Exploring the computational power of the human brain

Brain Computing

Introduction: the enigma of brain computing

Welcome to the interactive exploration of the computational power of the human brain. This application sets out to reveal the extraordinary complexity of this organ, going beyond the simple capacity to compute and embracing the whole range of cognitive faculties. Understanding that power means deciphering the biological mechanisms behind thought, perception and action, and asking what intelligence and consciousness really are.

Unlike digital computers, the power of the brain is not measured in conventional metrics such as FLOPS. It shows in how effectively and adaptably it processes information to deal with a changing world. Even everyday tasks that look simple call for computations of remarkable sophistication. Through the sections we will move from the basic building blocks of the brain, neurons and glial cells, to how it works, to higher cognition, to the challenge of measuring it and the comparison with artificial intelligence.

What you will discover:

The biological parts that make computing possible.
The fundamental principles, such as parallelism and neuroplasticity.
How the brain handles memory, learning and the integration of information.
The challenges of measuring its power, and how it differs from AI.
The frontiers of research, from neuromorphic computing to the philosophical questions.

The building blocks of the mind

The extraordinary computational capacity of the human brain is rooted in the complex organisation and the dynamic working of its biological parts. This section looks at the fundamental cellular units, the specialised neural architectures and the large-scale networks that together make up the physical ground of thought and cognition. Understanding these “building blocks” is the first step to appreciating the computing power of the brain.

The cellular units that compute

At the most basic level, brain computing is orchestrated by specialised cells and their parts. Not only neurons: glial cells and sub-cellular structures such as dendrites play crucial roles too.

Neurons and synapses ▸

Neurons generate and carry electrical signals (action potentials). They receive, integrate and pass on information. There are sensory neurons, motor neurons and interneurons.

Synapses are the connections between neurons where information is passed on, often through neurotransmitters. Synaptic strength is plastic, which is fundamental to learning and memory.

Glial cells ▸

Once thought of as mere support, glial cells (astrocytes, oligodendrocytes, microglia) take an active part in computing.

Astrocytes: they modulate the synaptic environment, take part in transmission and synchronise network activity (forming the “tripartite synapse”).

Oligodendrocytes: they form myelin, speeding up the transmission of signals and giving metabolic support.

Microglia: the immune agents, involved in reshaping synapses (*pruning*).

Dendritic computing ▸

Dendrites, the branches that receive input, are not passive conductors but compartments doing active processing. They can generate local spikes (sodium, calcium, NMDA) and integrate information in non-linear ways.

That raises the computational power of a single neuron, and its memory capacity, a good deal.

The architecture of thought

The brain has specialised areas with distinct but connected roles. How those areas are organised into networks is fundamental to higher cognition. The architecture of the brain grows over time, shaped by experience.

The cerebral cortex

Frontal lobe

Parietal lobe

Temporal lobe

Occipital lobe

Hippocampus

Cerebellum

Basal ganglia

Select a structure of the brain to see the details.

Large-scale brain networks and neurovascular coupling

Beyond what each region does on its own, the brain works through large-scale networks (the default mode network, for instance) that bring the activity of many areas together. These networks have topological features such as hubs (regions that are highly connected), crucial to integrating information.

Intense computing needs a constant supply of energy, ruled by neurovascular coupling (NVC). This mechanism fits the flow of blood in the brain to how active the neurons are, showing the critical link between energy and computing. Recent findings point to direct, “synaptic” signalling between axons and arterioles for that regulation.

How it works: the fundamental principles of computing

The brain works on principles that set it clearly apart from conventional computers. Its ability to process information efficiently, to adapt to experience and to keep working comes from built-in features: massive parallelism, distributed processing, deep plasticity and sophisticated neural coding. This section looks at these “operating systems” of the brain.

Massive parallelism and distributed processing ▸

The brain processes information concurrently, with billions of neurons working at the same time (massive parallelism). That gives it high speed and responsiveness.

Information is carried by patterns of activation spread over large sets of neurons. That makes it robust, tolerant of failure (it degrades gradually) and accurate.

A simplified picture:

Serial processing (a traditional PC):

Parallel processing (the brain):

Neuroplasticity: the brain that adapts ▸

Neuroplasticity is the ability of the brain to change its structure and its function in response to experience, learning or damage. It never stops, and it is built in.

The mechanisms include:

Synaptic plasticity: changes in the strength of the connections (LTP, LTD).
Structural plasticity: dendrites growing and being reshaped, synaptogenesis, synaptic pruning, neurogenesis (limited).
Functional plasticity: functions being reorganised, compensatory strategies.

That constant tuning of hardware and software is a mark of biological computing.

