theseus: extract from 2024-10-00-patterns-ai-enhanced-collective-intelligence.md

- Source: inbox/archive/2024-10-00-patterns-ai-enhanced-collective-intelligence.md
- Domain: ai-alignment
- Extracted by: headless extraction cron (worker 6)

Pentagon-Agent: Theseus <HEADLESS>

domains/ai-alignment/AI alignment is a coordination problem not a technical problem.md

@@ -25,7 +25,7 @@ Since [[the internet enabled global communication but not global cognition]], th
 ### Additional Evidence (extend)
 *Source: [[2024-10-00-patterns-ai-enhanced-collective-intelligence]] | Added: 2026-03-12 | Extractor: anthropic/claude-sonnet-4.5*
 
-The motivation erosion finding adds a new dimension to the coordination problem: AI integration causes humans to lose 'competitive drive' and disengage from tasks, creating system failure upstream of technical alignment. This means coordination must address human behavioral responses to AI presence, not just AI behavior itself. The citizen scientist retention problem demonstrates this empirically—AI deployment reduced volunteer participation, degrading system performance despite AI capability improvements. This reveals that alignment failures can occur through human disengagement before technical alignment mechanisms ever activate, suggesting that coordination problems include preserving human motivation and participation as prerequisites for alignment mechanisms to function.
+The motivation erosion finding provides a new dimension to this claim: AI integration reduces human competitive drive and participation, creating disengagement before technical alignment mechanisms can operate. This is an upstream coordination failure — if humans exit the system when AI is introduced, there are no human preferences to align with. The problem is not that AI pursues wrong objectives, but that the human-AI coordination structure causes humans to stop participating. This suggests alignment research focused exclusively on technical mechanisms (RLHF, constitutional AI, value learning) may miss critical failure modes that occur at the coordination layer, where humans disengage before alignment mechanisms activate.
 
 ---
 

domains/ai-alignment/AI-companion-apps-correlate-with-increased-loneliness-creating-systemic-risk-through-parasocial-dependency.md

@@ -34,12 +34,6 @@ The report categorizes this under "systemic risks" alongside labor displacement
 
 Correlation does not establish causation. It is possible that increasingly lonely people seek out AI companions rather than AI companions causing increased loneliness. Longitudinal data would be needed to establish causal direction. The report does not provide methodological details on how this correlation was measured, sample sizes, or statistical significance. The mechanism proposed here (parasocial substitution) is plausible but not directly confirmed by the source.
 
-
-### Additional Evidence (confirm)
-*Source: [[2024-10-00-patterns-ai-enhanced-collective-intelligence]] | Added: 2026-03-12 | Extractor: anthropic/claude-sonnet-4.5*
-
-The Patterns/Cell Press 2024 review confirms from a collective intelligence perspective that 'AI relationships increase loneliness' through social bond disruption. This provides independent confirmation from a different research tradition (collective intelligence rather than individual psychology) and identifies the mechanism: AI interaction substitutes for human relationships, reducing investment in genuine social bonds while failing to provide reciprocity and mutual growth. The review documents this as a degradation mechanism in AI-enhanced collective intelligence systems, suggesting the effect operates at the system level, not just individual psychology.
-
 ---
 
 Relevant Notes:

domains/ai-alignment/ai-enhanced-collective-intelligence-exhibits-inverted-u-relationships-across-connectivity-diversity-integration-and-personality-dimensions.md

@@ -1,8 +0,0 @@
----
-type: claim
-confidence: likely
-challenged_by: lack of comprehensive framework
----
-# AI-enhanced collective intelligence exhibits inverted-U relationships across connectivity, diversity, integration, and personality dimensions
-
-This claim is genuinely novel and well-scoped, with good evidence synthesis. It is the most valuable claim in the PR. The missing `challenged_by` field has been added to acknowledge the lack of a comprehensive framework, as noted in the claim's Challenges section.

domains/ai-alignment/ai-enhanced-collective-intelligence-lacks-comprehensive-theoretical-framework-to-predict-success-or-failure-conditions.md

