The AtomSpace

The AtomSpace is the heart of Hyperon. It is a specialized, distributed database designed to store complex, multi-layered knowledge in a way that AI algorithms can easily traverse and reason about.

Atoms: The Building Blocks

Everything in the AtomSpace is an Atom. There are two primary types:

• Nodes: Represent individual concepts or data points (e.g., "Human", "Socrates").
• Links: Represent relationships between Atoms. Since it's a hypergraph, a link can connect any number of atoms (not just two).

Examples of Atoms

• ConceptNode "intelligence"
• InheritanceLink (ConceptNode "Socrates") (ConceptNode "Human")
• EvaluationLink (PredicateNode "is-mortal") (ListLink (ConceptNode "Socrates"))

Key Characteristics

1. Unified Knowledge Representation

Symbols, data, rules, and even executable code are all stored as Atoms. This means the system can treat "code" like "data" (homoiconicity), which is essential for self-modification.

2. Meta-Information (Truth Values)

Each Atom can have associated metadata, most notably Truth Values:

• Strength: How true the relationship is (0 to 1).
• Confidence: How much evidence the system has for this truth value.

3. Attention Values

Used for resource management. Atoms that are frequently used gain "STI" (Short-Term Importance), while those rarely used are eventually "forgotten" or moved to long-term storage.

The Distributed AtomSpace (DAS)

In a production AGI system, the knowledge base will be too large for a single machine. The DAS provides:

• Scalability: Sharding knowledge across a cluster.
• Query Optimization: Efficiently finding patterns across millions of Atoms.
• Persistence: Ensuring knowledge survives system restarts.

Interaction with Search and Reasoners

Reasoners like PLN use the AtomSpace to find facts, combine them using rules, and write back new derived facts as new Atoms.

---

Next: MeTTa Basics