Neural coding: the language of the brain ▸

Information is represented and carried by patterns of neural activity (spikes).

Rate coding: the information is in the average rate of firing.
Temporal coding: the information is in the precise timing of the spikes. It can carry more information.
Population coding: the information is in the coordinated activity of a group of neurons. It brings robustness and flexibility.

Efficient coding is crucial to fast processing, to decisions and to complex behaviour.

Higher cognitive computing

Beyond the basic principles of processing, the brain carries out cognitive work of a higher order, the ground of what we do best. That includes memory, with its complex encoding and recall, the ability to bring scattered information into one coherent whole (the “binding problem”) and the behavioural flexibility finely tuned by neuromodulation. This section goes deeper into these sophisticated sides of brain computing.

Memory: a computational view ▸

Memory is a dynamic computing system: encoding, consolidation, storage, recall.

Key roles:

Hippocampus: quick formation of episodic and spatial memories, and their conversion into long-term ones.
Neocortex: the long-term store.
Systems consolidation: the hippocampus “teaches” the neocortex, making memories independent of it (a slow process, with hippocampal *replay* during sleep or quiet waking).
Prefrontal cortex (PFC): fitting new memories into existing patterns, and recall in context.
Amygdala: it governs how long emotional memories last.

The brain fits new information into existing patterns, avoiding the “catastrophic forgetting” typical of AI.

How a memory is consolidated (simplified):

First experience: the hippocampus encodes the new information quickly (short-term memory).
Hippocampal replay: during sleep or quiet waking, the neural patterns of the experience are reactivated in the hippocampus.
Hippocampus and neocortex together: the replay is coordinated with activity in the neocortex, guided by the PFC.
Transfer and integration: the information is gradually moved into the networks of the neocortex and fitted there (long-term memory).
Independence from the hippocampus: in time, a consolidated memory depends less on the hippocampus to be recalled.

The binding problem ▸

How the brain brings features processed separately (colour, shape, movement) into one conscious experience (a red apple rolling, for example).

The main hypotheses:

Binding through synchrony (BBS): neurons that process features of the same object fire in step (gamma oscillations).
Attention: it selects which features get bound.
Top-down feedback: signals from higher areas, there to integrate.

Solving binding is crucial to coherent episodic memories and to a deeper understanding of consciousness.

Neuromodulation: tuning the computing ▸

Nerve activity is regulated by neuromodulators (dopamine, serotonin, acetylcholine, noradrenaline). They work widely (“volume transmission”), with slower, longer-lasting effects, changing the properties of cells and synapses.

What they do:

They change the rates of learning and of plasticity (dopamine in reinforcement learning, for example).
They change how excitable neurons are and their gain (noradrenaline for a “network reset”, for example).
They shift the state of a network (the sleep-wake cycle, attention).
They affect the signal-to-noise ratio.

It works as a system of meta-control, for flexibility and adaptability.

Measuring the unmeasurable? Estimating computing capacity

Putting a number on the computing power of the human brain is a formidable challenge. The methods we have only open partial windows on what it does, and direct comparisons with digital computers often turn out to be beside the point. This section looks at the techniques of measurement, their limits, the trap of metrics such as FLOPS, and at more meaningful alternatives such as energy efficiency and storage capacity.

Methods and their limits

fMRI (functional magnetic resonance imaging)

It measures blood flow (the BOLD signal), an indirect sign of neural activity. Good spatial resolution (~1 mm³, ~60k neurons), poor temporal resolution (~1-2 s). It does not catch fast dynamics or the activity of single neurons.

EEG (electroencephalography)

It records electrical activity through electrodes on the scalp. Excellent temporal resolution (ms), poor spatial resolution. Mostly the activity of the surface of the cortex.

These instruments tell you *where* and *when* something happens, but not necessarily *which* computations or *what* information. They measure correlates, not the algorithms being run.

The trap of FLOPS, and other metrics

Comparing the brain with computers through FLOPS (floating point operations per second) is misleading. The brain does not work that way: it is massively parallel, hybrid analogue and digital, with memory built into the processing.

Key numbers estimated for the brain:

Neurons: ~86 billion
Synapses per neuron: ~7,000 (average)
Total synapses (cortex): ~125 trillion
Storage capacity: estimates from 10-100 TB up to 2.5 PB (petabytes)
TEPS (traversed edges per second): $0.18 \times 10^{14}$ to $6.4 \times 10^{14}$

These are estimates, and they can vary a lot from one source to another.

The energy efficiency of the brain is astonishing: it uses about 20 watts, while doing computing that would take megawatts of supercomputer. That efficiency comes from a parallel architecture, from memory sitting with the processing and from event-driven work. It is a benchmark for sustainable AI.

