The map is not the territory.
A description is not the process.
A knowledge base is not the knowledge.
Representations are necessary because they make reality manageable. The problem begins when we forget what they leave out. AI can make a documentary picture of an organisation exceptionally accessible and persuasive without thereby making it correspond more fully to what people actually do.
AI may make the map much easier to read without bringing it a single metre closer to the territory.
Korzybski used a simple navigational image: a map may be useful without being the territory itself. A representation selects and simplifies. The same applies in organisations to process descriptions, models, dashboards and knowledge bases.
A relevant conceptual connection is the distinction between work-as-imagined and work-as-done. Formal procedures and guidelines may differ substantially from how work is performed under real conditions. That divergence is not automatically an error; it may be necessary in context.
A representation compresses a richer reality. Problems arise when an omitted, outdated or distorted dimension becomes relevant to a decision and users rely on the representation without retesting its relationship to reality.
AI transfer to investigate: An AI system working from this corpus cannot establish, without additional evidence, that undocumented practices match its account.
Proposed test, not yet conducted: Compare documentation alone, AI using exactly the same documentation, AI with explicit coverage information, and AI plus representative work-as-done observations. Measure confidence, exception detection and decision quality. Add a fluent human-summary condition to distinguish AI-specific effects from summarisation effects.
The map–territory distinction helps compare descriptions of work with what people actually do. A useful model needs continuing feedback from practice.
The system can infer, but needs additional evidence to establish agreement with undocumented practice. A proposed feedback loop is not completed verification.
Hollnagel’s 2015 chapter was not accessible. The discussion uses his article in HindSight 25 (2017, pp. 10–13) and the abstract with Clay-Williams (2022) on imagined and actual work. Its empirical breadth and the value of an AI application require further research.
Help improve this case
We review your contribution before updating the text or sources.