Liquid Brains

Cognitive networks have evolved a broad range of solutions to the problem of gathering, storing and responding to information. Some of these networks are describable as static sets of neurons linked in an adaptive web of connections. These are "solid" networks, with a well-defined and physically persistent architecture. But nature has also found a very different solution: systems formed by sets of agents that exchange, store and process information without persistent connections, moving relative to each other in physical space. We refer to these networks that lack stable connections and static elements as liquid brains, a category that includes ant and termite colonies, immune systems, some microbiomes and slime moulds.

What are the key differences between solid and liquid brains? Do liquid systems face intrinsic computational limits, or do they instead exploit their physical fluidity to solve problems that rigid architectures cannot? Despite lacking neural-like elements, many of these systems solve complex problems, exhibit learning and memory, and make collective decisions in response to environmental conditions. Ant colonies, for example, behave as excitable neural networks displaying collective synchronization phenomena. The immune system processes information through an idiotypic network that self-organizes into blocks of strongly and weakly influential nodes, performing self/non-self discrimination without any centralized controller. At a very different scale, social insects have been enormously successful ecosystem engineers, transforming the planet at massive scales — a parallel intelligence to our own, built on radically different principles.

Using a comparative approach grounded in statistical physics, network theory and computational models, our lab studies the generic properties of these distributed cognitive architectures. We have shown that the attractors found in liquid brains are not always based on connection weights, as in standard neural networks, but instead on population abundances. However, some liquid systems use fluctuations in ways remarkably similar to those found in cortical networks, suggesting a relevant role for criticality as a way of rapidly reacting to external signals. Phase transitions from zero to high collective activity emerge at well-defined thresholds, and the dynamics at these boundaries display properties that may be universal across cognitive systems.

Our programme also extends to language networks and other cognitive structures where heterogeneity and ambiguity play essential roles. We have discovered that language networks are typically scale-free, with most words having few connections while a handful act as hubs, and that this architecture is strongly tied to ambiguity — a feature that, far from being a defect, provides a source of language efficiency. In the study of language acquisition in children, we observed a sharp transition between a tree-like and a scale-free web around the two-year critical zone, suggesting that something is hardwired in the brain to enable the explosion of linguistic complexity.

A central goal of our research is to map the full space of cognitive architectures — small and large, distributed and centralized, alive and artificial — in order to establish the basis of a general theory of cognitive networks and explore the ultimate limits of both biological and artificial intelligence.