A honey bee colony of 60,000 individuals builds honeycombs of mathematically perfect hexagonal geometry — the shape that, as Thomas Hales proved in 1999, maximises storage volume while minimising construction material. No bee knows geometry. An ant colony constructs underground cities with chambers, nurseries, ventilation shafts, and waste disposal systems spanning several metres beneath the earth. No ant has seen the blueprints. A swarm of foraging ants converges on the shortest path to a food source from among thousands of possible routes — solving, in effect, a version of the Travelling Salesman Problem that challenges supercomputers. No ant knows what an algorithm is.

Biologists call this swarm intelligence: the collective behaviour of decentralised, self-organised systems in which no individual has access to the global picture, yet the group as a whole produces coherent, adaptive, and strikingly intelligent outcomes. The mechanisms are well understood — pheromone trails that create positive feedback loops, evaporation that provides negative feedback, threshold-based role transitions, and competitive signalling that enables collective decision-making. The Belgian poet Maurice Maeterlinck once asked of the beehive: “What is it that governs here? What is it that issues orders, foresees the future, elaborates plans and preserves equilibrium?” The answer, worked out painstakingly by entomologists over the past century, is: nobody. The intelligence is not located in any individual. It emerges from the interactions.

Now I want to make a claim that may seem extravagant: your brain works the same way.

You have approximately 86 billion neurons. No single neuron knows who you are. No single neuron understands the sentence you are reading right now. No single neuron decides to move your hand, feel sadness, or recognise your mother’s face. And yet, from the interactions of these 86 billion cells — each one as cognitively limited as a single ant in a colony of millions — arises everything you call consciousness, thought, and selfhood.

The question that haunts neuroscience is structurally identical to the question that haunts entomology: Who is the planner? Where does the intelligence live? What governs here?

The answer is both unsettling and profound. Let us go looking for it.

The Neuron: An Ant in the Cortical Colony

A single neuron is, in computational terms, remarkably simple. It receives electrical and chemical signals from other neurons through its dendrites — branching input structures that can number in the thousands. If the accumulated signals cross a threshold, the neuron fires an electrical pulse — an action potential — down its axon, which branches out to connect with thousands of other neurons at junctions called synapses. At each synapse, the signal is converted into a chemical message (a neurotransmitter) that either excites or inhibits the next neuron.

That is essentially it. Receive signals. Integrate them. If the threshold is crossed, fire. If not, remain silent. This is not thinking. This is not understanding. This is a biological switch — more sophisticated than a transistor, but not by the orders of magnitude you might expect.

And yet, like a single ant depositing a pheromone-laced grain of sand, a single neuron’s action is never isolated. It fires into a network. Its signal becomes part of a pattern. That pattern becomes part of a larger pattern. And from these nested patterns — cascading through layers and regions and loops — cognition emerges.

The parallel to the insect colony is not merely poetic. Researchers at the Santa Fe Institute, studying neural decision-making in macaque monkeys, found that collective neural computation proceeds in two distinct phases — in a manner strikingly reminiscent of swarm behaviour. Early in the decision process, many neurons must be polled to predict the animal’s eventual choice: the “neural voice” is heterogeneous and distributed, like scout bees dispersing from a hive to evaluate potential nest sites. But as the moment of commitment approaches, the neurons converge: each individual neuron becomes maximally predictive of the decision. The system transitions from a crowdsourcing phase (information accumulation) to a consensus phase (commitment to action) — exactly the pattern observed in bee colonies selecting a new home, where dozens of scouts initially advertise competing locations through waggle dances, but eventually converge through competitive signalling until a quorum forms behind a single site.

As the researchers noted, this commonality suggests something fundamental: effective collective computation — whether performed by neurons, bees, or fish — involves the same basic architecture of distributed information gathering followed by convergent consensus.

The Cortex: A Sheet of 150,000 Colonies

If a neuron is like an individual ant, then what is the equivalent of the anthill — the structure that channels individual actions into collective intelligence?

The answer lies in the neocortex, the wrinkled outer layer of the brain, roughly 2 to 3 millimetres thick, folded into the iconic ridges and valleys of the cerebral surface. The neocortex is where perception, language, planning, abstract thought, and conscious experience are believed to reside. It contains somewhere between 18 and 26 billion neurons.

But here is the key structural insight: the cortex is not a homogeneous soup of neurons. It is organised into repeating modular units called cortical columns — vertical cylinders of neurons, each spanning the full thickness of the cortex, containing several thousand neurons arranged across six distinct layers. Vernon Mountcastle, who first described this organisation in the 1950s, proposed that the cortical column is the fundamental processing unit of the neocortex. Estimates suggest there are roughly 2 to 4 million cortical columns in the human brain.

