Wiki source code of Automation
Version 6.1 by Robert Schaub on 2025/12/14 18:59
Show last authors
| author | version | line-number | content |
|---|---|---|---|
| 1 | = Automation = | ||
| 2 | |||
| 3 | Automation in FactHarbor amplifies human capability while implementing risk-based oversight. | ||
| 4 | |||
| 5 | This chapter defines: | ||
| 6 | * Risk-based publication model | ||
| 7 | * Quality gates for AI-generated content | ||
| 8 | * What must remain human-only | ||
| 9 | * What AI (AKEL) can draft and publish | ||
| 10 | * What can be fully automated | ||
| 11 | * How automation evolves through POC → Beta 0 → Release 1.0 | ||
| 12 | |||
| 13 | == POC v1 (AI-Generated Publication Demonstration) == | ||
| 14 | |||
| 15 | The goal of POC v1 is to validate the automated reasoning capabilities and demonstrate AI-generated content publication. | ||
| 16 | |||
| 17 | === Workflow === | ||
| 18 | |||
| 19 | 1. **Input**: User pastes a block of raw text. | ||
| 20 | 1. **Deep Analysis (Background)**: The system autonomously performs the full pipeline **before** displaying the text: | ||
| 21 | * Extraction & Normalisation | ||
| 22 | * Scenario & Sub-query generation | ||
| 23 | * Evidence retrieval with **contradiction search** | ||
| 24 | * Quality gate validation | ||
| 25 | * Verdict computation | ||
| 26 | 1. **Visualisation (Extraction & Marking)**: The system displays the text with claims extracted and marked. | ||
| 27 | * **Verdict-Based Coloring**: The extraction highlights (e.g. Orange/Green) are chosen **according to the computed verdict** for each claim. | ||
| 28 | * **AI-Generated Label**: Clear indication that content is AI-produced | ||
| 29 | 1. **Inspection**: User clicks a highlighted claim to see the **Reasoning Trail**, showing exactly which evidence and sub-queries led to that verdict. | ||
| 30 | |||
| 31 | === Technical Scope === | ||
| 32 | |||
| 33 | * **AI-Generated Publication**: Content published as Mode 2 (AI-Generated, no prior human review) | ||
| 34 | * **Quality Gates Active**: All automated quality checks enforced | ||
| 35 | * **Contradiction Search Demonstrated**: Shows counter-evidence and reservation detection | ||
| 36 | * **Risk Tier Classification**: POC shows tier assignment (demo purposes) | ||
| 37 | * **No Human Approval Gate**: Demonstrates scalable AI publication | ||
| 38 | * **Structured Sub-Queries**: Logic generated by decomposing claims into the FactHarbor data model | ||
| 39 | |||
| 40 | ---- | ||
| 41 | |||
| 42 | == Publication Model == | ||
| 43 | |||
| 44 | FactHarbor implements a risk-based publication model with three modes: | ||
| 45 | |||
| 46 | === Mode 1: Draft-Only === | ||
| 47 | * Failed quality gates | ||
| 48 | * High-risk content pending expert review | ||
| 49 | * Internal review queue only | ||
| 50 | |||
| 51 | === Mode 2: AI-Generated (Public) === | ||
| 52 | * Passed all quality gates | ||
| 53 | * Risk tier B or C | ||
| 54 | * Clear AI-generated labeling | ||
| 55 | * Users can request human review | ||
| 56 | |||
| 57 | === Mode 3: Human-Reviewed === | ||
| 58 | * Validated by human reviewers/experts | ||
| 59 | * "Human-Reviewed" status badge | ||
| 60 | * Required for Tier A content publication | ||
| 61 | |||
| 62 | See [[AKEL page>>FactHarbor.Specification.AI Knowledge Extraction Layer (AKEL).WebHome]] for detailed publication mode descriptions. | ||
| 63 | |||
| 64 | ---- | ||
| 65 | |||
| 66 | == Risk Tiers and Automation Levels == | ||
| 67 | |||
| 68 | === Tier A (High Risk) === | ||
| 69 | * **Domains**: Medical, legal, elections, safety, security | ||
| 70 | * **Automation**: AI can draft, human review required for "Human-Reviewed" status | ||
| 71 | * **AI publication**: Allowed with prominent disclaimers and warnings | ||
| 72 | * **Audit rate**: Recommendation: 30-50% | ||
| 73 | |||
| 74 | === Tier B (Medium Risk) === | ||
| 75 | * **Domains**: Complex policy, science, causality claims | ||
| 76 | * **Automation**: AI can draft and publish (Mode 2) | ||
