Lesson 6 — Plasticity: The Brain That Rewires Itself

Understanding the Complex: How Does the Brain Actually Work?

---

In the 1990s, Eleanor Maguire at University College London began studying London taxi drivers. To obtain a license in London, drivers must pass "The Knowledge" — a notoriously demanding examination requiring them to memorize 25,000 streets within a six-mile radius of Charing Cross, plus thousands of points of interest and the optimal routes between them. Preparation takes on average four years.

Maguire used MRI to compare the brains of licensed cab drivers to non-driver controls. The finding was striking: taxi drivers had significantly larger posterior hippocampi — the region associated with spatial navigation and memory. The longer a driver had been licensed, the larger the difference. And in a follow-up study, when drivers retired, the structural difference diminished.

The brain had physically changed in response to sustained experience. That physical change wasn't metaphorical.

---

What plasticity means

Neuroplasticity is the brain's ability to reorganize itself — structurally and functionally — in response to experience, learning, injury, or environmental change. It operates at multiple levels.

At the synaptic level: long-term potentiation (LTP), described in Lesson 4, strengthens connections between co-active neurons. Long-term depression (LTD) weakens connections between neurons that fire out of sync. Over time, these microscopic adjustments in synaptic strength accumulate into something macroscopic: the circuit reconfigures.

At the structural level: synapses can form anew or be pruned away. Dendritic spines — the tiny protrusions on dendrites where synapses form — grow and retract on timescales of hours to days. In humans, hundreds of billions of synapses are pruned during adolescence as the brain refines its wiring based on lived experience.

At the regional level: as with the taxi drivers, sustained activity can change the relative size of brain regions — not by generating new neurons (with limited exceptions) but by expanding the dendritic arbors and synaptic density in active areas.

---

Learning as physical change

This reframing of learning — not as the acquisition of information stored somewhere, but as the physical reshaping of a network — has profound implications.

A skill practiced repeatedly isn't just "remembered." Its neural representation becomes more efficient, more distributed, more robust. Expert pianists show less cortical activation when playing than beginners — not because their brains are working less hard, but because the circuit has been optimized. The task that required effortful, widespread recruitment initially has been offloaded to a tight, automatized circuit.

This is why the 10,000-hour rule (however oversimplified) captures something real: sustained practice doesn't just improve performance, it changes the hardware. And it's why sleep matters — research by Jan Born and others has shown that memory consolidation predominantly happens during slow-wave sleep, when the hippocampus replays the day's experiences and transfers patterns to cortical long-term storage.

---

What AI borrowed — and what it didn't

The architecture of modern artificial neural networks was directly inspired by this neuroscience. Warren McCulloch and Walter Pitts formalized the neuron as a binary threshold unit in 1943. Frank Rosenblatt built the perceptron in 1957. The backpropagation algorithm — which underlies virtually all deep learning — adjusts connection weights to minimize error, directly mirroring Hebbian plasticity.

In this sense, the AI revolution was built on a neuroscience foundation. The key insight — that intelligence could emerge from adjusting the weights of a network of simple units — came from the brain.

But the analogy has significant limits.

The brain doesn't use backpropagation. Backprop requires knowing the "correct answer" in advance, computing an error signal, and propagating that signal backward through the network to adjust weights layer by layer. The brain has no such global error signal. It learns locally — through mechanisms like LTP and reward signals from dopamine — without any central supervisor.

The brain sleeps. Current AI systems don't need to. Sleep in biological brains serves memory consolidation, synaptic homeostasis (resetting overexcited synapses), metabolic restoration, and possibly cellular repair. These are constraints of biology, not intelligence per se — but they reveal how deeply the brain's operation is embedded in its physical substrate.

The brain forgets. Not as a bug but as a feature. Selective forgetting prevents older memories from interfering with new learning and frees up capacity. AI systems have no natural forgetting — they require explicit regularization techniques to prevent "catastrophic forgetting" when learning new tasks.

The brain has a body. It receives constant proprioceptive input from muscles, joints, and skin. Many cognitive functions — spatial reasoning, emotional regulation, even language — are shaped by embodiment in ways that disembodied AI systems completely lack. The philosopher Hubert Dreyfus argued in the 1970s that this was a fundamental barrier. The debate continues.

---

Limits of plasticity

Plasticity isn't unlimited. The brain has "critical periods" — windows of heightened plasticity in early life during which certain capacities (language, binocular vision) are most easily acquired. Miss the critical period for language (as happens in rare cases of extreme social deprivation) and some aspects of grammar never fully develop.

Adult plasticity is real — the taxi driver hippocampi are proof — but it's slower, more effortful, and more constrained than childhood plasticity. The idea that adults can simply "rewire" their brains at will through positive thinking is an overreading of the evidence.

Still, the evidence for adult neuroplasticity is substantial enough to change how we think about learning, rehabilitation after brain injury, and the cognitive effects of sustained practice across a lifetime.

---

Next lesson: Who Does What? Why? Who Pays? — The BRAIN Initiative, the Human Brain Project, Neuralink vs. academic BCIs.

---

Reading time: approx. 10–11 minutes