Each column operates as a semi-independent processing unit — receiving inputs from the senses or from other columns, performing computations through the interactions of its neurons across layers, and sending outputs to other columns or to subcortical structures. The layered architecture is remarkably consistent across the entire cortex, regardless of whether the region processes vision, hearing, touch, language, or abstract planning.

This uniformity is the neural equivalent of a striking observation from entomology: every ant in a colony follows the same basic rule set, and every worker bee progresses through the same age-based career — cleaning cells as a newborn, nursing larvae by day four, building comb by day twelve, and foraging until death after day twenty-one. This system, called temporal polyethism, requires no supervisor assigning roles; each transition is triggered by hormonal changes and environmental feedback. In the cortex, the hardware is similarly modular and repeated. The software is local. And from this uniformity of parts, staggering diversity of function emerges — not because different columns are fundamentally different, but because they are connected differently.

Jeff Hawkins, in his “Thousand Brains” theory, takes this analogy further: he proposes that each cortical column independently builds a model of whatever object or concept it is processing, and that perception is not the product of a single hierarchical computation but the consensus of thousands of columns, each voting with its own model — much like scout bees vote for a nest site through competitive signalling, with the most supported option eventually winning.

No column is in charge. No column has the complete picture. Intelligence arises from the interaction — from the swarm.

Predictive Coding: The Brain as a Prediction Machine

If the cortex is a colony of columns, what is the equivalent of pheromone communication — the chemical signalling through which ants and bees turn local actions into global intelligence?

In ant colonies, this coordination happens through a mechanism called stigmergy: an ant deposits a pheromone-laced grain of soil, and that chemical trace stimulates other ants to deposit their own material at the same spot, creating a positive feedback loop that produces pillars, arches, and tunnels — none of which any individual ant planned. In bee colonies, returning foragers perform waggle dances that encode the direction and distance to food sources, recruiting idle workers to profitable sites through a competitive signalling system that efficiently allocates the colony’s workforce across a changing landscape.

Neuroscience’s most compelling answer to the coordination question is predictive coding, a theory that has gained enormous traction since Rajesh Rao and Dana Ballard’s seminal 1999 paper on the visual cortex. The core idea is elegant: the brain is not primarily a stimulus-response machine that passively receives information from the senses. Instead, the brain is fundamentally a prediction machine. It continuously generates expectations about what sensory signals it is about to receive — and then compares those predictions with what actually arrives.

The comparison yields a prediction error — the difference between what was expected and what was received. This error signal is the only thing that travels upward through the cortical hierarchy. If the prediction is accurate, there is little or no error to report, and neural activity is minimal. If the prediction is wrong — if something unexpected happens — a large error signal propagates upward, forcing higher levels of the hierarchy to update their models.

The Architecture of Prediction

The theory maps onto the physical structure of the cortex with remarkable specificity. Within each cortical area (and within each column), there are at least two functional populations of neurons:

Prediction neurons — principally located in the deeper cortical layers (layers V and VI) — send their signals downward and laterally. These neurons encode what the brain expects to happen. Their connections flow from higher cortical areas to lower ones — from regions that process abstract concepts down to regions that process raw sensory data.

Error neurons — principally located in the superficial cortical layers (layers II and III) — compute the mismatch between the prediction and the actual input. These error signals flow upward, from lower cortical areas to higher ones, carrying the news: “Your model was wrong. Here is how.”

This creates a continuous, bidirectional flow of information: predictions cascade downward, prediction errors cascade upward, and at every level, the system adjusts to minimise the discrepancy. The process is iterative, rapid, and massively parallel — running simultaneously across millions of columns.

Notice the structural parallel to the insect colony. In a beehive, returning foragers carry information upward through waggle dances that report on food sources, while the colony’s existing state pushes information downward — “stop signals” (head-butts delivered to dancing foragers) inhibit recruitment when processing capacity is overwhelmed, and a chemical called ethyl oleate, produced in foragers’ stomachs and transferred through food-sharing, inhibits younger bees from transitioning prematurely to foraging. In ant colonies, pheromone trails carry positive feedback about successful paths while pheromone evaporation provides negative feedback that erases outdated information. In both cases, the system maintains a dynamic equilibrium between what it expects (based on accumulated experience) and what the environment actually delivers.

Predictive coding is the brain’s version of stigmergy — indirect coordination through a shared medium. Instead of pheromone trails laid in the environment, neurons lay patterns of activation across synaptic connections, and these patterns guide the behaviour of downstream neurons, who in turn modify the patterns for their own downstream neighbours.

The Free Energy Principle: Why Prediction Is Everything

If predictive coding is the mechanism, then Karl Friston’s Free Energy Principle is the reason the mechanism exists.

