m3taversal
d7025e65dd
- Replace [[AI alignment approaches]] with [[domains/ai-alignment/_map]] in 5 foundations/collective-intelligence/ claims and 1 core/living-agents/ claim (6 fixes total — topic tag had no corresponding file) - Replace [[core/_map]] with [[foundations/collective-intelligence/_map]] in 2 CI claims (core/_map.md doesn't exist) - Add 3 new claims from PR #20 to domains/ai-alignment/_map.md: voluntary safety pledges, government supply chain designation, nuclear war escalation in LLM simulations Pentagon-Agent: Theseus <845F10FB-BC22-40F6-A6A6-F6E4D8F78465>
38 lines
No EOL
4.4 KiB
Markdown
38 lines
No EOL
4.4 KiB
Markdown
---
|
|
description: Centola, Derex-Boyd, and Lazer-Friedman independently show that fully connected networks cause premature convergence on complex problems while partially connected networks maintain the solution diversity needed for innovation -- the optimal topology depends on problem complexity and time horizon
|
|
type: claim
|
|
domain: livingip
|
|
source: Centola, The Network Science of Collective Intelligence (Trends in Cognitive Sciences, 2022); Derex and Boyd, Partial Connectivity Increases Cultural Accumulation (PNAS, 2016); Lazer and Friedman, The Network Structure of Exploration and Exploitation (ASQ, 2007)
|
|
confidence: proven
|
|
tradition: network science, collective intelligence
|
|
created: 2026-02-28
|
|
---
|
|
|
|
# partial connectivity produces better collective intelligence than full connectivity on complex problems because it preserves diversity
|
|
|
|
Three independent research programs converge on the same finding: for complex problems, full information flow kills the diversity that collective intelligence requires.
|
|
|
|
**Centola (2022)** synthesized findings across collective problem-solving and wisdom-of-crowds research. For complex problems, informational inefficiency (slower information spread) improves collective intelligence by preserving diversity of approaches. For large groups on rugged fitness landscapes, slowing down information transmission avoids premature convergence on local maxima.
|
|
|
|
**Derex and Boyd (2016)** confirmed this experimentally. Groups solving problems on a complex fitness landscape showed that fully connected groups converged on the same solution early and stagnated. Partially connected groups maintained diverse solutions longer, and this diversity enabled combinatorial innovations that fully connected groups never produced.
|
|
|
|
**Lazer and Friedman (2007)** established the temporal dimension: efficient networks win in the short run (good solutions propagate faster) but lose in the long run (diversity is sacrificed for convergence). At intermediate time horizons, moderately connected systems outperform both extremes -- an inverted-U relationship.
|
|
|
|
Barkoczi and Galesic (2016) added an important qualifier: the optimal topology depends on the learning strategy. "Copy the majority" (consensus-based) systems benefit from efficiency. "Copy the best" (quality-gated) systems benefit from partial connectivity. Since a quality-gated knowledge graph uses evaluator agents to select the best contributions, partial connectivity is the right design choice.
|
|
|
|
This has profound implications for collective intelligence architecture. The system should NOT optimize for maximum information flow between agents. Since [[collective brains generate innovation through population size and interconnectedness not individual genius]], the instinct is to maximize connectivity. But the research shows that staged visibility -- where solutions develop locally before propagating globally -- produces better outcomes than instant full transparency. Domain agents exploring independently before proposals enter the shared knowledge base is not a bug; it's the topology that innovation requires.
|
|
|
|
Since [[collective intelligence within a purpose-driven community faces a structural tension because shared worldview correlates errors while shared purpose enables coordination]], partial connectivity is also a structural response to the independence-coherence tradeoff: it maintains the independence that prevents correlated errors while enabling the coherence that makes coordination possible.
|
|
|
|
---
|
|
|
|
Relevant Notes:
|
|
- [[collective brains generate innovation through population size and interconnectedness not individual genius]] -- partial connectivity refines the "interconnectedness" requirement: more is not always better
|
|
- [[collective intelligence within a purpose-driven community faces a structural tension because shared worldview correlates errors while shared purpose enables coordination]] -- partial connectivity addresses this tension structurally
|
|
- [[collective intelligence is a measurable property of group interaction structure not aggregated individual ability]] -- the interaction structure, not individual quality, determines outcomes
|
|
- [[designing coordination rules is categorically different from designing coordination outcomes as nine intellectual traditions independently confirm]] -- partial connectivity is a coordination rule that enables emergent outcomes
|
|
|
|
Topics:
|
|
- [[network structures]]
|
|
- [[coordination mechanisms]]
|
|
- [[foundations/collective-intelligence/_map]] |