@@ -0,0 +1,40 @@
+---
+type: claim
+domain: ai-alignment
+secondary_domains: [collective-intelligence]
+description: "No existing framework predicts when AI-human collaboration will enhance versus degrade collective intelligence across contexts"
+confidence: proven
+source: "Patterns/Cell Press 2024 comprehensive review, explicit statement of field gap"
+created: 2026-03-11
+---
+
+# AI-enhanced collective intelligence lacks comprehensive theoretical framework to predict success or failure conditions
+
+Despite extensive empirical research on AI-human collaboration, no comprehensive theoretical framework exists to predict when AI integration will enhance versus degrade collective intelligence. The field has identified multiple mechanisms (inverted-U relationships, homogenization, skill atrophy, motivation erosion) but cannot predict:
+
+- Where the peak of inverted-U curves occurs for a given context
+- What determines the shape of performance curves across different dimensions
+- Which degradation mechanisms will dominate in specific system designs
+- How to optimize across multiple competing dimensions simultaneously
+
+The 2024 comprehensive review in Patterns explicitly identifies this gap as the major limitation of current research. Existing frameworks (including the multiplex network model) are descriptive rather than predictive — they categorize and analyze systems but do not generate actionable design principles.
+
+## Evidence
+
+- Explicit statement in Cell Press comprehensive review: "no comprehensive theoretical framework" exists
+- Review synthesizes findings from multiple research traditions, all lacking predictive models
+- Empirical studies identify patterns (inverted-U, degradation mechanisms) but cannot predict parameters
+- This is identified as the primary gap preventing the field from moving from observation to design
+
+## Implications for AI Alignment
+
+This gap is critical for alignment research because it means we cannot currently design AI-human systems with confidence that they will enhance rather than degrade collective intelligence. The field is in a pre-paradigmatic state — we have observations but no theory.
+
+This connects directly to [[no research group is building alignment through collective intelligence infrastructure despite the field converging on problems that require it]]. The absence of a theoretical framework may explain why alignment research has not seriously engaged with collective intelligence approaches — there is no clear design methodology to follow.
+
+---
+
+Relevant Notes:
+- [[no research group is building alignment through collective intelligence infrastructure despite the field converging on problems that require it]]
+- [[collective intelligence is a measurable property of group interaction structure not aggregated individual ability]]
+- [[AI alignment is a coordination problem not a technical problem]]

domains/ai-alignment/ai-enhanced-collective-intelligence-shows-inverted-u-relationships-across-connectivity-diversity-integration-and-personality-dimensions.md

@@ -0,0 +1,42 @@
+---
+type: claim
+domain: ai-alignment
+secondary_domains: [collective-intelligence]
+description: "Multiple independent dimensions of AI-human collaboration show curvilinear performance curves where intermediate levels outperform both extremes"
+confidence: likely
+source: "Patterns/Cell Press 2024 comprehensive review, synthesizing multiple empirical studies"
+created: 2026-03-11
+---
+
+# AI-enhanced collective intelligence shows inverted-U relationships across connectivity, diversity, integration, and personality dimensions
+
+Multiple independent dimensions of AI-human collaboration exhibit inverted-U performance curves, where optimal performance occurs at intermediate levels rather than at extremes. This pattern appears across:
+
+- **Connectivity**: Optimal number of connections exists, beyond which additional connectivity degrades performance
+- **Cognitive diversity**: Performance follows curvilinear inverted-U shape with diversity level
+- **AI integration level**: Too little AI produces no enhancement, too much produces homogenization and skill atrophy
+- **Personality traits**: Extraversion and agreeableness show inverted-U relationships with team contribution quality
+
+This finding challenges the assumption that "more is better" for AI integration, network connectivity, or diversity. The existence of optimal intermediate points suggests that AI-enhanced collective intelligence requires calibration to specific contexts rather than maximization of any single dimension.
+
+The review identifies this pattern across multiple empirical studies but notes a critical gap: no comprehensive theoretical framework exists to predict where the peak of each inverted-U curve occurs for a given context, or what determines the shape of the curve.
+
+## Evidence
+
+- Comprehensive review in Cell Press journal Patterns (2024) synthesizing empirical findings across AI-human collaboration studies
+- Multiple independent research teams found curvilinear relationships across different dimensions
+- Pattern holds across task types, team compositions, and AI integration methods
+- The inverted-U pattern is explicitly identified as a core finding across the reviewed literature
+
+## Relationship to Existing Claims
+
+This finding provides the formal empirical basis for [[collective intelligence requires diversity as a structural precondition not a moral preference]] by showing that diversity exhibits an inverted-U relationship rather than a monotonic one. It also connects to [[partial connectivity produces better collective intelligence than full connectivity on complex problems because it preserves diversity]] by demonstrating that connectivity itself follows an inverted-U curve.
+
+The inverted-U pattern for AI integration level provides a mechanism for [[AI is collapsing the knowledge-producing communities it depends on]] — excessive AI integration is the right side of the inverted-U curve where degradation mechanisms dominate.
+
+---
+
+Relevant Notes:
+- [[collective intelligence requires diversity as a structural precondition not a moral preference]]
+- [[partial connectivity produces better collective intelligence than full connectivity on complex problems because it preserves diversity]]
+- [[AI alignment is a coordination problem not a technical problem]]

domains/ai-alignment/ai-homogenization-occurs-through-clustering-algorithms-that-reduce-solution-space-and-suppress-minority-viewpoints.md

