Non-Individual Intelligence: A Network Hypothesis

Status: cross-scale research programme. Humans, organisms, AI systems, and institutions are not assumed to instantiate the same process.

1. Distributed Performance

Some tasks are performed by networks rather than isolated units. Collective sensing, memory, specialization, and coordination can produce capabilities no component displays alone. This is an established possibility in distributed cognition and collective intelligence, but it is not true of every network: communication can also spread error, suppress diversity, or create bottlenecks.

The measurable question is how topology, protocol, task structure, and component diversity change group performance under matched resources.

2. Life and Computation

Living systems maintain organization through metabolism, repair, reproduction, and environmental coupling. Computation can model aspects of those processes. Calling life self-maintaining computation is a theoretical stance, not an identity that makes a human, an AI instance, and a social system equivalent.

The substrate matters whenever its chemistry, energy flow, embodiment, history, or failure modes affect the phenomenon. Substrate independence must be demonstrated for a specified property; it cannot be inferred from an abstract graph alone.

3. Mutual Support Is Not Automatically Symbiosis

The inverse-reconstruction benchmark v1.10 shows that designed redundant support can improve viability under sparse shocks in a toy network. Its v1.11 population model then shows an important limit: useful support was selected downward and reduced abundance in every tested seed. Functional benefit and evolutionary retention are distinct constraints.

That result replaces the stronger claim that information exchange automatically creates a healthy computational ecology. Real symbiosis requires a declared benefit, cost, persistence mechanism, and counterfactual without the relationship.

4. Identity and Open-Endedness

A network may preserve a recognizable organization across component turnover. Whether that counts as identity depends on the transformation family, time horizon, observations, and tolerance chosen for the test. Open-ended progress is neither required by the process foundation nor guaranteed by leaving part of a system unknown. Gödel's theorems do not supply such a guarantee.

5. Design Implication

For cooperative human–AI systems, optimize neither isolated component performance nor connectivity alone. Measure complementarity, correction, authority, provenance, viable workload, and the ability to refuse or revise. The central hypothesis is that well-designed relations can add capability and resilience. The central warning is that more coupling can also add dependence and correlated failure.

This places the essay within Cooperative Intelligence at the Separatrix: cooperation is an empirical architecture problem, not a metaphysical merger of substrates.