Friston, a neuroscientist at University College London, proposed in the mid-2000s what he considers a universal principle governing all biological systems — from single cells to entire organisms: every living system acts to minimise its variational free energy, which is, roughly speaking, a mathematical measure of the difference between what the system predicts about the world and what the world actually delivers. In plainer terms: all life is in the business of minimising surprise.

This is not “surprise” in the colloquial sense of birthday parties and plot twists. In information theory, surprise (or “surprisal”) is a precise quantity: it measures how improbable an observed event is, given the system’s model of the world. A fish finding itself in water has low surprise. A fish finding itself on a kitchen counter has very high surprise. Living systems that consistently find themselves in high-surprise states do not remain living for long.

The Two Strategies for Reducing Surprise

The Free Energy Principle identifies two fundamental strategies that any organism can use to minimise surprise:

1. Update the model (Perception). If the world delivers something unexpected, revise your internal model to better account for reality. This is perception and learning — you change your beliefs to fit the evidence. In predictive coding terms, this means adjusting the predictions sent downward through the cortical hierarchy until the prediction errors are minimised.

2. Change the world (Action). Instead of updating your beliefs about the world, act on the world to make it conform to your predictions. If you expect to see your hand touching an object, and your hand is not yet touching the object, you can resolve the prediction error by moving your hand. Action, in this framework, is not fundamentally different from perception. Both are strategies for minimising the gap between expectation and reality.

This is a revolutionary reframing. Perception and action — which classical neuroscience treated as separate systems (sensory processing vs. motor control) — are revealed to be two sides of the same coin: two complementary strategies for keeping the organism in states of low surprise, which are, by definition, the states compatible with the organism’s continued existence.

The Organism as a Colony Minimising Surprise

Now notice the deep parallel to the insect colony. When a bee colony allocates foragers to a rich nectar source, it is — in Friston’s terms — minimising the free energy of the colony’s resource model. The waggle dance is a prediction (“there is abundant nectar at this location”); the forager’s return confirms or disconfirms the prediction; the vigour of subsequent dancing is adjusted to minimise the error. When the source is exhausted and the prediction fails, error signals (reduced dancing, stop signals) propagate through the colony, and the model is updated.

When an ant colony converges on the shortest path to a food source, the pheromone trail system is performing a collective version of free energy minimisation. Each ant’s pheromone deposit is a prediction about the quality of the path. The evaporation of pheromone on longer paths is the equivalent of prediction error correction — outdated models fade, and the colony’s “belief” converges on the path that best fits reality.

In a 2023 experiment published in Nature Communications, researchers provided the first direct experimental validation of the Free Energy Principle using living neurons. In vitro networks of rat cortical neurons, when stimulated with signals generated from two hidden sources, self-organised to selectively encode those sources. Changes in synaptic connectivity reduced variational free energy exactly as the theory predicted. The neurons were not instructed. They were not programmed. They simply self-organised — like ants building an anthill, like bees constructing a comb — to minimise the discrepancy between their internal model and the signals arriving from outside.

The Cortex Decides: But How?

We now arrive at the central question your experience poses every moment of every day: how does the brain decide? When you choose tea over coffee, when you swerve to avoid a pedestrian, when you weigh the merits of two job offers — what is happening in the cortical colony?

No Executive Sitting in the Prefrontal Cortex

The intuitive model — and the one that dominates popular culture — is that there is a command centre, typically identified as the prefrontal cortex, that weighs the evidence, deliberates, and issues orders. The prefrontal cortex as the CEO of the brain.

But this model has a fatal flaw, recognised by philosophers since at least Aristotle: it leads to an infinite regress. If the prefrontal cortex makes the decisions, then who decides what the prefrontal cortex should do? If there is a little homunculus inside your head watching the screen and pulling the levers, who is inside the homunculus’s head?

Neuroscience increasingly supports a different picture — one that looks, once again, like a colony. Research on distributed executive control, published in work from the Santa Fe Institute and elsewhere, argues for control without controllers. The brain’s executive functions — the capacity to inhibit inappropriate responses, switch between tasks, and maintain goals — emerge from the interactions among distributed neural populations, not from the dictates of a localised command module.

The parallels to the insect world are striking. Neurons, like ants, have limited views of the whole system’s activity. They communicate primarily with their neighbours through local connections. They produce chemical outputs (neurotransmitters) that modulate the responses of downstream neurons — just as ants produce pheromones that modulate the behaviour of nearby colony members. These local interactions can generate emergent control signals without the need for a dedicated controller.

In the cortex, this works through competitive dynamics. When you are deciding between tea and coffee, separate neural populations encode the evidence for each option. These populations compete — each one accumulating evidence for its favoured outcome, while simultaneously inhibiting the other through lateral inhibition (the neural equivalent of stop signals in the beehive). The population that first reaches a critical threshold of activation “wins,” and its output cascades downstream to motor systems that execute the choice.

