Skip to Content

Changes for page FAQ

Last modified by Robert Schaub on 2025/12/24 20:33
From version 1.1
edited by Robert Schaub
on 2025/12/14 22:27
Change comment: Imported from XAR
To version 3.2
edited by Robert Schaub
on 2025/12/16 21:39
Change comment: Renamed back-links.

Summary

Details

Page properties
Content
... ... @@ -6,23 +6,74 @@
6 6  
7 7  == How do facts get input into the system? ==
8 8  
9 -FactHarbor uses a hybrid model:
9 +FactHarbor uses a hybrid model combining three complementary approaches:
10 10  
11 -**1. AI-Generated (scalable)**: System dynamically researches claims—extracting, generating structured sub-queries, performing mandatory contradiction search (actively seeking counter-evidence, not just confirmations), running quality gates. Published with clear "AI-Generated" labels.
11 +=== 1. AI-Generated Content (Scalable) ===
12 12  
13 -**2. Expert-Authored (authoritative)**: Domain experts directly author, edit, and validate content—especially for high-risk domains (medical, legal). These get "Human-Reviewed" status and higher trust.
13 +**What**: System dynamically researches claims using AKEL (AI Knowledge Extraction Layer)
14 14  
15 -**3. Audit-Improved (continuous quality)**: Sampling audits (30-50% high-risk, 5-10% low-risk) where expert reviews systematically improve AI research quality.
15 +**Process**:
16 16  
17 -**Why both matter**:
18 -* AI research handles scale—emerging claims, immediate responses with transparent reasoning
19 -* Expert authoring provides authoritative grounding for critical domains
20 -* Audit feedback ensures AI quality improves based on expert validation patterns
17 +* Extracts claims from submitted text
18 +* Generates structured sub-queries
19 +* Performs **mandatory contradiction search** (actively seeks counter-evidence, not just confirmations)
20 +* Runs automated quality gates
21 +* Publishes with clear "AI-Generated" labels
21 21  
22 -Experts can author high-priority content directly, validate/edit AI outputs, or audit samples to improve system-wide performance—focusing their time where expertise matters most.
23 +**Publication**: Mode 2 (public, AI-labeled) when quality gates pass
23 23  
24 -POC v1 demonstrates the AI research pipeline (fully automated with transparent reasoning); full system supports all three pathways.
25 +**Purpose**: Handles scale — emerging claims get immediate responses with transparent reasoning
25 25  
27 +=== 2. Expert-Authored Content (Authoritative) ===
28 +
29 +**What**: Domain experts directly author, edit, and validate content
30 +
31 +**Focus**: High-risk domains (medical, legal, safety-critical)
32 +
33 +**Publication**: Mode 3 ("Human-Reviewed" status) with expert attribution
34 +
35 +**Authority**: Tier A content requires expert approval
36 +
37 +**Purpose**: Provides authoritative grounding for critical domains where errors have serious consequences
38 +
39 +=== 3. Audit-Improved Quality (Continuous) ===
40 +
41 +**What**: Sampling audits where experts review AI-generated content
42 +
43 +**Rates**:
44 +
45 +* High-risk (Tier A): 30-50% sampling
46 +* Medium-risk (Tier B): 10-20% sampling
47 +* Low-risk (Tier C): 5-10% sampling
48 +
49 +**Impact**: Expert feedback systematically improves AI research quality
50 +
51 +**Purpose**: Ensures AI quality evolves based on expert validation patterns
52 +
53 +=== Why All Three Matter ===
54 +
55 +**Complementary Strengths**:
56 +
57 +* **AI research**: Scale and speed for emerging claims
58 +* **Expert authoring**: Authority and precision for critical domains
59 +* **Audit feedback**: Continuous quality improvement
60 +
61 +**Expert Time Optimization**:
62 +
63 +Experts can choose where to focus their time:
64 +
65 +* Author high-priority content directly
66 +* Validate and edit AI-generated outputs
67 +* Audit samples to improve system-wide AI performance
68 +
69 +This focuses expert time where domain expertise matters most while leveraging AI for scale.