| 77 | * **Human review**: Optional, audit-based | ||
| 78 | * **Audit rate**: Recommendation: 10-20% | ||
| 79 | |||
| 80 | === Tier C (Low Risk) === | ||
| 81 | * **Domains**: Definitions, established facts, historical data | ||
| 82 | * **Automation**: AI publication default | ||
| 83 | * **Human review**: On request or via sampling | ||
| 84 | * **Audit rate**: Recommendation: 5-10% | ||
| 85 | |||
| 86 | ---- | ||
| 87 | |||
| 88 | == Human-Only Tasks == | ||
| 89 | |||
| 90 | These require human judgment and cannot be automated: | ||
| 91 | |||
| 92 | * **Ethical boundary decisions** (especially medical, political, psychological harm assessment) | ||
| 93 | * **Dispute resolution** between conflicting expert opinions | ||
| 94 | * **Governance policy** setting and enforcement | ||
| 95 | * **Final authority** on Tier A "Human-Reviewed" status | ||
| 96 | * **Audit system oversight** and quality standard definition | ||
| 97 | * **Risk tier policy** adjustments based on societal context | ||
| 98 | |||
| 99 | ---- | ||
| 100 | |||
| 101 | == AI-Draft with Audit (Semi-Automated) == | ||
| 102 | |||
| 103 | AKEL drafts these; humans validate via sampling audits: | ||
| 104 | |||
| 105 | * **Scenario structures** (definitions, assumptions, context) | ||
| 106 | * **Evaluation methods** and reasoning chains | ||
| 107 | * **Evidence relevance** assessment and ranking | ||
| 108 | * **Reliability scoring** and source evaluation | ||
| 109 | * **Verdict reasoning** with uncertainty quantification | ||
| 110 | * **Contradiction and reservation** identification | ||
| 111 | * **Scenario comparison** explanations | ||
| 112 | * **Public summaries** and accessibility text | ||
| 113 | |||
| 114 | Most Tier B and C content remains in AI-draft status unless: | ||
| 115 | * Users request human review | ||
| 116 | * Audits identify errors | ||
| 117 | * High engagement triggers review | ||
| 118 | * Community flags issues | ||
| 119 | |||
| 120 | ---- | ||
| 121 | |||
| 122 | == Fully Automated Structural Tasks == | ||
| 123 | |||
| 124 | These require no human interpretation: | ||
| 125 | |||
| 126 | * **Claim normalization** (canonical form generation) | ||
| 127 | * **Duplicate detection** (vector embeddings, clustering) | ||
| 128 | * **Evidence metadata extraction** (dates, authors, publication info) | ||
| 129 | * **Basic reliability heuristics** (source reputation scoring) | ||
| 130 | * **Contradiction detection** (conflicting statements across sources) | ||
| 131 | * **Re-evaluation triggers** (new evidence, source updates) | ||
| 132 | * **Layout generation** (diagrams, summaries, UI presentation) | ||
| 133 | * **Federation integrity checks** (cross-node data validation) | ||
| 134 | |||
| 135 | ---- | ||
| 136 | |||
| 137 | == Quality Gates (Automated) == | ||
| 138 | |||
| 139 | Before AI-draft publication (Mode 2), content must pass: | ||
| 140 | |||
| 141 | 1. **Source Quality Gate** | ||
| 142 | * Primary sources verified | ||
| 143 | * Citations complete and accessible | ||
| 144 | * Source reliability scored | ||
| 145 | |||
| 146 | 2. **Contradiction Search Gate** (MANDATORY) | ||
| 147 | * Counter-evidence actively sought | ||
| 148 | * Reservations and limitations identified | ||
| 149 | * Bubble detection (echo chambers, conspiracy theories) | ||
| 150 | * Diverse perspective verification | ||
| 151 | |||
| 152 | 3. **Uncertainty Quantification Gate** | ||
| 153 | * Confidence scores calculated | ||
| 154 | * Limitations stated | ||
| 155 | * Data gaps disclosed | ||
| 156 | |||
| 157 | 4. **Structural Integrity Gate** | ||
| 158 | * No hallucinations detected | ||
| 159 | * Logic chain valid | ||
| 160 | * References verifiable | ||
| 161 | |||
| 162 | See [[AKEL page>>FactHarbor.Specification.AI Knowledge Extraction Layer (AKEL).WebHome]] for detailed quality gate specifications. | ||
| 163 | |||
| 164 | ---- | ||
| 165 | |||
| 166 | == Audit System == | ||
| 167 | |||
| 168 | Instead of reviewing all AI output, systematic sampling audits ensure quality: | ||