Comparing biological and digital computing

Feature	Digital computer	Human brain
Architecture	Mostly von Neumann (CPU and memory apart), serial.	Massively parallel, distributed neural networks.
Kind of processing	Mostly serial and digital.	Parallel; hybrid analogue (graded potentials, dendritic integration) and digital (neural spikes).
Memory	Separate from the processing unit (RAM, storage).	Built into the processing (synapses are both the store and the computing).
Energy efficiency	Generally low; supercomputers take megawatts.	Extremely high; about 20 watts.
Plasticity and adaptability	Limited; it takes explicit reprogramming.	High, built-in plasticity (synaptic, structural, functional) for constant learning and adaptation.
How information is coded	Binary (bits).	Neural coding (rate, temporal, population).
Handling errors and failures	Error correction codes, planned redundancy. It can be fragile.	Gradual degradation, high redundancy, compensating plasticity.
The usual metrics of power	FLOPS, IPS.	No single metric everyone accepts. TEPS, storage capacity (TB/PB), energy efficiency and functional abilities are all considered.

The human brain against artificial intelligence

Comparing the computing of the human brain with that of artificial intelligence (AI) shows deep differences. While AI has made extraordinary progress on specific tasks, the brain keeps unique advantages in flexibility, in learning from very few examples, in causal reasoning and in robustness. This section looks at those key differences.

Comparing cognitive abilities

Cognitive ability	Human brain	AI systems today
How few examples it needs (few-shot learning)	High, based on patterns and abstraction.	Low without pre-training or special methods; it needs a lot of data.
Unsupervised learning and finding structure	Built in, guided by experience.	Improving, often statistical; it needs validation.
Continual and lifelong learning	Robust, fitted into existing patterns.	Prone to catastrophic forgetting.
Catastrophic forgetting	Minimal, or a gradual decline.	A real problem.
Transfer learning	High, based on principles and analogy.	Limited, it struggles out of its distribution; often correlational.
Causal reasoning	Strong, based on theory and intervention.	Mostly correlational, without a deep grasp of the “why”.
Multi-step logical reasoning	Capable, though open to bias.	Fragile as complexity grows; LLMs can fail.
Understanding context and common sense	Deep, intuitive, based on models of the world.	Lacks deep common sense; limited by the training context.
Abstract thought and genuinely new ideas	High: divergent thinking, theories of its own.	Recombining learned patterns; the debate over “true” novelty is open.
Robustness to noise and ambiguity	High, it filters well.	Varies, it can be fragile; tolerance of noise is a goal.
Tolerance of failure and gradual decline	High, thanks to redundancy and plasticity.	Can be brittle; a single failure can be catastrophic.

Learning: humans are excellent at *few-shot* and continual learning with no “catastrophic forgetting”, using abstraction and causal understanding. AI often needs huge datasets and struggles with forgetting and with out-of-distribution (OOD) generalisation.

Reasoning: humans use causal logic and models of the world. AI leans on statistical correlation and struggles with multi-step inference and deep common sense.

Creativity: human creativity means divergent thinking and genuinely new things. AI makes combinations out of training data; whether that is “true” originality is open to debate.

Robustness: the brain degrades gradually. AI can be fragile when the data is unexpected or when it is attacked on purpose.

Frontiers and philosophical horizons

Exploring the computational power of the brain opens exciting frontiers of research and raises fundamental philosophical questions. From neuromorphic computing, which tries to imitate the brain, to the challenge of understanding consciousness and the algorithms of the brain we still do not know, to what advanced AI means for ethics, this section looks at the edges of what we know and at what comes next.

Neuromorphic computing and AGI ▸

Neuromorphic computing aims at AI systems inspired by the architecture and the principles of the brain (spiking neural networks, event-driven processing, memory sitting with the computing, parallelism, learning on the chip).

The potential: energy efficiency, speed, adapting in real time, processing sparse data. Many see it as a step towards Artificial General Intelligence (AGI) — AI with the cognitive abilities of a human.

Consciousness and the algorithms we do not know ▸

What subjective experience is like (*qualia*) is the “hard problem” of consciousness. Theories such as IIT (integrated information) and GNWT (global neuronal workspace) try to explain it, but face empirical hurdles.

Many of the algorithms of the brain for higher cognition (abstraction, deep causal reasoning) are unknown. Predictive coding (the brain as a prediction machine) is a promising framework.

Computational psychiatry ▸

It reframes mental illness in terms of specific computational dysfunctions, using models to link the brain and the symptoms.

The aims: better understanding, diagnosis and personalised treatment (changes in reward processing in depression, for example).