70 +
71 +=== Current Status ===
72 +
73 +**POC v1**: Demonstrates the AI research pipeline (fully automated with transparent reasoning and quality gates)
74 +
75 +**Full System**: Will support all three pathways with integrated workflow
76 +
26 26  ----
27 27  
28 28  == What prevents FactHarbor from becoming another echo chamber? ==
... ... @@ -30,6 +30,7 @@
30 30  FactHarbor includes multiple safeguards against echo chambers and filter bubbles:
31 31  
32 32  **Mandatory Contradiction Search**:
84 +
33 33  * AI must actively search for counter-evidence, not just confirmations
34 34  * System checks for echo chamber patterns in source clusters
35 35  * Flags tribal or ideological source clustering
... ... @@ -36,21 +36,25 @@
36 36  * Requires diverse perspectives across political/ideological spectrum
37 37  
38 38  **Multiple Scenarios**:
91 +
39 39  * Claims are evaluated under different interpretations
40 40  * Reveals how assumptions change conclusions
41 41  * Makes disagreements understandable, not divisive
42 42  
43 43  **Transparent Reasoning**:
97 +
44 44  * All assumptions, definitions, and boundaries are explicit
45 45  * Evidence chains are traceable
46 46  * Uncertainty is quantified, not hidden
47 47  
48 48  **Audit System**:
103 +
49 49  * Human auditors check for bubble patterns
50 50  * Feedback loop improves AI search diversity
51 51  * Community can flag missing perspectives
52 52  
53 53  **Federation**:
109 +
54 54  * Multiple independent nodes with different perspectives
55 55  * No single entity controls "the truth"
56 56  * Cross-node contradiction detection
... ... @@ -62,20 +62,23 @@
62 62  This is exactly what FactHarbor is designed for:
63 63  
64 64  **Scenarios capture contexts**:
121 +
65 65  * Each scenario defines specific boundaries, definitions, and assumptions
66 66  * The same claim can have different verdicts in different scenarios
67 67  * Example: "Coffee is healthy" depends on:
68 - ** Definition of "healthy" (reduces disease risk? improves mood? affects specific conditions?)
69 - ** Population (adults? pregnant women? people with heart conditions?)
70 - ** Consumption level (1 cup/day? 5 cups/day?)
71 - ** Time horizon (short-term? long-term?)
125 +** Definition of "healthy" (reduces disease risk? improves mood? affects specific conditions?)
126 +** Population (adults? pregnant women? people with heart conditions?)
127 +** Consumption level (1 cup/day? 5 cups/day?)
128 +** Time horizon (short-term? long-term?)
72 72  
73 73  **Truth Landscape**:
131 +
74 74  * Shows all scenarios and their verdicts side-by-side
75 75  * Users see *why* interpretations differ
76 76  * No forced consensus when legitimate disagreement exists
77 77  
78 78  **Explicit Assumptions**:
137 +
79 79  * Every scenario states its assumptions clearly
80 80  * Users can compare how changing assumptions changes conclusions
81 81  * Makes context-dependence visible, not hidden
... ... @@ -85,6 +85,7 @@
85 85  == What makes FactHarbor different from traditional fact-checking sites? ==
86 86  
87 87  **Traditional Fact-Checking**:
147 +
88 88  * Binary verdicts: True / Mostly True / False
89 89  * Single interpretation chosen by fact-checker
90 90  * Often hides legitimate contextual differences
... ... @@ -91,6 +91,7 @@
91 91  * Limited ability to show *why* people disagree
92 92  
93 93  **FactHarbor**:
154 +
94 94  * **Multi-scenario**: Shows multiple valid interpretations
95 95  * **Likelihood-based**: Ranges with uncertainty, not binary labels
96 96  * **Transparent assumptions**: Makes boundaries and definitions explicit
... ... @@ -103,6 +103,7 @@
103 103  == How do you prevent manipulation or coordinated misinformation campaigns? ==
104 104  
105 105  **Quality Gates**:
167 +
106 106  * Automated checks before AI-generated content publishes
107 107  * Source quality verification
108 108  * Mandatory contradiction search
... ... @@ -109,11 +109,13 @@
109 109  * Bubble detection for coordinated campaigns
110 110  
111 111  **Audit System**:
174 +
112 112  * Stratified sampling catches manipulation patterns
113 113  * Expert auditors validate AI research quality
114 114  * Failed audits trigger immediate review
115 115  
116 116  **Transparency**:
180 +
117 117  * All reasoning chains are visible
118 118  * Evidence sources are traceable
119 119  * AKEL involvement clearly labeled
... ... @@ -120,11 +120,13 @@
120 120  * Version history preserved
121 121  
122 122  **Moderation**:
187 +
123 123  * Moderators handle abuse, spam, coordinated manipulation
124 124  * Content can be flagged by community
125 125  * Audit trail maintained even if content hidden
126 126  
127 127  **Federation**:
193 +
128 128  * Multiple nodes with independent governance
129 129  * No single point of control
130 130  * Cross-node contradiction detection
... ... @@ -137,6 +137,7 @@
137 137  FactHarbor is designed for evolving knowledge:
138 138  
139 139  **Automatic Re-evaluation**:
206 +
140 140  1. New evidence arrives
141 141  2. System detects affected scenarios and verdicts
142 142  3. AKEL proposes updated verdicts
... ... @@ -145,16 +145,19 @@
145 145  6. Old versions remain accessible
146 146  
147 147  **Version History**:
215 +
148 148  * Every verdict has complete history
149 149  * Users can see "as of date X, what did we know?"
