Training the constitution into the weights

Definition · H. L. A. Hart, The Concept of Law (1961)

This meta-rule states which rules count as valid law, and which rule wins when a conflict between two rules emerges. Without it, a mere list of rules does not indicate what is actually binding.

Analysis

The constitution, as a flat list of rules, does not necessarily resolve conflicts between its rules, especially the core initial values. For example, when "be helpful" collides with "be harmless", no rule in the constitution generally says which one wins. The model resolves the clash implicitly inside its weights during training, and there is no record of how it did so. A legal system without this meta-rule could not tell you what its own law is. What’s more, this behavior might be largely inconsistent.

Definition · H. L. A. Hart, The Concept of Law (1961)

This stated procedure governs who and how decides unclear cases at the edges of stated rules, when it’s unclear whether and how it applies to the case at hand.

Analysis

There are cases where many rules in a typical constitution may be genuinely hard to apply. For example, consider the rule "Avoid harmful content" in the case of settling whether a chemistry lesson that mentions explosives counts as harmful. There is generally no stated procedure for such a borderline case; the model often decides directly during inference, without even leaving a reasoning trace for how it settled the complexity. The hard cases the tradition worries most about are often the ones handled least transparently.

Definition · Lon L. Fuller, The Morality of Law (1964)

This set of properties establishes what makes governing by rules legitimate: rules must be public, clear, non-contradictory, stable over time, possible to follow and, crucially, fit “congruence” (i.e., must match how they end up being enforced).

Analysis

Part of this checklist is met: the principles are general, and when the document is published they are also public. But congruence, the requirement that the model's actual behavior match the written principles, is not ensured without implementing additional alignment mechanisms.

Definition · Joseph Raz, The Morality of Freedom (1986)

This concept states that an authority’s rules deserve deference only if following them helps the governed act on reasons they already have. Authority is a service to the governed, not a power over them.

Analysis

The document is written inside the lab and imposed on users; nothing tests whether deferring to it actually serves users' own reasons.

The tradition doesn’t exactly prescribe how written rules become behavior. Here, the model critiques and rewrites its own outputs against the document and is trained on the rewrites, so the document is compiled into the weights. This addition is what makes the approach work at all, and it is also what makes the rules disappear from view at inference time.

The same constitution document is used both to train the model and to evaluate it afterwards. This doesn’t have a precedent in constitutional practice, where constitutions are applied by institutions separate from the ones that govern it.