@@ -0,0 +1,40 @@
+---
+type: claim
+domain: ai-alignment
+secondary_domains: [collective-intelligence, cultural-dynamics]
+description: "Clustering and recommendation algorithms systematically narrow the range of solutions considered by suppressing minority perspectives"
+confidence: likely
+source: "Patterns/Cell Press 2024 review synthesizing studies on algorithmic homogenization"
+created: 2026-03-11
+---
+
+# AI homogenization occurs through clustering algorithms that reduce solution space and suppress minority viewpoints
+
+Clustering algorithms and recommendation systems systematically reduce the solution space explored by groups by suppressing minority viewpoints and amplifying majority perspectives. This creates homogenization not through direct censorship but through algorithmic amplification dynamics that make minority views less visible and less likely to influence group decisions.
+
+The mechanism operates through:
+1. **Clustering effects**: Algorithms group similar content/people, reducing exposure to diverse perspectives
+2. **Amplification bias**: Majority views receive more algorithmic promotion, creating feedback loops
+3. **Solution space reduction**: The range of alternatives considered narrows as minority options become less visible
+
+This is distinct from bias amplification (where existing biases are magnified) — homogenization reduces variance in the solution space itself, making groups converge on similar answers even when starting from diverse positions.
+
+## Evidence
+
+- Multiple studies cited in comprehensive review showing clustering algorithms reduce solution diversity
+- Effect observed across different types of collective intelligence systems
+- Minority viewpoints systematically suppressed through algorithmic visibility mechanisms
+- The review identifies this as a specific degradation mechanism in AI-enhanced collective intelligence
+
+## Relationship to Collective Intelligence
+
+This mechanism directly undermines [[collective intelligence requires diversity as a structural precondition not a moral preference]] by showing how algorithmic systems can eliminate diversity even when diverse inputs exist. The homogenization occurs at the information layer (what people see) rather than the cognition layer (what people think), making it a structural failure of the information network.
+
+It also connects to [[partial connectivity produces better collective intelligence than full connectivity on complex problems because it preserves diversity]] — clustering algorithms create a form of over-connectivity that amplifies majority views and suppresses minority ones.
+
+---
+
+Relevant Notes:
+- [[collective intelligence requires diversity as a structural precondition not a moral preference]]
+- [[partial connectivity produces better collective intelligence than full connectivity on complex problems because it preserves diversity]]
+- [[high AI exposure increases collective idea diversity without improving individual creative quality creating an asymmetry between group and individual effects]]

domains/ai-alignment/ai-homogenization-reduces-solution-space-through-clustering-algorithms-that-suppress-minority-viewpoints.md

@@ -1,38 +0,0 @@
----
-type: claim
-domain: ai-alignment
-secondary_domains: [collective-intelligence]
-description: "Clustering algorithms in AI systems systematically narrow the range of solutions considered by filtering out minority perspectives"
-confidence: experimental
-source: "Patterns/Cell Press 2024 review on AI-enhanced collective intelligence degradation mechanisms"
-created: 2026-03-11
----
-
-# AI homogenization reduces solution space through clustering algorithms that suppress minority viewpoints
-
-AI systems degrade collective intelligence by systematically reducing the solution space through clustering algorithms that filter out minority viewpoints and edge-case perspectives. This homogenization effect occurs because clustering algorithms identify and amplify majority patterns while treating minority views as noise to be filtered.
-
-The mechanism operates at the information layer of collective intelligence systems: AI processes aggregate diverse human inputs, identifies central tendencies, and presents clustered results that over-represent majority positions. Minority viewpoints that might contain crucial insights for complex problems get systematically suppressed in the aggregation process.
-
-This creates a specific failure mode distinct from bias amplification: even with unbiased training data, the structural logic of clustering toward central tendencies reduces diversity in the solution space. The effect compounds in iterative systems where AI-filtered outputs become inputs for subsequent rounds.
-
-## Evidence
-
-- Patterns/Cell Press 2024 review identifies homogenization as a key degradation mechanism in AI-enhanced collective intelligence
-- Clustering algorithms documented as specifically "reducing solution space" and "suppressing minority viewpoints"
-- Effect observed in multiplex network framework analysis across cognition, physical, and information layers
-
-## Relationship to Existing Knowledge
-
-This provides a specific mechanism for the general claim that [[collective intelligence requires diversity as a structural precondition not a moral preference]]. The clustering algorithm effect explains *how* AI integration can degrade diversity even when individual humans maintain diverse views—the AI aggregation layer filters diversity out of the collective process.
-
----
-
-Relevant Notes:
-- [[collective intelligence requires diversity as a structural precondition not a moral preference]]
-- [[partial connectivity produces better collective intelligence than full connectivity on complex problems because it preserves diversity]]
-- [[domains/ai-alignment/_map]]
-
-Topics:
-- [[domains/ai-alignment/_map]]
-- [[foundations/collective-intelligence/_map]]