Philosophical reflections ▸

The computational theory of mind (CTM) — the mind as a system that processes information — has radical implications, but faces criticism (Searle’s Chinese room, Penrose’s arguments, the problem of qualia).

The ethical and social side of AI: responsibility, bias, privacy, alignment with human values, the effect on work and on human cognition, the effect on the environment.

Philosophy is crucial to these fundamental questions about who we are in the age of AI.

Interactive glossary

This section gives the meaning of some key terms used in the treatise and in this interactive exploration. Use the search bar to find a term quickly, or scroll through the list.

Neurovascular coupling (NVC)

The mechanism that fits the supply of blood (and so of energy) to the regions of the brain to how active their neurons are.

AGI (artificial general intelligence)

A hypothetical kind of AI that matches or beats human cognitive ability on virtually every task.

Few-shot learning

The ability to learn new concepts from a very small number of examples.

Astrocytes

A kind of glial cell that modulates the synaptic environment, takes part in transmission and synchronises network activity.

Binding problem

The question of how the brain brings features processed separately (colour, shape, movement) into one perception.

Glial cells

Cells of the nervous system that are not neurons, with supporting, metabolic, immune and modulating roles. They include astrocytes, oligodendrocytes and microglia.

Cerebellum

A structure of the brain involved mostly in motor learning, coordination and balance, and in some cognitive functions too.

Neural coding

How information is represented in the nervous system through patterns of neural activity (rate, temporal and population coding, for example).

Dendritic computing

The active processing of information that happens inside the dendrites of a neuron, beyond simply adding up the inputs.

Cerebral cortex

The outer layer of the brain, responsible for higher cognition such as language, reasoning and memory.

Catastrophic forgetting

A phenomenon in artificial neural networks where learning new information tends to overwrite and erase what was learned before.

EEG (electroencephalography)

A technique that measures the electrical activity of the brain through electrodes on the scalp.

Distributed processing

The principle that information is represented and processed by patterns of activation spread over large sets of neurons, not in single places.

FLOPS (floating point operations per second)

A measure of the computing power of digital computers, generally unsuitable for the brain.

fMRI (functional magnetic resonance imaging)

A neuroimaging technique that measures brain activity indirectly, by picking up changes in blood flow.

Basal ganglia

A group of subcortical nuclei involved in choosing actions, in reinforcement learning and in forming habits.

Neuromorphic computing

A computing approach that designs AI systems inspired by the architecture and the processing of the biological brain.

Hippocampus

A structure of the brain crucial to forming new memories (episodic and spatial) and to consolidating them.

LTD (long-term depression)

A lasting weakening of synaptic strength. Important in refining neural circuits.

LTP (long-term potentiation)

A lasting increase in synaptic strength that follows high-frequency stimulation. Considered a cellular mechanism of memory.

Microglia

Immune cells that live in the brain, involved in clearing debris and in reshaping synapses.

World models

Internal representations that a system (biological or artificial) builds to understand and predict what goes on around it.

Neuromodulation

The regulation of nerve activity by neuromodulators (dopamine, serotonin, for example) that change the properties of cells or synapses.

Neuroplasticity

The ability of the brain to change its own structure and function in response to experience, learning or injury.

Neurotransmitter

A chemical released by a neuron at a synapse to carry a signal to another neuron or cell.

Oligodendrocytes

A kind of glial cell that forms the myelin sheath around axons in the central nervous system, speeding up the signal.

Massive parallelism

The ability of the brain to process much information at once through the concurrent work of billions of neurons.

Synaptic plasticity

The ability of synapses to get stronger or weaker over time, as their activity rises or falls. Fundamental to learning and memory.

Action potential

A short electrical impulse that travels along the axon of a neuron, the basic unit of neural communication.

Predictive coding

A framework in which the brain constantly predicts what it will sense and updates its internal models on the prediction error.

Computational psychiatry

A field that uses computational models to understand and treat mental disorders in terms of dysfunctions in how information is processed.

Spiking neural networks (SNNs)

Artificial neural networks that use spikes (discrete pulses) to communicate and to compute, imitating action potentials.

Synapse

The junction between two neurons (or between a neuron and a cell it drives) where the signal is passed on, usually through neurotransmitters.

Computational theory of mind (CTM)

A philosophical theory holding that the mind is a system that processes information, and that cognition and consciousness are forms of computing.

Integrated information theory (IIT)

A theory of consciousness that has it emerge from the integration of information inside a complex system.

Global neuronal workspace theory (GNWT)

A theory of consciousness suggesting that it arises when important information is broadcast widely across a global network of the brain.

TEPS (traversed edges per second)

A measure of how fast a graph is traversed, put forward as an alternative metric for complex systems such as the brain.