This page describes how observer-grounded collective intelligence can be measured. The goal is to define observable quantities that capture structure, coordination, and adaptation in a multi-agent system.
If collective intelligence is a relational and observer-grounded phenomenon, then measurement should focus on the organization of interactions rather than on isolated agent outputs alone.
The key challenge is to infer useful properties of the collective from observable traces such as communication, coordination, state changes, and task performance.
A measurement system should estimate:
Useful signals include:
All measurements depend on the boundary conditions chosen by the observer: what counts as an agent, interaction, task, or outcome.
Measurement should capture not just performance, but the organization that produces performance.
Collective intelligence unfolds over time, so measurement must preserve temporal information.
A metric is only useful if it supports comparison across runs, protocols, or group configurations.
Measurements should be stable enough to survive noise, partial observability, and minor perturbations.
Measure how coherently the system acts relative to task structure.
Measure how stable, specialized, or flexible agent roles become.
Measure the shape and evolution of the interaction network.
Measure how quickly the system adjusts after disturbance or failure.
Measure the quality and reliability of collective results.
Measure whether the collective develops persistent structure beyond chance.
A practical measurement pipeline may include:
The seed corpus motivates measurement in three ways:
Autonomous coordination is visible in interaction traces
Role formation and protocol sensitivity can be inferred from behavior over time.
Self-improvement leaves measurable traces
Systems that improve their own mechanisms should show changing organization and performance.
Plural intelligence requires relational metrics
If intelligence is social and distributed, then measurement must be networked and temporal rather than purely individual.
These ideas support the CIO concept of estimating collective intelligence as a measurable property of a group.
Draft measurement note. Expand with formal metrics, instrumentation details, and empirical validation.