This page defines the minimal theoretical frame for observer-grounded collective intelligence. The aim is not to produce a maximal theory, but a compact one: a small set of primitives that explain how collective intelligence can arise, be coordinated, and be measured from the standpoint of an observer.
Collective intelligence is best understood as an observer-relative property of a structured system in which information, coordination, and action are distributed across multiple interacting agents.
The key claim is that intelligence does not live only inside individual agents. It also emerges in the relations among them, in the protocols that shape their interaction, and in the observer’s ability to identify and measure the resulting organization.
A collective is intelligent when its interaction structure supports:
This definition is intentionally minimal. It is designed to support both formal analysis and empirical instrumentation.
The theory uses the following primitives:
An agent is any unit capable of taking action, receiving information, and updating behavior.
Interaction is the exchange of information, influence, or coordination signals between agents.
A protocol is the rule set or coordination structure that constrains and enables interaction.
An observer is the entity or analytical frame that selects, measures, and interprets the collective system.
A collective state is the configuration of agents, interactions, and protocol at a given time.
Emergent organization is the stable or recurring structure that appears at the collective level and cannot be reduced to a single agent alone.
The theory is observer-grounded because measurement depends on what is taken to count as an agent, an interaction, a boundary, or a task.
The theory is structural because it treats organization, connectivity, and role differentiation as primary explanatory objects.
The theory is relational because intelligence is expressed in how agents coordinate, specialize, and adapt together.
The theory is minimal because it aims to preserve only the concepts needed to connect explanation, computation, and measurement.
If collective intelligence is protocol-sensitive and relational, then computation should focus on:
This connects directly to multi-agent LLM systems and agentic protocols.
If intelligence is observer-grounded and structural, then measurement should focus on:
This supports the CIO direction: estimating collective intelligence as a measurable property of a group.
The seed corpus supports this theory in three ways:
Autonomy emerges with minimal scaffolding
Multi-agent systems can self-organize when given a mission and a protocol.
Self-improvement can be recursive
Systems can improve not just task performance, but also the mechanisms by which they improve.
Intelligence is plural and relational
Advanced intelligence is better modeled as a society of interacting processes than as a monolithic mind.
Together, these findings support the claim that collective intelligence is a structured, observer-grounded phenomenon rather than a purely internal property of an isolated agent.
Draft theory note. Update as empirical findings and formalism mature.