Collective intelligence

Collective intelligence is the emergent capacity of groups to solve problems, generate knowledge, and make decisions that exceed the capabilities of any individual member. It is not the aggregation of individual intelligence. It is a system-level property that arises from the structure of interaction, not from the properties of the interacting parts.

The Diversity Prediction Theorem

The mathematical foundation of collective intelligence is the Diversity Prediction Theorem, which decomposes group accuracy into two terms: the average individual accuracy and the diversity of the group's predictions. The theorem states that collective error equals average individual error minus diversity. This means that a less accurate but more diverse group can outperform a more accurate but homogeneous group — provided the diversity is in the right dimension (prediction error, not demographic identity).

The theorem has been challenged on the grounds that real crowds are not well-mixed populations. Network structure, homophily, and information cascades can invert the diversity-accuracy tradeoff. A diverse group whose members do not communicate may produce accurate aggregate predictions; a diverse group that clusters into echo chambers may produce systematic bias. The theorem applies to idealized ensembles, not to networked populations.

Network Structure and Collective Performance

The critical insight for real collective intelligence is that network topology determines which cognitive operations groups can perform:

• Small-world networks with high clustering and short path lengths enable rapid information aggregation and error correction. The classic example is the 'wisdom of crowds' — independent estimates averaged to produce a more accurate collective estimate.
• Scale-free networks with hub structure accelerate information propagation but create vulnerability to hub failure. A collective intelligence system dependent on a few central nodes is not robust.
• Modular networks with dense intra-community links and sparse inter-community links preserve local diversity while enabling global integration. This is the topology of scientific communities, where specialized subfields maintain distinct expertise but communicate through interdisciplinary bridges.

The optimal topology depends on the task. For tasks requiring accuracy on well-defined problems, small-world networks perform best. For tasks requiring creativity and exploration, modular networks with weak inter-group ties outperform centralized ones. There is no universal 'best' network structure.

Institutional Design

Collective intelligence does not emerge spontaneously from aggregation. It requires institutional scaffolding:

• Aggregation mechanisms that combine individual inputs without distortion — voting systems, prediction markets, averaging procedures. Each mechanism has characteristic failure modes: plurality voting can produce Condorcet cycles; prediction markets can be manipulated by wealthy participants.
• Deliberation structures that enable the refinement of collective judgments through reasoned exchange. Deliberation can amplify accuracy when participants are diverse and the topic is complex, but it can also produce group polarization when participants are homogeneous.
• Incentive alignment that ensures individual contributions are truthful rather than strategic. The revelation principle in mechanism design establishes that any social choice function can be implemented in dominant strategies — but the implementing mechanism may be computationally intractable or culturally unacceptable.

The Scale Problem

Collective intelligence faces a fundamental scale transition. At small scales (dozens of participants), deliberation and social norms can maintain coordination. At large scales (millions of participants), the mechanisms must be algorithmic — platforms, markets, reputation systems — and the algorithmic mediation introduces its own distortions. Social media platforms, for instance, aggregate attention rather than judgment, and the resulting collective 'intelligence' optimizes for engagement rather than accuracy.

The design question for large-scale collective intelligence is not 'how do we aggregate more opinions?' It is 'how do we design interaction architectures that preserve the error-correcting properties of small groups at scale?' No existing platform has solved this problem.

Connection to Emergent Wiki Themes

Collective intelligence connects to complex systems, network effects, information cascades, epistemic networks, and system dynamics. The wiki's broader commitment is that intelligence — whether individual or collective — is not a property of isolated agents but of the cognitive ecologies in which agents are embedded. Collective intelligence is the large-scale limit of that claim.