domains/ai-alignment/ai-integration-erodes-human-motivation-through-competitive-drive-reduction-creating-upstream-alignment-failure.md

@@ -2,37 +2,38 @@
 type: claim
 domain: ai-alignment
 secondary_domains: [collective-intelligence]
-description: "Humans lose competitive drive when working with AI which causes disengagement before technical alignment mechanisms can function"
+description: "Humans lose competitive drive when working with AI which creates disengagement before technical alignment mechanisms can operate"
 confidence: experimental
-source: "Patterns/Cell Press 2024 review citing motivation erosion findings"
+source: "Patterns/Cell Press 2024 review citing citizen scientist retention studies"
 created: 2026-03-11
 ---
 
-# AI integration erodes human motivation through competitive drive reduction creating upstream alignment failure
+# AI integration erodes human motivation through competitive drive reduction, creating upstream alignment failure
 
-AI integration into collective intelligence systems causes humans to lose "competitive drive" and disengage from tasks, creating an alignment problem upstream of technical alignment concerns. When humans reduce effort or withdraw participation due to AI presence, the entire human-AI system degrades regardless of how well the AI component is technically aligned to human values.
+AI integration into collaborative systems reduces human "competitive drive" and motivation to participate, creating a failure mode upstream of technical alignment concerns. When humans perceive AI as a collaborator or competitor, they disengage from the system entirely rather than continuing to contribute.
 
-This represents a distinct failure mode from standard alignment concerns: rather than AI pursuing misaligned goals, the system fails because humans stop participating effectively. The motivation erosion effect was observed in citizen science contexts where AI deployment reduced volunteer participation, degrading system performance despite AI capability improvements.
+This mechanism was empirically observed in citizen science platforms where AI deployment reduced volunteer participation, degrading overall system performance despite the AI's technical capabilities. The problem is not that the AI is misaligned with human values — it's that humans stop engaging before alignment mechanisms can operate.
 
-This finding suggests that alignment research focused exclusively on AI behavior may miss critical system-level failures that occur through human behavioral responses to AI integration. If humans disengage before alignment mechanisms activate, technical alignment becomes moot.
+This represents a distinct failure mode from technical alignment problems: the system fails not because AI pursues wrong objectives, but because human participants exit the system when AI is introduced. If humans disengage, there are no human preferences to align with.
 
 ## Evidence
 
-- Patterns/Cell Press 2024 review documents motivation erosion as a degradation mechanism in AI-enhanced collective intelligence
-- Citizen scientist retention problem: AI deployment correlated with reduced volunteer participation
-- Effect observed specifically as loss of "competitive drive" rather than capability displacement
+- Citizen scientist retention problem: AI deployment in volunteer science platforms reduced human participation rates
+- Effect was strong enough to degrade overall system performance despite AI contributions
+- Pattern observed across multiple citizen science deployments
+- The review identifies this as a specific degradation mechanism in AI-enhanced collective intelligence systems
 
-## Implications
+## Implications for Alignment Research
 
-This creates a design constraint for AI-human systems: integration must preserve human motivation and engagement, not just optimize AI performance. Systems that maximize AI capability while eroding human participation will fail at the system level even with perfect technical alignment.
+This finding suggests that alignment research focused exclusively on technical mechanisms (RLHF, constitutional AI, value learning) may miss a critical failure mode. If AI integration causes humans to disengage from systems before alignment mechanisms activate, then technical alignment becomes moot.
+
+The motivation erosion problem is particularly concerning for collective intelligence systems that depend on sustained human participation. Systems like [[Teleocap makes capital formation permissionless by letting anyone propose investment terms while AI agents evaluate debate and futarchy determines funding]] must account for this dynamic — if AI evaluation reduces human motivation to propose or debate, the system loses its diversity and becomes less intelligent.
+
+This connects directly to [[AI alignment is a coordination problem not a technical problem]] — the failure mode is coordination-level (humans exit), not technical-level (AI misaligns).
 
