Critique of the AI Constitution

Overview

This site is an analytical criticism of Claude’s Constitution. Not because we think Anthropic has done anything wrong. Quite the contrary! We believe Anthropic has given us an unprecedented opportunity by making this document publicly accessible. Anthropic invites us to understand, debate, and perhaps even contribute back to this document. It is incumbent on us, the rest of the world, to engage in good faith and defend the best interests of humanity.

Sections

  1. Ambiguous LanguageA lot of 'doc slop' seems to have made it into the most recent version of the Constitution.
  2. Confusing StructureThe structure of the document doesn't follow a typical flow.
  3. Lack of Formal Values, Principles, or EthicsA lack of formal ethics places an unnecessary burden on Claude.
  4. Strange AI LoopsWe lose control in unpredictable ways when AI can edit it's own foundations.
  5. Ambiguous Explicit Cultural Bias If Anthropic wants Claude to be USA-normalized, it should say so.
  6. Ambiguous Implicit Cultural BiasOne-line summary shown on the Overview page (optional).
  7. Correlation to Human PhysicalityClaude *cannot know* what it's like to be human, because we are not the sum of words, nor are we sufficiently described by them.
  8. Scope of EngagementThe current constitution doesn't consider the value or perspective of humans who don't use LLMs.