The knowledge graph

Why a graph?

What is a knowledge graph?

Knowledge as a network instead of a list: things are nodes, their relationships are edges. A graph answers not only “what is this?” but also “what is it connected to – and how?”

1. Nodes and edges

   Every article is a node. Its edges say what it is about and where it belongs – here the gecko article, exactly as the graph above stores it.

2. Four dimensions as coordinates

   Every article is placed in all four at once – that is what makes a shrimp, a gecko and a robot comparable.

   • When? Time Cambrian · Modern era · Age of AI
   • Where? Space Coral reefs · Microcosm · Outer space
   • Which laws? Rules Mechanics · Fluid dynamics · Optics
   • Which disciplines? Adjacent sciences Biology · Computer science · Bionics

3. Lenses as bridges

   A lens is one way of looking at the same phenomenon. In the graph, each lens is an edge from the article to its discipline.

   • BiologyBiology
   • PhysicsRules
   • MathematicsMathematics
   • ChemistryChemistry
   • Computer scienceComputer science
   • Physical AIPhysical AI

4. Shared concepts connect

   Articles that seem unrelated meet at shared nodes. With every new article the web gets denser, and the next thing worth reading gets easier to find. Each article also describes its subjects as structured data (JSON-LD), so that search engines and AI assistants understand it.

5. Why AI needs it, too

   Language models are strong at language but weak at facts they cannot look up: they may invent plausible answers, and they have to read long texts to find a single relationship. A knowledge graph hands them explicit, sourced connections – that is why WildBionics publishes its graph as open data, adds JSON-LD to every page and offers an llms.txt.

   Without a graph

   Read every article – 15 so far

   ≈ 77,000 tokens

   With the graph

   Follow two edges

   article:gecko-adhesion → term:mechanics
article:gecko-adhesion → term:physical-ai

   ≈ 45 tokens

   Answer Gecko adhesion

   The article side grows with every new article; the answer from the graph stays two edges. Estimated from this site’s current data at about 4 characters per token – a rule of thumb for English text; the exact count depends on the model and the language.

• More accurate answers

  In a benchmark of business questions, GPT-4 answered 16% correctly on a raw SQL database – and 54% when the questions were posed over a knowledge graph of the same data. [1]

• Better overviews

  For questions about a whole collection, such as “what are the main themes?”, a graph index gave substantially more comprehensive and diverse answers than conventional retrieval (RAG). [2]

• Traceable reasoning, lower cost

  A model that explores a knowledge graph step by step can show its reasoning path; in some tasks a small model with a graph outperformed GPT-4. [3]

• Fewer hallucinations

  Grounding answers in a graph of facts is a promising way to reduce plausible but false answers – still an active field of research. [4] [5]

• No mix-ups

  WildBionics links its topics to Wikidata: “gecko” then means the lizard (Q1008888), not the browser engine of the same name.

What you can do

• Explore

  Hover over or tap a node to see its neighbours, and filter the dimensions.

  To the graph

• Find gaps

  Grey concepts have no article yet – an invitation to write one.

  Write the missing article

• Reuse the data

  The graph is open data under CC BY-SA 4.0 – build your own analyses or visualisations.

  Download graph.json