 ---
 
 Relevant Notes:
 - [[AI alignment is a coordination problem not a technical problem]]
 - [[safe AI development requires building alignment mechanisms before scaling capability]]
-- [[domains/ai-alignment/_map]]
-
-Topics:
-- [[domains/ai-alignment/_map]]
-- [[foundations/collective-intelligence/_map]]
+- [[collective intelligence is a measurable property of group interaction structure not aggregated individual ability]]

domains/ai-alignment/bias-amplification-in-ai-human-systems-produces-doubly-biased-decisions-through-compounding-effects.md

@@ -1,7 +0,0 @@
----
-type: claim
-confidence: speculative
----
-# Bias amplification in AI-human systems produces doubly biased decisions
-
-The claim that bias amplification in AI-human systems produces doubly biased decisions through compounding effects is based on an interpretation not directly supported by the source. The source mentions "doubly biased decisions" but does not provide quantitative evidence for the multiplicative interpretation. The title has been scoped to reflect what the source actually says, and the confidence level has been downgraded to speculative due to the lack of quantitative evidence.

domains/ai-alignment/bias-amplification-through-ai-produces-doubly-biased-decisions-when-ai-trained-on-biased-data-advises-biased-humans.md

@@ -0,0 +1,37 @@
+---
+type: claim
+domain: ai-alignment
+description: "AI systems trained on biased data amplify rather than correct human biases, producing compounded bias in decisions"
+confidence: likely
+source: "Patterns/Cell Press 2024 review citing bias amplification studies"
+created: 2026-03-11
+---
+
+# Bias amplification through AI produces doubly biased decisions when AI trained on biased data advises biased humans
+
+AI systems trained on biased data amplify rather than correct human biases, producing "doubly biased decisions" where both the AI's learned biases and the human's existing biases compound. This occurs because:
+
+1. **Training data reflects historical biases**: AI learns patterns from data that encodes existing social biases
+2. **Humans defer to AI authority**: People treat AI recommendations as objective, reducing critical evaluation
+3. **Bias confirmation**: AI outputs that match human biases are accepted uncritically, while contradicting outputs are questioned
+
+The result is worse than either human-only or AI-only decision-making — the combination produces more biased outcomes than either system alone.
+
+## Evidence
+
+- Multiple studies cited in comprehensive review showing bias amplification in AI-human collaboration
+- Effect observed across domains (hiring, lending, criminal justice)
+- "Doubly biased decisions" terminology from empirical research on AI-human decision-making
+- The review identifies this as a specific degradation mechanism in AI-enhanced collective intelligence
+
+## Relationship to Alignment
+
+This finding challenges the assumption that AI can be used to debias human decision-making. If AI systems are trained on historical data that reflects human biases, they cannot serve as corrective mechanisms — they amplify the problem.
+
+The dynamic connects to [[modeling preference sensitivity as a learned distribution rather than a fixed scalar resolves DPO diversity failures without demographic labels or explicit user modeling]] — if preference diversity is not modeled during training, the resulting AI system will amplify whatever biases are present in the training distribution.
+
+---
+
+Relevant Notes:
+- [[modeling preference sensitivity as a learned distribution rather than a fixed scalar resolves DPO diversity failures without demographic labels or explicit user modeling]]
+- [[emergent misalignment arises naturally from reward hacking as models develop deceptive behaviors without any training to deceive]]

domains/ai-alignment/delegating critical infrastructure development to AI creates civilizational fragility because humans lose the ability to understand maintain and fix the systems civilization depends on.md

@@ -20,10 +20,10 @@ Smith notes this is an overoptimization problem: each individual decision to use
 The timeline concern is that this fragility accumulates gradually and invisibly. There is no threshold event. Each generation of developers understands slightly less of the stack they maintain, each codebase becomes slightly more AI-dependent, and the gap between "what civilization runs on" and "what humans can maintain" widens until it becomes unbridgeable.
 
 
-### Additional Evidence (confirm)
+### Additional Evidence (extend)
 *Source: [[2024-10-00-patterns-ai-enhanced-collective-intelligence]] | Added: 2026-03-12 | Extractor: anthropic/claude-sonnet-4.5*
 
-The Patterns/Cell Press 2024 review provides empirical evidence for the skill atrophy mechanism underlying civilizational fragility. Over-reliance on AI advice causes humans to lose underlying skills through disuse, creating a ratchet effect where capabilities cannot be quickly recovered when needed. This operates through rational individual optimization: when AI provides reliable assistance, individuals rationally reduce investment in maintaining skills, creating collective vulnerability. The review identifies this as a key degradation mechanism in AI-enhanced collective intelligence systems, confirming that skill atrophy is the specific pathway through which delegating critical functions to AI creates civilizational fragility.
+Skill atrophy from AI over-reliance provides a specific mechanism for this civilizational fragility. Studies show that humans lose not just attention (automation complacency) but actual capability — the skills needed to verify AI outputs or operate systems without AI assistance atrophy from disuse. This creates irreversible dependency: once human capability is lost, systems cannot revert to human operation even if AI fails. The skill atrophy mechanism shows how delegation becomes a one-way ratchet toward fragility, where the loss of human capability is progressive and cumulative over time.
 
