In 1869, Dmitri Mendeleev arranged the known chemical elements into a periodic table. The arrangement wasn’t arbitrary — it reflected deep structural relationships in matter that physicists wouldn’t fully explain for another 50 years.
The periodic table wasn’t just a classification system. It was a discovery engine. Gaps in the table predicted elements that hadn’t been found yet. The structure of the table revealed the structure of the atom.
A periodic table for the digital world
Holosynthics began with a deceptively simple question: is there an analogous structure in artificial intelligence?
Not a metaphor. Not a loose analogy. A literal organizational structure — irreducible building blocks from which all AI cognition is composed, organized by type, predictable in their properties, and universally present across different AI systems.
We found it.
The 56 elements
Every model in the published validation set — 11 architectures, ranging from 117M to 9B parameters — contains the same 56 organizational types, present in consistent proportions.
These elements are organized into 8 fundamental types:
- Boolean — truth values, polarity, existence
- Number — quantities, magnitudes, proportions
- Symbol — tokens, code, structure
- Identity — people, places, organizations
- Space — positions, directions, regions
- Time — dates, eras, sequences
- Relation — causation, contrast, reference
- Entropy — uncertainty, confidence, ambiguity
These aren’t engineered categories. They emerged from analysis of how AI models organize their internal representations. The structure was observed, not designed.
Why this is surprising
The 11 models in the validation set were built by different organizations, trained on different data, and designed with different architectures. There’s no reason to expect them to organize knowledge the same way.
Yet they do. All 56 elements appear in 10 of 11 models. The proportions are different across models — a model trained on more code has more Symbol elements; a model trained on news has more Time elements — but the organizational types are the same.
This suggests the 56-element structure isn’t an artifact of any particular training approach. It’s something deeper: the natural way that distributed numerical systems organize semantic content.
What this enables
The periodic table of chemistry enabled not just classification, but manipulation. Knowing the properties of elements let chemists design compounds, predict reactions, and synthesize new materials.
The periodic table of digital matter enables the same for AI. Knowing an AI model’s element composition lets us:
- Diagnose where problems originate
- Predict failure modes before they appear
- Apply corrections without retraining
- Monitor for drift over time
This is Holosynthics: the science of digital matter.