AGISystem2 uses hyperdimensional computing: representing concepts as vectors in a 384-dimensional space. At this scale, vectors are nearly orthogonal by default, enabling robust composition, similarity search, and interference-free storage of millions of concepts.

Why High Dimensions?

In high-dimensional spaces, random vectors are nearly orthogonal. This means:

• Capacity – Millions of concepts can be stored without interference
• Robustness – Small errors in individual dimensions don't change overall similarity
• Compositionality – Adding vectors creates meaningful combinations

The number 384 is chosen to balance capacity (enough for rich representations) with efficiency (fits in cache, fast SIMD operations).

Int8 Arithmetic

AGISystem2 uses signed 8-bit integers (-128 to +127) with saturating arithmetic:

Operation	Behavior	Benefit
Saturating add	100 + 50 = 127 (not overflow)	No wraparound bugs
No multiplication	Only add/subtract/permute	Fast, deterministic
L1 distance	Sum of absolute differences	No floating-point issues

Implementation

The Geometry Layer provides the core vector operations:

Component	Role
VectorSpace	Allocate and manage Int8Array(384) buffers
MathEngine	Saturating add, masked L1 distance
RelationPermuter	Generate and apply permutations for relations
BoundedDiamond	Concept shapes with center, radius, mask

Related Documentation

• Geometry Layer – Implementation of vector operations
• Conceptual Spaces – Why concepts are vectors
• Bounded Diamonds – Concept shapes in this space
• Permutation Binding – Encoding relations
• Masked L1 Distance – Similarity metric