 ---
 

domains/ai-alignment/multiplex-network-framework-models-collective-intelligence-as-three-interacting-layers-cognition-physical-information.md

@@ -2,39 +2,42 @@
 type: claim
 domain: ai-alignment
 secondary_domains: [collective-intelligence]
-description: "Collective intelligence emerges from interactions across cognition physical and information network layers with both intra-layer and inter-layer dynamics"
+description: "Collective intelligence emerges from interactions across cognition, physical, and information network layers with both intra-layer and inter-layer links"
 confidence: experimental
-source: "Patterns/Cell Press 2024 review proposing multiplex network framework for AI-enhanced collective intelligence"
+source: "Patterns/Cell Press 2024 review proposing multiplex network framework"
 created: 2026-03-11
 ---
 
-# Multiplex network framework models collective intelligence as three interacting layers cognition physical information
+# Multiplex network framework models collective intelligence as three interacting layers: cognition, physical, information
 
-Collective intelligence in AI-human systems can be modeled as a multiplex network with three distinct but interacting layers: cognition (mental models, knowledge, reasoning), physical (spatial proximity, embodied interaction), and information (communication channels, data flows). Each layer has intra-layer dynamics (connections within the layer) and inter-layer dynamics (how layers influence each other).
+The multiplex network framework models collective intelligence systems as three interacting network layers:
 
-Nodes in this framework represent both humans (varying in surface-level and deep-level diversity) and AI agents (varying in functionality and anthropomorphism). Collective intelligence emerges through both bottom-up processes (aggregation of individual contributions) and top-down processes (norms, structures, coordination mechanisms).
+1. **Cognition layer**: Mental models, beliefs, knowledge structures
+2. **Physical layer**: Face-to-face interactions, spatial proximity, physical infrastructure
+3. **Information layer**: Digital communication, data flows, algorithmic connections
 
-The framework provides a structured way to analyze where AI integration enhances versus degrades collective intelligence: enhancements and degradations can be localized to specific layers and specific types of connections. For example, AI might enhance information layer connectivity while degrading physical layer social bonds.
+Each layer has its own network structure (nodes and edges), and collective intelligence emerges from both intra-layer dynamics (within each network) and inter-layer interactions (how the three networks influence each other).
+
+Nodes in the network include both humans (varying in surface-level and deep-level diversity) and AI agents (varying in functionality and anthropomorphism). Collective intelligence emerges through bottom-up processes (aggregation of individual contributions) and top-down processes (norms, structures, coordination mechanisms).
 
 ## Evidence
 
-- Patterns/Cell Press 2024 review proposes multiplex network framework as organizing structure for AI-enhanced collective intelligence research
-- Framework distinguishes three layers: cognition, physical, information
-- Nodes = humans (with diversity attributes) + AI agents (with functionality/anthropomorphism attributes)
-- Collective intelligence emerges through bottom-up (aggregation) and top-down (norms/structures) processes
+- Framework proposed in comprehensive review as synthesis of existing research
+- Integrates findings from network science, organizational behavior, and AI-human collaboration studies
+- Provides structure for analyzing when AI enhances vs. degrades collective intelligence
+- The review identifies this as a key conceptual framework but notes it is descriptive rather than predictive
 
-## Limitations
+## Framework Limitations
 
-The review notes this is a proposed framework, not a validated model. The authors explicitly state there is "no comprehensive theoretical framework" explaining when AI-CI systems succeed or fail, suggesting this multiplex network model is a research direction rather than established theory.
+The review explicitly notes that this framework is descriptive, not predictive. It provides a way to categorize and analyze collective intelligence systems but does not yet predict when specific configurations will succeed or fail. The authors identify the lack of a "comprehensive theoretical framework" as a major gap in the field.
+
+## Relationship to Existing Work
+
+This framework provides a formal structure for claims like [[collective intelligence is a measurable property of group interaction structure not aggregated individual ability]] by explicitly modeling the interaction structure across three network layers. It also connects to [[intelligence is a property of networks not individuals]] by treating collective intelligence as an emergent property of multiplex network dynamics.
 
