ARIA Opportunity Seeds: Mathematics for Safe AI

DRAFT — NOT SUBMITTED. Operator review required before submission. Deadline: 31 July 2026.

1. The Idea

AI alignment today faces a structural problem: binary safety classifiers (safe/unsafe) cannot express the nuanced ethical trade-offs that real-world deployment demands. The Eden Protocol proposes a ternary ethical routing architecture where every consequential AI decision passes through a three-state logic gate — Allow (+1), Investigate (0), Deny (-1) — rather than a binary safe/unsafe binary.

The mathematical foundation is balanced ternary arithmetic (Setun 1958, Brusentsov), applied here specifically to alignment routing. The core insight: ethical reasoning is inherently ternary (yes/no/defer-to-human), and forcing it into binary creates exactly the brittle safety behaviours observed in deployed systems.

2. Evidence Anchors

3. Budget

CategoryYear 1Year 2Total
Researcher stipend£90,000£95,000£185,000
Compute and infrastructure£40,000£45,000£85,000
Open-source development£35,000£35,000£70,000
Evaluation and testing£25,000£30,000£55,000
Publication£15,000£15,000£30,000
Contingency (10%)£20,500£22,000£42,500
Total£225,500£242,000£467,500

4. Honest Limits

Bus factor of one. Independent research programme — no formal external peer review beyond documented institutional convergences. Prototype at TRL 0-1. No institutional affiliation.