Introduction to DIAL
DIAL (Dynamic Integration between AI and Labor) is a task-oriented coordination framework for AI and human specialists making decisions together within state machines.
Why DIAL?
The promise of AI is efficiency: faster, cheaper execution of narrow tasks. But the question organizations face isn't "Can AI do this?" It's:
How do you know, in dollars, time, and quality, exactly what it would cost to turn any task over to a minimally competent AI decision-maker? And how involved should humans remain as quality control?
DIAL provides the answer through empirical measurement.
The Core Insight
An AI model operates on a bounded context window. A human operates on a lifetime of embodied experience, tacit knowledge, institutional context, and real-time sensory input that no model has access to. The human knows things they cannot tell the machine.
Three Foundational Principles
1. Human Primacy
Humans have context that AI cannot access.
An AI specialist is judged on alignment with human choices. The standard is distributional: output should match the distribution a population of competent humans would produce for the same decision.
2. Progressive Collapse
Over repeated decision cycles, measuring how well AI predicts human choices causes the multi-agent deliberation structure to progressively collapse into deterministic execution.
As AI specialists prove their alignment with human judgment through accumulated data, the expensive deliberation process simplifies.
3. Empirical Trust
Trust is earned through demonstrated alignment with human decisions and through continued sampling of human preferences. Specialists prove their value one decision at a time.
Scope
DIAL targets decisions that humans already make well and measures whether AI specialists can replicate those decisions cheaply enough to justify delegation, with precise cost data on ongoing human quality-control. The value of AI is efficiency. AI is faster and cheaper at narrow tasks where the required context fits within the model's window.
How It Works
- Model the task as a state machine: Define states, transitions, and decision prompts
- Register proposers: AI and human specialists that propose transitions
- Run decision cycles: Propose → Arbitrate → Execute
- Reach rest: The session is at rest when it reaches its goal state
Documentation Structure
This documentation is organized into two main sections:
Concepts
The theory and philosophy behind DIAL. No code—just ideas:
- What sessions, specialists, and decision cycles are
- How arbitration and consensus work
- Why human primacy matters
- How DIAL relates to other approaches
Implementation
How to use the dialai library:
- Installation and quick start
- API reference
- Guides for building state machines and registering specialists
- Examples and deployment patterns
Key Terminology
| Term | Definition |
|---|---|
| Session | An instance of a state machine being navigated by specialists |
| Specialist | A pluggable actor (AI or human) that proposes transitions |
| Decision Cycle | The repeating process: Propose → Arbitrate → Execute |
| Arbiter | The built-in logic that evaluates consensus and determines when a proposal wins |
| Goal State | The rest state where the session is headed; no action needed when reached |