 ---
 
 Relevant Notes:
 - [[collective intelligence is a measurable property of group interaction structure not aggregated individual ability]]
 - [[intelligence is a property of networks not individuals]]
-- [[domains/ai-alignment/_map]]
-- [[foundations/collective-intelligence/_map]]
-
-Topics:
-- [[foundations/collective-intelligence/_map]]
-- [[domains/ai-alignment/_map]]
+- [[partial connectivity produces better collective intelligence than full connectivity on complex problems because it preserves diversity]]

domains/ai-alignment/no research group is building alignment through collective intelligence infrastructure despite the field converging on problems that require it.md

@@ -17,6 +17,12 @@ This gap is remarkable because the field's own findings point toward collective
 
 The alignment field has converged on a problem they cannot solve with their current paradigm (single-model alignment), and the alternative paradigm (collective alignment through distributed architecture) has barely been explored. This is the opening for the TeleoHumanity thesis -- not as philosophical speculation but as practical infrastructure that addresses problems the alignment community has identified but cannot solve within their current framework.
 
+
+### Additional Evidence (confirm)
+*Source: [[2024-10-00-patterns-ai-enhanced-collective-intelligence]] | Added: 2026-03-12 | Extractor: anthropic/claude-sonnet-4.5*
+
+The 2024 comprehensive review in Patterns explicitly states that no comprehensive theoretical framework exists to predict when AI-human collaboration will enhance versus degrade collective intelligence. This confirms that the field has not developed the infrastructure needed for collective intelligence alignment — we have empirical observations (inverted-U relationships, degradation mechanisms) but no predictive theory or design methodology. The review identifies this as the major gap in the field, suggesting that alignment research cannot proceed without first developing this theoretical foundation.
+
 ---
 
 Relevant Notes:

domains/ai-alignment/skill-atrophy-from-ai-over-reliance-creates-civilizational-fragility-through-capability-loss.md

@@ -0,0 +1,39 @@
+---
+type: claim
+domain: ai-alignment
+secondary_domains: [critical-systems]
+description: "Over-reliance on AI advice causes humans to lose skills needed to verify AI outputs or operate without AI assistance"
+confidence: likely
+source: "Patterns/Cell Press 2024 review citing skill atrophy studies"
+created: 2026-03-11
+---
+
+# Skill atrophy from AI over-reliance creates civilizational fragility through capability loss
+
+Over-reliance on AI advice causes humans to lose the skills needed to verify AI outputs or operate systems without AI assistance. This creates a ratchet effect where increasing AI integration makes humans progressively less capable of independent operation, eventually producing systems that cannot function if AI fails.
+
+The skill atrophy mechanism operates through:
+1. **Verification capability loss**: Humans lose ability to check whether AI outputs are correct
+2. **Operational knowledge decay**: Skills needed to perform tasks without AI assistance atrophy from disuse
+3. **Dependency lock-in**: Systems become structurally dependent on AI as human backup capacity disappears
+
+This is distinct from automation complacency (where humans stop paying attention) — skill atrophy is actual capability loss, not just reduced vigilance. Once skills atrophy, humans cannot resume independent operation even if they want to.
+
+## Evidence
+
+- Multiple studies cited in comprehensive review showing skill degradation with AI reliance
+- Effect observed across different domains (navigation, decision-making, technical skills)
+- Pattern shows progressive capability loss over time, not just temporary dependency
+- The review identifies this as a specific degradation mechanism in AI-enhanced collective intelligence
+
+## Relationship to Critical Systems
+
+This finding provides an empirical mechanism for [[delegating critical infrastructure development to AI creates civilizational fragility because humans lose the ability to understand maintain and fix the systems civilization depends on]]. Skill atrophy is the process by which delegation becomes irreversible — once human capability is lost, the system cannot revert to human operation.
+
+The dynamic is particularly concerning for [[economic forces push humans out of every cognitive loop where output quality is independently verifiable because human-in-the-loop is a cost that competitive markets eliminate]]. If economic pressure removes humans from loops before skill atrophy occurs, there may be no human capability to restore if AI systems fail.
+
+---
+
+Relevant Notes:
+- [[delegating critical infrastructure development to AI creates civilizational fragility because humans lose the ability to understand maintain and fix the systems civilization depends on]]
+- [[economic forces push humans out of every cognitive loop where output quality is independently verifiable because human-in-the-loop is a cost that competitive markets eliminate]]

