What is an AI model made of?
At the hardware level: silicon transistors. At the mathematical level: billions of floating-point numbers. These answers are true but not useful. They’re like describing a book as “ink and paper” — accurate, but missing what actually matters.
Holosynthics answers the question at the level that matters: what are the semantic building blocks of AI cognition?
Digital matter
We define digital matter as the organized semantic content inside AI language models — the structured representations that allow these systems to process meaning, not just process tokens.
Physical matter is made of atoms, organized into elements, which combine into molecules. Digital matter is made of what we call elements — 56 distinct organizational types that appear universally across different AI models, organized into 8 fundamental types.
These elements aren’t physical things. They’re patterns — recurring organizational structures in the numerical weight space of the model. But they’re consistent enough across different models that we can study them the way chemists study chemical elements: by their properties, their behaviors, and their relationships.
The 8 fundamental types
The 56 elements organize into 8 fundamental types, each representing a different aspect of how AI models process meaning:
Boolean — the model’s truth-handling apparatus. These elements process yes/no, affirmation/denial, existence/absence.
Number — quantitative processing. Cardinals, ordinals, magnitudes, fractions, money, statistics.
Symbol — linguistic structure. The machinery that processes tokens, code, and formal syntax.
Identity — named entities. People, places, organizations, products, roles, and works.
Space — spatial relationships. Positions, directions, regions, boundaries, layout, and containment.
Time — temporal relationships. Dates, durations, sequences, eras, events, and frequency.
Relation — logical connections. Causation, contrast, conjunction, reference, part-whole, and analogy.
Entropy — uncertainty handling. Probability, ambiguity, variability, noise, approximation, and confidence.
Why the same structure appears everywhere
The consistency of these 56 elements across models from different organizations, with different architectures and different training data, is the central finding of Holosynthics research.
It suggests that these elements aren’t specific to any particular AI system. They’re the natural organizational structure that emerges when any large numerical system is trained to process language at scale.
Digital matter, like physical matter, has a structure that exists independent of the particular instance you’re looking at. Understanding that structure is what Holosynthics is for.
What this enables
The Holosynthics framework turns AI models from black boxes into structured systems that can be mapped, monitored, and corrected.
Once a system is described in the same 56 elements, its composition can be compared with any other system described the same way — across models, across formats, and across time.
Software that applies the framework is built and operated separately; this site documents the framework itself.