Some early notes on this (old) idea. Not had time yet to document, but hopefully we can introduce this into a forthcoming pilot.
# Work-in-progress Thinking about names... - Deliberative Move Recognition + **Jury Pass** (when tied to Deliberative Move Recognition) - Free Justice - Deliberative Jury (JUDE) - **First-Pass Jury-of-AIs** (plain) - **Affordable Adjudication Layer** (infrastructure)
# AI Role: Affordability-First Adjudication In this framework, AI’s value is **affordability**, not grandeur: quick, cheap, *OK-but-not-great* first passes on complex judgments that would otherwise be too costly to attempt.
We use a **first-pass jury-of-AIs** to produce a provisional view inside Living Documents, then route only the hard or contentious parts to human arbitration and continuous improvement (a Common Law-style evolution of precedent).
# First-Pass Jury-of-AIs - Multiple small “judge” models give **independent first opinions** on a matter. - Each judge discloses **reasons** (short rationale), **confidence**, and **known limits**. - An ensemble report aggregates: **majority view**, **minority views**, and **points of disagreement**. - The result is an **affordable first adjudication** that is good enough to triage and structure the human work that follows.
# Human-Centred Control - **Training set design**: curators assemble balanced, documented exemplars; every data addition has provenance and review notes. - **Diverse judge panel**: include judges trained on different sources, methods, and value framings to surface blind spots. - **Arbitration loop**: humans (and affected stakeholders) review contested points, add counter-examples, and **set binding precedent**. - **Continuous integration of precedent**: new cases and rulings are merged into the training sets via a versioned, auditable workflow (like policy CI in Policy as Code). - **Appeal paths**: any party can trigger re-examination; appeals produce new exemplars for future cases.
# Why Affordability Matters - Like the internet drove the **marginal cost of information** toward zero, first-pass AI drives the **marginal cost of context-aware adjudication** toward zero. - We **spend human attention where it counts**: on disagreements, harms, edge cases, and norm-setting—rather than on routine sorting and summarizing. - Lower adjudication cost enables **more frequent, earlier, and broader** deliberation in everyday governance, not just in rare formal processes.
# Workflow Inside Living Documents - **Submit**: a proposal or dispute is filed with minimal structure (claims, obligations, context). - **Jury pass**: the jury-of-AIs issues a provisional finding with reasons, disagreements, and suggested Deliberative Primitives to apply next. - **Arbitrate**: humans apply primitives (e.g., **Mirror role-swap**, **Care ledger attach**, **Boundary proof**), revise the plan, and record precedent. - **Integrate**: accepted precedents and new counter-examples are merged; judges are re-tested on held-out cases. - **Publish**: the document remains “living,” with versioned decisions, obligations, and monitoring triggers.
# Minimal Safeguards - **Disclosure**: every AI opinion includes model identity, data snapshot, and confidence. - **Diversity checks**: enforce panel diversity (sources, methods, values); rotate judges. - **Precedent guardrails**: precedent cannot be added without human sign-off and a short justification. - **Red-team & drift tests**: periodic stress tests; alert when distribution shifts or disagreement spikes. - **Recusal logic**: detect conflicts (e.g., data overlap) and swap in alternate judges.
# Metrics - **Time-to-first-pass** (minutes) - **Appeal rate** and **overturn rate** (health of arbitration) - **Human minutes per resolved case** (attention saved) - **Disagreement index** (where to focus improvement) - **Precedent uptake** (how often a new rule resolves future cases without appeal) - **Harm & care outcomes** (from attached ledgers and boundary checks)
# See - Living Document - Deliberative Primitive and Deliberative Move Recognition - Policy as Code - Common Law and Design Justice - Jury-of-AIs