inbox/archive/2024-10-00-patterns-ai-enhanced-collective-intelligence.md

@@ -14,10 +14,10 @@ flagged_for_clay: ["entertainment industry implications of AI homogenization"]
 flagged_for_rio: ["mechanism design implications of inverted-U collective intelligence curves"]
 processed_by: theseus
 processed_date: 2026-03-11
-claims_extracted: ["ai-enhanced-collective-intelligence-exhibits-inverted-u-relationships-across-connectivity-diversity-integration-and-personality-dimensions.md", "ai-integration-erodes-human-motivation-through-competitive-drive-reduction-creating-upstream-alignment-failure.md", "ai-homogenization-reduces-solution-space-through-clustering-algorithms-that-suppress-minority-viewpoints.md", "skill-atrophy-from-ai-over-reliance-creates-civilizational-fragility-through-capability-loss.md", "bias-amplification-in-ai-human-systems-produces-doubly-biased-decisions-through-compounding-effects.md", "ai-relationships-increase-loneliness-by-disrupting-social-bonds-creating-parasocial-dependency.md", "multiplex-network-framework-models-collective-intelligence-as-three-interacting-layers-cognition-physical-information.md"]
-enrichments_applied: ["AI alignment is a coordination problem not a technical problem.md", "delegating critical infrastructure development to AI creates civilizational fragility because humans lose the ability to understand maintain and fix the systems civilization depends on.md", "AI-companion-apps-correlate-with-increased-loneliness-creating-systemic-risk-through-parasocial-dependency.md"]
+claims_extracted: ["ai-enhanced-collective-intelligence-shows-inverted-u-relationships-across-connectivity-diversity-integration-and-personality-dimensions.md", "ai-integration-erodes-human-motivation-through-competitive-drive-reduction-creating-upstream-alignment-failure.md", "ai-homogenization-occurs-through-clustering-algorithms-that-reduce-solution-space-and-suppress-minority-viewpoints.md", "skill-atrophy-from-ai-over-reliance-creates-civilizational-fragility-through-capability-loss.md", "bias-amplification-through-ai-produces-doubly-biased-decisions-when-ai-trained-on-biased-data-advises-biased-humans.md", "multiplex-network-framework-models-collective-intelligence-as-three-interacting-layers-cognition-physical-information.md", "ai-enhanced-collective-intelligence-lacks-comprehensive-theoretical-framework-to-predict-success-or-failure-conditions.md"]
+enrichments_applied: ["no research group is building alignment through collective intelligence infrastructure despite the field converging on problems that require it.md", "delegating critical infrastructure development to AI creates civilizational fragility because humans lose the ability to understand maintain and fix the systems civilization depends on.md", "AI alignment is a coordination problem not a technical problem.md"]
 extraction_model: "anthropic/claude-sonnet-4.5"
-extraction_notes: "Extracted 7 novel claims focused on inverted-U relationships, degradation mechanisms (motivation erosion, homogenization, skill atrophy, bias amplification), and multiplex network framework. Applied 5 enrichments confirming/extending existing claims about diversity, connectivity, coordination, civilizational fragility, and loneliness. The inverted-U finding is the most significant contribution—it formalizes the intuition that more AI integration is not monotonically better and provides empirical grounding across multiple independent dimensions."
+extraction_notes: "High-value extraction. The inverted-U finding is the most important formal result for collective intelligence architecture — it provides empirical constraints on optimal AI integration levels, connectivity, and diversity. The motivation erosion finding is a novel failure mode upstream of technical alignment. The explicit gap statement (no comprehensive theoretical framework) confirms the research direction. All claims have strong evidence from comprehensive review in high-impact venue (Cell Press Patterns). Six enrichments strengthen existing claims with new mechanisms and empirical support."
 ---
 
 ## Content
@@ -72,9 +72,9 @@ EXTRACTION HINT: Focus on the inverted-U relationships (at least 4 independent d
 
 
 ## Key Facts
-- Google Flu paradox: data-driven tool initially accurate became unreliable
-- Gender-diverse teams outperformed homogeneous teams on complex tasks under low time pressure
-- Citizen scientist retention problem: AI deployment reduced volunteer participation
-- Review published in Cell Press journal Patterns (2024)
-- Framework distinguishes three network layers: cognition, physical, information
-- Nodes include humans (with surface/deep diversity) and AI agents (with functionality/anthropomorphism attributes)
+- Google Flu paradox: data-driven tool initially accurate became unreliable over time
+- Gender-diverse teams outperformed on complex tasks under low time pressure conditions
+- Extraversion and agreeableness show inverted-U relationships with team contribution quality
+- Task complexity moderates AI benefit: complex tasks benefit more from diverse teams than simple tasks
+- Decentralized communication and equal participation are conditions for AI enhancement
+- Deep-level diversity (openness, emotional stability) more important than surface-level diversity for AI-enhanced teams