150 150  * Timeline shows how understanding evolved
151 151  
152 152  **Transparent Updates**:
221 +
153 153  * Reason for re-evaluation documented
154 154  * New evidence clearly linked
155 155  * Changes explained, not hidden
156 156  
157 157  **User Notifications**:
227 +
158 158  * Users following claims are notified of updates
159 159  * Can compare old vs new verdicts
160 160  * Can see which evidence changed conclusions
... ... @@ -166,6 +166,7 @@
166 166  **Anyone** - even without login:
167 167  
168 168  **Readers** (no login required):
239 +
169 169  * Browse and search all published content
170 170  * Submit text for analysis
171 171  * New claims added automatically unless duplicates exist
... ... @@ -172,6 +172,7 @@
172 172  * System deduplicates and normalizes
173 173  
174 174  **Contributors** (logged in):
246 +
175 175  * Everything Readers can do
176 176  * Submit evidence sources
177 177  * Suggest scenarios
... ... @@ -178,6 +178,7 @@
178 178  * Participate in discussions
179 179  
180 180  **Workflow**:
253 +
181 181  1. User submits text (as Reader or Contributor)
182 182  2. AKEL extracts claims
183 183  3. Checks for existing duplicates
... ... @@ -194,6 +194,7 @@
194 194  Risk tiers determine review requirements and publication workflow:
195 195  
196 196  **Tier A (High Risk)**:
270 +
197 197  * **Domains**: Medical, legal, elections, safety, security, major financial
198 198  * **Publication**: AI can publish with warnings, expert review required for "Human-Reviewed" status
199 199  * **Audit rate**: Recommendation 30-50%
... ... @@ -200,6 +200,7 @@
200 200  * **Why**: Potential for significant harm if wrong
201 201  
202 202  **Tier B (Medium Risk)**:
277 +
203 203  * **Domains**: Complex policy, science causality, contested issues
204 204  * **Publication**: AI can publish immediately with clear labeling
205 205  * **Audit rate**: Recommendation 10-20%
... ... @@ -206,6 +206,7 @@
206 206  * **Why**: Nuanced but lower immediate harm risk
207 207  
208 208  **Tier C (Low Risk)**:
284 +
209 209  * **Domains**: Definitions, established facts, historical data
210 210  * **Publication**: AI publication default
211 211  * **Audit rate**: Recommendation 5-10%
... ... @@ -212,6 +212,7 @@
212 212  * **Why**: Well-established, low controversy
213 213  
214 214  **Assignment**:
291 +
215 215  * AKEL suggests tier based on domain, keywords, impact
216 216  * Moderators and Experts can override
217 217  * Risk tiers reviewed based on audit outcomes
... ... @@ -221,6 +221,7 @@
221 221  == How does federation work and why is it important? ==
222 222  
223 223  **Federation Model**:
301 +
224 224  * Multiple independent FactHarbor nodes
225 225  * Each node has own database, AKEL, governance
226 226  * Nodes exchange claims, scenarios, evidence, verdicts
... ... @@ -227,6 +227,7 @@
227 227  * No central authority
228 228  
229 229  **Why Federation Matters**:
308 +
230 230  * **Resilience**: No single point of failure or censorship
231 231  * **Autonomy**: Communities govern themselves
232 232  * **Scalability**: Add nodes to handle more users
... ... @@ -234,6 +234,7 @@
234 234  * **Trust diversity**: Multiple perspectives, not single truth source
235 235  
236 236  **How Nodes Exchange Data**:
316 +
237 237  1. Local node creates versions
238 238  2. Builds signed bundle