| 169 | |||
| 170 | === Stratified Sampling === | ||
| 171 | * Risk tier (A > B > C sampling rates) | ||
| 172 | * Confidence scores (low confidence → more audits) | ||
| 173 | * Traffic/engagement (popular content audited more) | ||
| 174 | * Novelty (new topics/claim types prioritized) | ||
| 175 | * User flags and disagreement signals | ||
| 176 | |||
| 177 | === Continuous Improvement Loop === | ||
| 178 | Audit findings improve: | ||
| 179 | * Query templates | ||
| 180 | * Source reliability weights | ||
| 181 | * Contradiction detection algorithms | ||
| 182 | * Risk tier assignment rules | ||
| 183 | * Bubble detection heuristics | ||
| 184 | |||
| 185 | === Transparency === | ||
| 186 | * Audit statistics published | ||
| 187 | * Accuracy rates by tier reported | ||
| 188 | * System improvements documented | ||
| 189 | |||
| 190 | ---- | ||
| 191 | |||
| 192 | == Automation Roadmap == | ||
| 193 | |||
| 194 | Automation capabilities increase with system maturity while maintaining quality oversight. | ||
| 195 | |||
| 196 | === POC (Current Focus) === | ||
| 197 | |||
| 198 | **Automated:** | ||
| 199 | * Claim normalization | ||
| 200 | * Scenario template generation | ||
| 201 | * Evidence metadata extraction | ||
| 202 | * Simple verdict drafts | ||
| 203 | * **AI-generated publication** (Mode 2, with quality gates) | ||
| 204 | * **Contradiction search** | ||
| 205 | * **Risk tier assignment** | ||
| 206 | |||
| 207 | **Human:** | ||
| 208 | * High-risk content validation (Tier A) | ||
| 209 | * Sampling audits across all tiers | ||
| 210 | * Quality standard refinement | ||
| 211 | * Governance decisions | ||
| 212 | |||
| 213 | === Beta 0 (Enhanced Automation) === | ||
| 214 | |||
| 215 | **Automated:** | ||
| 216 | * Detailed scenario generation | ||
| 217 | * Advanced evidence reliability scoring | ||
| 218 | * Cross-scenario comparisons | ||
| 219 | * Multi-source contradiction detection | ||
| 220 | * Internal Truth Landscape generation | ||
| 221 | * **Increased AI-draft coverage** (more Tier B content) | ||
| 222 | |||
| 223 | **Human:** | ||
| 224 | * Tier A final approval | ||
| 225 | * Audit sampling (continued) | ||
| 226 | * Expert validation of complex domains | ||
| 227 | * Quality improvement oversight | ||
| 228 | |||
| 229 | === Release 1.0 (High Automation) === | ||
| 230 | |||
| 231 | **Automated:** | ||
| 232 | * Full scenario generation (comprehensive) | ||
| 233 | * Bayesian verdict scoring across scenarios | ||
| 234 | * Multi-scenario summary generation | ||
| 235 | * Anomaly detection across federated nodes | ||
| 236 | * AKEL-assisted cross-node synchronization | ||
| 237 | * **Most Tier B and all Tier C** auto-published | ||
| 238 | |||
| 239 | **Human:** | ||
| 240 | * Tier A oversight (still required) | ||
| 241 | * Strategic audits (lower sampling rates, higher value) | ||
| 242 | * Ethical decisions and policy | ||
| 243 | * Conflict resolution | ||
| 244 | |||
| 245 | ---- | ||
| 246 | |||
| 247 | == Automation Levels Diagram == | ||
| 248 | |||
| 249 | {{include reference="FactHarbor.Specification.Diagrams.Automation Level.WebHome"/}} | ||
| 250 | |||
| 251 | ---- | ||
| 252 | |||
| 253 | == Automation Roadmap Diagram == | ||
| 254 | |||
| 255 | {{include reference="FactHarbor.Specification.Diagrams.Automation Roadmap.WebHome"/}} | ||
| 256 | |||
| 257 | ---- | ||
| 258 | |||
| 259 | == Manual vs Automated Matrix == | ||
| 260 | |||
| 261 | {{include reference="FactHarbor.Specification.Diagrams.Manual vs Automated matrix.WebHome"/}} | ||
| 262 | |||
| 263 | ---- | ||
| 264 | |||
| 265 | == Related Pages == | ||
| 266 | |||
| 267 | * [[AKEL (AI Knowledge Extraction Layer)>>FactHarbor.Specification.AI Knowledge Extraction Layer (AKEL).WebHome]] | ||
| 268 | * [[Requirements (Roles)>>FactHarbor.Specification.Requirements.WebHome]] | ||
| 269 | * [[Workflows>>FactHarbor.Specification.Workflows.WebHome]] | ||
| 270 | * [[Governance>>FactHarbor.Organisation.Governance]] |