Intelligence is usually measured by giving people tasks and studying the pattern among their scores. But the mind doing those tasks is not a collection of test scores. It is an enormous developing system of facts, procedures, concepts, strategies and cognitive processes, connected through years of learning. Network theories ask what happens if we take that interconnectedness seriously. Instead of assuming that correlations among cognitive performances must all trace back to one latent common cause, a network account allows cognitive components themselves to become related. Learning one thing can change how easily another thing is learned. A procedure can become connected to several concepts. Existing knowledge can create new routes for reasoning. One influential version, Wiring Intelligence, uses this idea to explore how the positive manifold and hierarchical cognitive structure might emerge from the development of connected cognitive networks.
Why use a network?
A simplified common-factor diagram looks like:
g → test A
g → test B
g → test C
The tests covary because each is related to the same latent dimension. A network account asks whether some of that structure can instead arise because cognitive components themselves are connected. Conceptually:
component A ↔ component B ↔ component C
The arrows here are illustrative, not a claim that every relationship is bidirectionally causal. The important shift is from one hidden source shared by many indicators to a system in which components can support, constrain, or become connected with one another.
How knowledge becomes wired together
Imagine a child first learns:
4 + 4 = 8
Initially it may be an isolated memorized fact. Later it becomes connected to:
- doubling;
2 × 4;8 ÷ 2;40 + 40;- even-number patterns;
- repeated addition.
Now the fact is part of a structured network. When new mathematical information arrives, existing connections can provide routes for understanding it. A learner with a richer relevant network may therefore acquire some later material differently from a learner whose knowledge remains sparse or weakly connected. This example illustrates the proposed mechanism. It does not demonstrate that all individual differences in intelligence are caused by knowledge-network connectivity.
What is Wiring Intelligence?
Savi and colleagues proposed a formal developmental network model in which an individual’s cognitive structure consists of interrelated facts and procedures that are acquired and connected over time. Their framework was designed to account for several phenomena, including:
- the positive manifold;
- hierarchical psychometric structure;
- developmental growth;
- cumulative or Matthew effects.
A notable feature is its idiographic ambition: rather than beginning only with differences among people, it attempts to represent an individual’s evolving cognitive network.
How can a network produce a positive manifold?
Suppose acquiring one useful piece of knowledge makes several later pieces easier to learn. As the network grows, successful acquisition can create additional opportunities for successful acquisition. Across individuals, differences in the development and connectedness of those networks can therefore produce broad performance differences. When researchers later administer many cognitive tests, those broad differences can appear as positive correlations. A general factor may then summarize the covariance even though, in the network theory, the factor is not the single cause generating every component.
How is network theory different from mutualism?
They are close relatives, not synonyms. Mutualism focuses on reciprocal developmental interactions among cognitive processes or abilities. Improvement in one can facilitate development in another. Network theories formalize cognition in terms of connected components and examine how network structure and growth can generate observed cognitive patterns. Mutualistic interactions can help create or strengthen a network. A network model can contain mutually reinforcing relations. But the explanatory objects and formal models are not identical. That difference is large enough to justify separate reference articles.
Does network theory show that g is an illusion?
No. A network system can generate data with a strong general factor. The disagreement is over what the factor means causally. A factor model may accurately summarize population covariance while a network model proposes a different mechanism for how that covariance arose. Therefore the relevant question is not:
factor or network—which one exists?
It is:
which model best represents the structure for a given purpose, and which causal account makes better discriminating predictions about how that structure develops and changes?
What evidence would strongly support a network explanation?
Merely fitting a network to correlations is not enough. Stronger evidence would show that:
- proposed nodes correspond to meaningful cognitive components;
- proposed connections can be estimated or manipulated;
- network development predicts later learning;
- longitudinal changes follow the theory’s developmental predictions;
- network accounts predict observations that plausible common-factor alternatives do not predict as well.
The challenge is therefore the same one faced by other causal theories of intelligence: generate evidence that distinguishes the mechanism from rivals that can reproduce the same broad empirical pattern.
The promise—and burden—of a network account
Network theories propose that intelligence may partly emerge from the organization and development of connected cognitive components rather than from one common cause acting independently on every test. They can coexist with the statistical existence of g while giving g a different causal interpretation. Their importance lies in connecting intelligence more directly with learning, development, and within-person cognitive organization. Their causal superiority over rival accounts remains an empirical question.
References
- Savi, A. O., Marsman, M., van der Maas, H. L. J., & Maris, G. K. J. (2019). The wiring of intelligence. Perspectives on Psychological Science, 14, 1034–1061. DOI
- van der Maas, H. L. J., et al. (2006). A dynamical model of general intelligence: The positive manifold of intelligence by mutualism. Psychological Review, 113, 842–861. DOI
- Schmank, C. J., Goring, S. A., Kovacs, K., & Conway, A. R. A. (2019). Psychometric network analysis of the Hungarian WAIS. Journal of Intelligence, 7, 21. DOI
