teleo-codex/domains/collective-intelligence/individual-generative-models-compose-via-interaction-structure.md

type	domain	description	confidence	source	created	depends_on	challenged_by
claim	collective-intelligence	Individual agent generative models combine into group-level models through the structure of agent interactions, not through aggregation or averaging of individual beliefs	likely	Kagan et al. (2025), 'As One and Many: Relating Individual and Emergent Group-Level Generative Models in Active Inference', Entropy 27(2), 143	2026-03-10

Individual agent generative models compose into group-level generative models through interaction structure, not aggregation

Kagan et al. (2025) provide formal treatment showing that when multiple active inference agents interact, their individual generative models do not merge through statistical aggregation or averaging. Instead, the structure of their interactions—the pattern of how agents influence and inform each other—determines how the group-level generative model emerges. This compositional process means the collective's beliefs are structurally emergent rather than statistically averaged.

This finding has important implications for understanding collective intelligence: the quality and nature of a group's collective beliefs depends on the interaction architecture rather than the sum of individual agent capabilities. A poorly-structured collective of capable agents will produce weaker collective inference than a well-structured collective of less capable agents.

Evidence

• source:kagan-2025 — Paper formally relates individual agent generative models to emergent group-level generative model, showing compositional mechanism through interaction structure rather than aggregation
• Implication: collective intelligence is determined by interaction topology, not individual agent quality alone

Challenges

• None identified in this source

---

Relevant Notes:

• collective-intelligence-is-a-measurable-property-of-group-interaction-structure-not-aggregated-individual-ability — directly supported by this finding

Topics:

• _map