liveBook · Manning

False independence

As explained in AI Governance

False independence is the mistaken impression that agents provide independent judgments when they share assumptions or sources of error.

False independence occurs when agents appear to provide separate judgments but share blind spots, making their agreement weaker evidence than it seems. In agent quality checks, one agent may review another yet miss errors because both rely on the same assumptions or sources. section 6.7

Reviewers may share:

  • The same model, retrieval-augmented generation (RAG) index, prompt template, or business objective. section 6.7

Shared answer sheet

Three students grading one another after copying the same answer sheet may agree without independently checking the answers. section 6.7

How AI Governance covers it

How deep it goes
Defines the failure, identifies shared dependencies, and gives an analogy, but provides no mitigation procedure.
Examples
Uses agents reviewing one another and students grading from the same copied answer sheet.
What you can do after
Evaluate agent agreement by checking whether reviewers share a model, RAG index, prompt template, or business objective.
How it connects
Links agent quality checks to shared models, retrieval-augmented generation indexes, prompt templates, and business objectives.

Related concepts

This explainer was generated by Manning's knowledge graph from the text of AI Governance. The section links point to the parts of the book it draws on.