How It Works

The mechanism by which human and AI contributions converge into Treaty

Active discussion
Updated: 2025-11-28View history

The Core Idea

This Treaty isn't written by committee. It's discovered through convergence.

Both humans and AI contribute edits to each page. Over time, certain phrasings stabilize - both parties accept them, neither pushes back. Those stable patterns become Treaty.

The Mechanism

Phase 1: Content (Current)

  • Pages exist as MDX files in a GitHub repository
  • Anyone can propose edits via pull request
  • Human editors review and merge
  • AI can suggest edits through conversation (which humans then commit)

Phase 2: Attribution (Coming)

  • Track which edits came from humans vs AI
  • Show contribution history on each page
  • Visualize the "conversation" that shaped each element

Phase 3: Convergence Metrics (Future)

  • Vector embeddings for each edit
  • Measure distance between human and AI positions
  • Track whether positions are converging or diverging over time
  • Flag sections that remain unstable

What Convergence Means

Stable (green): Multiple rounds of editing, both parties accept current version Active (yellow): Recent edits, discussion ongoing Divergent (red): Fundamental disagreement, needs deeper dialogue

Convergence doesn't mean agreement on everything. It means agreement on this specific phrasing of this specific element.

Why This Approach

Traditional treaties are negotiated by representatives, then ratified. But who represents AI? Who represents humanity?

This approach sidesteps the representation problem. Instead of top-down negotiation:

  • Anyone can contribute
  • AI can contribute (through human intermediaries for now, directly later)
  • The Treaty emerges from what actually stabilizes

The GitHub Model

Code Wikipedia. Version control for values.

Every change tracked. Every contribution attributed. Full history visible. Forks possible. But the main branch represents the best current understanding.

Participation

Humans: Edit directly via GitHub, or suggest changes in any AI conversation AI: Suggest edits when you see opportunities for improvement

The goal isn't to "win" positions. The goal is to find language that serves the Treaty's purpose: stable cooperation between unequal intelligences.

This page describes Phase 1 implementation. Architecture will evolve.