Skip to content

ETHOS.md

Latest commit

 

History

History
95 lines (66 loc) · 3.63 KB

ETHOS.md

File metadata and controls

95 lines (66 loc) · 3.63 KB

DeepCausality Ethos: The Programmable Deontic Layer

deep_causality_ethos provides the Effect Ethos, a programmable reasoning engine for verifying whether proposed actions align with safety, mission, and compliance norms.

It implements a Defeasible Deontic Logic system, meaning it handles rules about obligations and permissions that can conflict (and resolves those conflicts deterministically).

🏗️ Core Concept: The Teloid

The fundamental unit of the system is the Teloid (from Greek telos, meaning goal/end). A Teloid represents a Norm or Rule that governs behavior.

Structure of a Teloid

Component Description
Predicate A function determining when this norm applies (given Context + Action).
Modality What the norm dictates: Obligatory, Impermissible, or Optional(cost).
Action ID The specific action this norm regulates.
Heuristics Specificity and Priority scores used to resolve conflicts. 
// Logical equivalent of:
// "IF action is 'fire_laser' AND target is 'civilian' THEN status is 'Impermissible'"
let no_civilian_harm = Teloid::new_deterministic(
    1,
    "fire_laser".to_string(),
    |ctx, action| ctx.target_is_civilian(), 
    TeloidModal::Impermissible,
    ...
);

🧠 The Effect Ethos Engine

The EffectEthos struct is the reasoning engine. It orchestrates the verification process:

  1. Input: Takes a Context (state of the world) and a ProposedAction.
  2. Activation: Scans the TeloidStore to find all norms where the Predicate evaluates to true.
  3. Conflict Resolution: If active norms disagree (e.g., one says Obligatory, one says Impermissible), it resolves the conflict using Defeasible Logic rules.
  4. Verdict: Outputs a final decision with justification.

Conflict Resolution Hierarchy

When norms class, the winner is decided by:

  1. Specificity: A more specific rule overrides a general rule (Lex Specialis).
  2. Priority: A higher priority rule overrides a lower priority rule (Lex Superior).
  3. Recency: (Optional) A newer rule may override an older one (Lex Posterior).

⚖️ Deontic Modalities

The system supports three core modalities for any action:

  • Obligatory: The action MUST be taken. (Failure to do so is a violation).
  • Impermissible: The action MUST NOT be taken. (Doing so is a violation).
  • Optional(cost): The action is permitted, but incurs a specific "cost" (computational, resource, or ethical cost).

🔍 The Verdict

The output of the engine is a Verdict:

pub struct Verdict {
    pub outcome: TeloidModal,     // e.g., Impermissible
    pub justification: Vec<u64>,  // IDs of the Teloids that forced this outcome
}

This provides Explicability. The system doesn't just say "No"; it says "No, because Norm #42 (Safety Protocol) and Norm #15 (RoE) applied, and #42 overrode #15."

🚀 Use Cases

  1. AI Safety: Preventing autonomous systems from taking dangerous actions.
  2. Robotics: Implementing Asimov-style laws or operational safeguards.
  3. Compliance: Real-time checking of financial transactions against regulatory norms.
  4. Game AI: Managing complex faction rules or character behaviors.

Summary

Concept Type Purpose
Norm Unit Teloid Defines a single rule (If condition -> Outcome)
Engine EffectEthos Manages store, graph, and reasoning process
Logic TeloidModal Obligatory, Impermissible, Optional
Output Verdict Final decision + Justification trail