Talking about AGI without basic brain literacy is like discussing aviation without ever looking at a wing. Here is the minimum biology needed to argue intelligently about whether artificial systems can match biological ones.
Neurons and synapses
A neuron is an electrochemical cell that integrates inputs from thousands of other neurons and fires a spike when a threshold is crossed. The human brain has roughly 86 billion neurons and 100–500 trillion synapses (chemical junctions). Each spike is binary, but the rate, timing, and population pattern of spikes carry the signal.
Synapses are plastic: they strengthen and weaken based on local rules (Hebbian plasticity, spike-timing-dependent plasticity, neuromodulation). Most learning in the brain happens at synapses, not by adding new neurons.
Cortex: the thin outer sheet
The cerebral cortex is a folded sheet about 2–4 mm thick. Most of what we associate with human cognition — perception, language, planning, abstract thought — happens there. It is striking that the cortex is largely uniform in microstructure: roughly the same six-layered column repeated everywhere, specialised by its inputs and outputs rather than by radically different circuitry.
This uniformity is one motivation for general learning algorithms in AI: if one cortical circuit can support vision, language, and motor control depending on what it's wired to, perhaps one neural architecture can too.
Subcortex: older, faster, essential
Under the cortex sits a zoo of structures that are evolutionarily older and indispensable. The basal ganglia handle action selection and reinforcement learning. The thalamus routes information in and out of cortex. The hippocampus binds episodes into memory. The cerebellum tunes prediction and timing for motor and cognitive tasks alike.
Modern AI mostly emulates cortex. Many researchers think subcortical functions — particularly reinforcement learning and episodic memory — will need their own dedicated machinery in AGI systems.
Connectomics in 2026
Connectomics is the project of mapping every neuron and synapse in a brain. As of 2026, complete connectomes exist for C. elegans, the fruit fly larva and adult (FlyWire, 2024), and small cubic-millimetre patches of human and mouse cortex. Whole human or mouse connectomes are still years away.
What we have already learned: brains are far more recurrently connected than typical artificial networks, individual neurons are computationally rich, and the wiring contains regular motifs that may suggest reusable circuit templates.
Key terms
- Spike
- A brief electrical pulse — the basic unit of neural communication.
- Synapse
- The plastic chemical junction between two neurons where learning happens.
- Cortical column
- A small repeating vertical unit of cortex spanning all six layers.
- Basal ganglia
- Subcortical structure central to action selection and reinforcement learning.
- Connectome
- A complete wiring diagram of a nervous system.
Connects to AGI
The brain is not a transformer, and it is not obvious that AGI must look brain-like. But every honest comparison between artificial and biological intelligence has to start with the actual hardware of the latter — not a 1980s caricature.