Coding5s Research Lab · Machine-to-Machine Semantic Notation
What If Machines Didn’t Need
Human Prose to Exchange Meaning?
M2M Semantic Notation is a Crazy Idea exploring whether AI systems could exchange meaning through a shared semantic representation designed primarily for machines—while remaining translatable back into human language. It is not intended to become another programming language.
Repeated instructions and state made dynamically generated prompts increasingly difficult to fit and maintain.
ECP explored externalizing and rehydrating selected state between otherwise disposable AI executions.
Could AI systems exchange semantic information without wrapping every machine-to-machine handoff in conventional human prose?
If no human is participating in an AI-to-AI exchange, must the communication still be represented as conventional human prose?
Four Non-Negotiable Design Principles
Whatever M2M eventually becomes, the experiment only makes sense if these properties survive.
Human-Translatable
Machines may exchange compact semantics without human prose, but the meaning must remain recoverable. Humans do not need to speak M2M fluently; they must be able to inspect what it means.
AI-Native
The long-term goal is direct machine interpretation—not silently translating every M2M message into English before the model can understand it.
not: M2M → English → AI
Linear, Multidimensional Meaning
The representation should remain serializable as a simple linear stream while Conceptualization Signs express grouping, parallel dimensions, relationships, and combined concepts.
Universal Semantic Layer
The notation should aim beyond one vendor, one model, one human language, one programming language, or one Coding5s workflow.
Shared notation without shared spoken language.
A musical score encodes relationships such as pitch, timing, duration, and structure. People from different linguistic backgrounds can interpret the same representation because they share the notation.
Linear symbols. Richer relationships.
Mathematical expressions are written sequentially, yet operators and grouping symbols let a short line represent relationships that are richer than its surface order.
linear representation → structured meaning
02 · The M2M Hypothesis
Linear on the Surface.
Multidimensional in Meaning.
The current M2M syntax is only a bootstrap notation: a human-inspectable experiment for exploring how meaning, state, constraints, and relationships might be compressed into a machine-oriented semantic stream.
Stage: 3
Learner State: 1
Current Problem: Unclosed do-block
Pedagogy: Strict Socratic guidance
Boundary: Do not generate the solution code
Conceptualization Signs
M2M remains a linear stream, but grouping and relationship symbols allow that line to represent multiple conceptual dimensions at the same time.
Connects a concept or category with its value.
Separates independent conceptual dimensions in the stream.
Groups related semantic structure into a shared conceptual unit.
Combines related or simultaneous concepts within a dimension.
Still partially dependent on human language.
Current identifiers remain understandable because they preserve abbreviated English semantics.
SOCRATIC_STRICT
NO_CODE_GEN
Meaning without a natural-language translation layer.
The Crazy Idea is that AI systems could eventually interpret and emit the semantic representation directly while preserving human translatability when inspection is required.
The Idea Gets Interesting at Agent-Swarm Scale.
M2M becomes most meaningful when many specialized agents exchange state, constraints, findings, and task context in workflows where humans may never read most intermediate messages.
↓
Natural-language explanation
↓
Agent B
↓
Another natural-language explanation
↓
Agent C
↓
Semantic State
↓
Agent B
↓
Semantic State
↓
Agent C
Three Related Research Ideas. Different Layers.
Explores how AI systems might represent meaning to one another.
Can use M2M as one possible representation for persistence and rehydration.
Adds language, tone, culture, pedagogy, and communication style when the machine state must become human-facing output.
03 · Research Frontier
Maybe It Works.
Maybe Something Else Wins.
M2M is not published as a finished protocol. It is published as a testable direction in a much larger research space around agent communication, semantic compression, interoperability, and machine-native representations.
What M2M Is Exploring
- A semantic representation more compact than conventional prose.
- Meaning that remains translatable and inspectable by humans.
- Direct AI interpretation as a long-term native capability.
- A shared semantic layer that could potentially cross models, vendors, languages, and domains.
- Communication patterns that may become relevant as agent-to-agent traffic scales.
What M2M Has Not Demonstrated
- A universal semantic grammar.
- Reliable interpretation across heterogeneous models.
- Native AI support for M2M semantics.
- Reliable token or latency reduction.
- Improved multi-agent performance.
- Superiority over natural language, JSON, existing protocols, or latent communication.
The Questions That Actually Matter
A compact notation is only useful if the representation survives real technical constraints. These are some of the problems that would need to be measured.
Fewer characters do not automatically produce fewer model tokens.
Compression becomes useless if essential meaning disappears.
Different models may interpret the same compact identifier differently.
Dense semantic structures may amplify malformed symbols or incorrect values.
Fine-tuning, tokenizer adaptation, specialized training—or something else entirely?
Neighboring Research
M2M exists inside a broader research space exploring alternatives to unrestricted natural-language communication between AI agents.
Agora — A Scalable Communication Protocol for Networks of Large Language Models
Explores agent networks that can use standardized routines, natural language, or dynamically generated communication routines depending on the interaction.
Read Research Paper →Interlat — Enabling Agents to Communicate Entirely in Latent Space
Explores inter-agent communication through model hidden representations rather than conventional textual messages.
Read Research Paper →LatentMAS — Latent Collaboration in Multi-Agent Systems
Explores multi-agent collaboration using continuous latent representations and shared latent working memory.
Read Research Paper →Rich, expressive, culturally meaningful, and naturally readable by people.
Schemas and explicit fields provide predictable machine-readable structure.
A proposed middle layer: machine-oriented semantics that remain translatable to humans.
Machine-internal representations that may be highly native but are not inherently human-translatable.
Could AI systems share a universal, human-translatable, AI-native semantic notation that remains linear in representation, multidimensional in meaning, and efficient enough to improve large-scale machine-to-machine communication?
Maybe natural language remains the best general-purpose interface. Maybe structured schemas are enough. Maybe latent communication eventually makes textual M2M unnecessary. Or maybe there is useful territory between those approaches. M2M exists to explore that possibility.