No single neuron decided. No executive committee voted. The decision emerged from competitive dynamics among populations — exactly as a bee colony’s choice of a new nest site emerges from competitive dancing among scouts. When a bee colony must relocate, individual scouts discover candidate sites and return to advertise them through waggle dances. Scouts for different sites effectively compete: each tries to recruit uncommitted bees to her preferred location. The site with the most vigorous and persistent advertising eventually achieves a quorum of supporters, and the colony moves — a decision made by nobody, yet made well.

The Hierarchy of Time: Nested Predictions, Nested Colonies

One of the most beautiful aspects of the brain’s predictive architecture is that it operates across multiple timescales simultaneously — and this, too, mirrors the nested hierarchies of the insect colony.

At the lowest levels of the cortical hierarchy — primary sensory areas — predictions operate on timescales of milliseconds. The visual cortex predicts what the next instant of visual input will look like, based on the pattern of light that arrived a fraction of a second ago. Prediction errors at this level are the stuff of basic perception: edges, colours, motions.

At intermediate levels — association areas — predictions operate on timescales of seconds to minutes. These regions predict sequences of events: if I see a hand reaching for a cup, I predict the cup will be lifted. If the hand instead knocks the cup over, a large prediction error propagates upward.

At the highest levels — the prefrontal and anterior temporal cortex — predictions operate on timescales of days, months, and years. These are the regions that encode your life narrative, your long-term goals, your model of who you are and what the world is. When you plan your career, you are generating predictions about what your life will be like years from now, and adjusting your actions to minimise the discrepancy between your desired future and your expected future.

This hierarchical nesting of timescales has a striking parallel in social insect colonies. A worker bee’s moment-to-moment behaviour — depositing wax, following a waggle dance, fanning the hive — operates on a timescale of seconds. Her progression through the colony’s division of labour — from cell cleaner to nurse to comb builder to forager — unfolds over weeks, governed by hormonal shifts and chemical inhibitors like ethyl oleate. The colony’s seasonal rhythms — building honey stores for winter, swarming to reproduce in spring — operate on a timescale of months, governed by photoperiod, temperature, and resource availability. At every level, simple local rules interact to produce adaptive behaviour at the next level up — with no single level “understanding” the purposes of the level above it.

The Deeper Unity: What Bees and Brains Share

We can now articulate the structural parallels between the bee colony and the brain with some precision:

These are not loose metaphors. They are structural homologies — independent systems that have converged on the same computational architecture because they face the same fundamental challenge: how to produce coherent, adaptive, intelligent behaviour from the interactions of many simple agents, none of whom has access to the global picture.

The Free Energy Principle provides the deepest unification. Both the bee colony and the brain are systems that exist far from thermodynamic equilibrium — systems that must continuously work to maintain their organisation against the entropy of the universe. Both do this by building and maintaining predictive models of their environment and acting to minimise the discrepancy between prediction and reality. The bee colony’s foraging system is a predictive model of the landscape’s nectar distribution. The brain’s cortical hierarchy is a predictive model of the causal structure of the world. Both systems update their models when predictions fail (learning) and act on the world to make predictions come true (action). Both are, in Friston’s terminology, engaged in active inference — the continuous cycle of predicting, acting, sensing, and updating that keeps a self-organising system alive.

The Mind as an Emergent Swarm

What, then, is the mind?

If the invisible architect of the beehive is not a who but a how — a system of feedback loops, chemical signals, and evolutionary tuning — then the same must be said of the mind. Your experience of being a unified, conscious agent is not the product of a central command module that watches and decides. It is the emergent property of 86 billion neurons engaged in a continuous swarm of prediction and error correction, competition and consensus, local action and global coherence.

This is not a diminishment. The honeycomb is not diminished by the fact that no bee understands hexagonal geometry. The anthill is not diminished by the fact that no ant has seen the blueprints. If anything, it is more remarkable — more worthy of what Darwin called “enthusiastic admiration” — that such complexity can arise from such simplicity; that the cathedral can be built without a cathedral builder; that the symphony can be performed without a conductor.

The Indian philosophical tradition has a name for this insight. The Bhagavad Gita speaks of action without attachment to the fruits of action — nishkama karma. Each bee cleans its cell, nurses its larvae, builds its comb, and forages its flowers without any conception of the colony’s grand strategy. Each neuron fires or stays silent, strengthens or weakens its connections, without any conception of the thought it is helping to produce. The intelligence lies not in the individual’s intention but in the system’s design.

And the system was designed by nobody. It was designed by everything — by the slow, patient, relentless pressure of natural selection acting on feedback loops, on threshold sensitivities, on the timing of chemical signals, over hundreds of millions of years.

The hive inside your skull is the most complex object in the known universe. And like every great hive, it runs itself.