's duty of clarity (1.2.1) or encoded objective law's requirement that moral constraints be auditable.

The failure mode is not that AI systems are deliberately opaque. It is that their architecture produces opacity as a side effect of scale and complexity. A protocol that relies on AI enforcement inherits this opacity — and the moral constraints the protocol encodes become unauditable. The institution claims to enforce specific rules. The enforcement mechanism cannot be inspected to verify whether those rules are actually what is being enforced.

Design response: Encoded objective law under AI conditions requires that the enforcement mechanism be inspectable — that the institution can demonstrate, at a structural level, that the AI system's outputs are consistent with the moral constraints the protocol claims to enforce. This may require interpretability tools, external audits, or structured testing against known cases. The standard is not that the AI system must be fully transparent in its internal operations — that may be structurally impossible — but that the correspondence between stated constraints and actual enforcement must be verifiable.

Consent Erosion at Scale

AI systems interact with millions of persons simultaneously, often through interfaces designed to optimize compliance rather than comprehension. Terms are long. Alternatives are absent. Refusal carries disproportionate cost. The consent mechanism — "agree to proceed" — meets the formal definition of consent while violating every structural condition that makes consent meaningful. Under AI conditions, this erosion scales: the same consent-extraction mechanism is deployed across populations, producing mass compliance that no individual meaningfully chose.

The failure mode for encoded objective law is that consent theater — named in the article's "What RO Leaves Behind" section — becomes the default rather than the exception. When AI systems mediate the consent process, the structural conditions for genuine consent (comprehension, genuine alternatives, capacity to refuse without disproportionate cost) are systematically degraded by the same optimization pressures that make the AI system efficient.

Design response: Protocols that incorporate AI-mediated consent must be designed so that the consent interface satisfies the structural conditions for consent, not merely the formal requirements. Comprehension must be verified, not assumed. Alternatives must be genuine, not decorative. The cost of refusal must be proportionate, not punitive. These are design requirements, not cultural aspirations — and they apply to the AI system's interface, not just to the policy it executes.

Protocol Modification Without Visibility

AI systems that learn and adapt can modify their effective enforcement rules without any explicit change to the protocol. The protocol's written rules remain unchanged. The AI system's interpretation of those rules drifts as the model updates, retunes, or incorporates new training data. The result is a protocol that says one thing and does another — not through institutional bad faith but through the structural dynamics of model evolution.

This is a new category of protocol capture. Traditional capture requires a human actor to modify the protocol's rules. AI-enabled capture requires only that the model's behavior drift from the protocol's stated constraints — and that the drift go undetected because the protocol's audit mechanisms were designed for static rules, not adaptive systems.

Design response: Protocols enforced by AI systems must include drift-detection mechanisms — periodic audits that compare the AI system's actual enforcement behavior against the protocol's stated constraints. The audits must be structural (not dependent on the AI system's self-reporting) and periodic (not triggered only by complaints). The standard is that the protocol's moral constraints must remain verifiable over time, not just at the point of deployment.

The Structural Requirement

AI does not invalidate encoded objective law. It sharpens the design requirements. The enforce-in-code / enforce-in-culture distinction becomes more consequential when "code" means an adaptive system rather than a static rule set. Moral harm becomes easier to produce at scale when enforcement is automated. The institutional duty of enforcement now includes the duty to ensure that AI-mediated enforcement actually carries the constraints it claims to carry.

The innovation was designed for a world in which moral constraints must survive operational implementation. AI systems are the most powerful implementation layer the world has produced. Encoded objective law must hold here — or it does not hold at all.