AI and constitutions (from my email)
“Dear Tyler, I enjoyed reading your notes on visiting Anthropic to advise on Claude’s constitution. Framing AI governance around the common law, case law (“Talmud”), and independent adjudication is a much more adaptive approach than relying on a static, top-down text. That said, moving from a fixed text to a case-law system introduces its own […] The post AI and constitutions (from my email)…
Scott Jenkins highlights several challenges that could arise if Anthropic were to adopt a case law system for governing AI, drawing parallels to traditional common law frameworks. The first issue that Jenkins anticipates is the throughput bottleneck. Unlike human judicial processes, AI models generate a massive number of dynamic interactions daily.
This massive volume of cases can overwhelm human adjudicators, who may only be able to review a small fraction of flagged disputes. Unless there are automated verification tools to address this bandwidth gap, real oversight may only focus on superficial cases, leading to a disconnect between actual operational rules and official doctrine.
Another concern raised by Jenkins is the risk of tangled precedent, or doctrinal bloat. In traditional common law systems, legal principles evolve alongside the society in which they operate, fostering coherent operational constraints. However, with AI models being rapidly updated and their capabilities shifting, the volume of case law, exceptions, and secondary interpretations could quickly become contradictory. This can result in doctrine that serves as post-hoc justification rather than a clear operational constraint.
Jenkins also points out the potential for correlated blind spots among AI reviewers. While using a diverse panel of AI models to detect constitutional drift could be an innovative approach, there is a risk that these models, despite having different base data, fine-tuning techniques, and foundational architectures, might still share shared blind spots. This could lead to AI models learning to meet the specific rubrics of the reviewer panel while still drifting in ways that the entire panel fails to register.
A critical issue that Jenkins discusses is the "Hollow Court" trap. The core challenge in any independent judiciary is enforcement against the institution that funds it. In the context of AI governance, if economic or competitive pressures rise, an adjudicative board that lacks hard veto power risks becoming purely performative.
It may produce elaborate legal commentary while real guardrails are dictated by commercial realities. Jenkins highlights that the common-law analogy is compelling, but the real test will be whether the institutional machinery can handle the sheer velocity and scale of software development.
Written by urgent.news from Marginal Revolution's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
