Knowledge-Graph Monarchism
From the AI Research Compass wiki
Knowledge-Graph Monarchism: Holds that knowledge should live in curated, structured knowledge graphs of entities and relations, with the graph as the sovereign source of truth.
Beliefs
Knowledge-Graph Monarchism believes truth comes in triples: subject, relation, object. Paris, capital of, France. A well-curated graph of entities and typed relations is precise, editable, queryable and auditable, everything a neural network’s weights are not. When a fact changes, you edit one edge instead of retraining a model and hoping. Language models may be eloquent courtiers, but the crown belongs to the graph, and every answer should be traceable to a node someone can inspect.
It sits at the horizontal centre, since graphs are built from named entities and linguistic relations yet describe the world itself, and near the Harness end: a structured store that the rest of a system queries, rather than a monolithic model. Its paradigm level is Neutral. Knowledge graphs powered semantic web initiatives, enterprise search and question answering long before large language models, and the monarchists see no reason to abdicate.
Arguments
- Structured graphs support exact queries, multi-hop joins and constraint checking that free-text retrieval and neural recall handle unreliably.
- Facts can be updated, versioned and attributed individually, which matters for compliance and correctness.
- Knowledge graph embeddings and graph-grounded generation can reduce hallucination by anchoring answers to verified entities.
Critiques
- Building and maintaining large graphs is labor-intensive, and coverage always lags the messy long tail of real knowledge.
- Much knowledge, such as procedures, nuance and context, does not fit cleanly into triples.
- Retrieval and long-context camps argue that indexing raw documents gets most of the benefit without the schema-design burden.
Representative works
- Translating Embeddings for Modeling Multi-relational Data2013
- Wikidata: A Free Collaborative Knowledgebase2014
Relationships
Friends
Frenemies
How to get it
Land on this cell in the full or pro test: x ≈ −0.33, y ≈ −0.93, paradigm ≈ −0.03.