239 239  3. Pushes to trusted neighbor nodes
... ... @@ -242,6 +242,7 @@
242 242  6. Local re-evaluation if needed
243 243  
244 244  **Trust Model**:
325 +
245 245  * Trusted nodes → auto-import
246 246  * Neutral nodes → import with review
247 247  * Untrusted nodes → manual only
... ... @@ -253,16 +253,19 @@
253 253  **Yes - and that's a feature, not a bug**:
254 254  
255 255  **Multiple Scenarios**:
337 +
256 256  * Experts can create different scenarios with different assumptions
257 257  * Each scenario gets its own verdict
258 258  * Users see *why* experts disagree (different definitions, boundaries, evidence weighting)
259 259  
260 260  **Parallel Verdicts**:
343 +
261 261  * Same scenario, different expert interpretations
262 262  * Both verdicts visible with expert attribution
263 263  * No forced consensus
264 264  
265 265  **Transparency**:
349 +
266 266  * Expert reasoning documented
267 267  * Assumptions stated explicitly
268 268  * Evidence chains traceable
... ... @@ -269,6 +269,7 @@
269 269  * Users can evaluate competing expert opinions
270 270  
271 271  **Federation**:
356 +
272 272  * Different nodes can have different expert conclusions
273 273  * Cross-node branching allowed
274 274  * Users can see how conclusions vary across nodes
... ... @@ -280,6 +280,7 @@
280 280  **Multiple Safeguards**:
281 281  
282 282  **Quality Gate 4: Structural Integrity**:
368 +
283 283  * Fact-checking against sources
284 284  * No hallucinations allowed
285 285  * Logic chain must be valid and traceable
... ... @@ -286,6 +286,7 @@
286 286  * References must be accessible and verifiable
287 287  
288 288  **Evidence Requirements**:
375 +
289 289  * Primary sources required
290 290  * Citations must be complete
291 291  * Sources must be accessible
... ... @@ -292,11 +292,13 @@
292 292  * Reliability scored
293 293  
294 294  **Audit System**:
382 +
295 295  * Human auditors check AI-generated content
296 296  * Hallucinations caught and fed back into training
297 297  * Patterns of errors trigger system improvements
298 298  
299 299  **Transparency**:
388 +
300 300  * All reasoning chains visible
301 301  * Sources linked
302 302  * Users can verify claims against sources
... ... @@ -303,6 +303,7 @@
303 303  * AKEL outputs clearly labeled
304 304  
305 305  **Human Oversight**:
395 +
306 306  * Tier A requires expert review for "Human-Reviewed" status
307 307  * Audit sampling catches errors
308 308  * Community can flag issues
... ... @@ -314,6 +314,7 @@
314 314  [ToDo: Business model and sustainability to be defined]
315 315  
316 316  Potential models under consideration:
407 +
317 317  * Non-profit foundation with grants and donations
318 318  * Institutional subscriptions (universities, research organizations, media)
319 319  * API access for third-party integrations
... ... @@ -330,5 +330,4 @@
330 330  * [[AKEL (AI Knowledge Extraction Layer)>>FactHarbor.Specification.AI Knowledge Extraction Layer (AKEL).WebHome]]
331 331  * [[Automation>>FactHarbor.Specification.Automation.WebHome]]
332 332  * [[Federation & Decentralization>>FactHarbor.Specification.Federation & Decentralization.WebHome]]
333 -* [[Mission & Purpose>>FactHarbor.Organisation.Mission & Purpose.WebHome]]
334 -
424 +* [[Mission & Purpose>>FactHarbor.Organisation V0\.9\.18.Core Problems FactHarbor Solves.WebHome]]