# What did Donald Hebb's 1949 book say about how the brain learns?

Date: 2019-12-02 · Author: Robert Scales · https://guidetoartificialintelligence.com/en/learn/what-is-hebbs-rule-organization-of-behavior/

**Answer:** In The Organization of Behavior (1949), the Canadian psychologist Donald Hebb proposed that when one brain cell repeatedly helps to fire another, the link between them grows stronger, and that groups of cells strengthened this way, which he called cell assemblies, are the building blocks of perception and thought. That idea, known as Hebb's rule, became one of the founding ideas of neural network research.

How does a brain remember anything? In 1949 a psychologist at McGill University in Montreal offered an answer simple enough to state in one sentence and rich enough to keep researchers busy for seventy years. Donald Hebb suggested that when one nerve cell keeps helping another to fire, something physical changes so that the first cell gets better at firing the second. Learning, in this picture, is the gradual strengthening of useful connections.

He set out the idea in a book called *The Organization of Behavior: A Neuropsychological Theory*, published in 1949 by John Wiley & Sons in New York. The preface is signed "Montreal, April 1949." Today the idea is usually called Hebb's rule or Hebbian learning, and anyone who studies neural networks meets it early.

## Who Donald Hebb was

Donald Olding Hebb was born in 1904 in Chester, Nova Scotia. He earned a master's degree in psychology at McGill in 1932 and a doctorate at Harvard in 1936. From 1937 to 1939 he worked with the neurosurgeon Wilder Penfield in Montreal, where he saw that patients could lose large parts of the brain without losing much measured intelligence. That puzzle stayed with him.

In the 1940s, many psychologists treated behavior as a matter of stimulus and response, with little interest in what happened inside the head. Hebb wanted a theory that connected what physiologists knew about neurons with what psychologists saw in the laboratory and the clinic. His preface thanks students in seminars at Harvard in the summer of 1947 and at McGill the following winter, and the colleagues he argued with each week, for five years, at the Yerkes primate laboratories.

## The rule in Hebb's own words

The key passage sits in the fourth chapter, under the heading "A Neurophysiological Postulate." A postulate is an assumption you make in order to build a theory. Hebb wrote: "When an axon of cell A is near enough to excite a cell B and repeatedly or persistently takes part in firing it," some growth or chemical change occurs in one or both cells, so that A becomes more efficient at firing B.

A few terms help here. A neuron is a nerve cell. Its axon is the long fiber that carries its signal out to other cells. A synapse is the tiny junction where one cell's axon meets the next cell. To "fire" means to send an electrical pulse.

Hebb guessed that the change might be the growth of small knobs at the synapse, which would enlarge the area of contact between the cells. He was candid that there was no direct evidence for this yet. He offered it as the most plausible mechanism, not a proven fact.

## Cell assemblies and phase sequences

The rule alone describes one pair of cells. Hebb's larger claim was about groups. If the same stimulation happens again and again, he argued, a scattered group of cells in the brain becomes linked into what he called a "cell assembly." Once formed, the assembly can stay active briefly on its own, like a small circuit that keeps itself going, and it can pass activity on to other assemblies.

A chain of assemblies firing one after another is what Hebb called a "phase sequence," and he identified it with the flow of thought. One assembly can be set off by a sense impression, by the assembly before it, or, most often, by both together.

An everyday analogy helps. Think of a path across a grassy field. The first person to cross leaves almost no trace. If many people take the same route, the grass wears down and the path becomes the easy way to go. Hebb's synapses are like that grass: repeated use makes a route easier to travel next time.

| Everyday picture | Hebb's term | What it means |
|---|---|---|
| One footprint | A single firing | Cell A helps fire cell B once |
| A worn path | A strengthened synapse | A now fires B more easily |
| A network of paths | A cell assembly | A group of cells that act together |
| A walk along several paths | A phase sequence | One assembly triggering the next, a thought |

> **Try it:** Pick a skill you practiced until it felt automatic, such as typing your password or a piano scale. Write down what changed between the first try and the hundredth. Then describe that change using Hebb's words: which "paths" do you think got worn in?

## Why the book mattered

The Canadian Encyclopedia credits the book with putting the mind back into the brain, at a moment when much of psychology had stopped talking about inner processes. Hebb's ideas also gave neuroscientists something to test. Work in Eric Kandel's laboratory on the sea slug *Aplysia*, for example, produced evidence that Hebb-like strengthening happens at real synapses.

For artificial intelligence, the book arrived at just the right time. Warren McCulloch and Walter Pitts had shown in 1943 that simple model neurons could compute logical functions, but their networks did not learn. Hebb supplied a plausible learning rule. Within a couple of years, Marvin Minsky and Dean Edmonds built SNARC, a learning machine of about forty artificial synapses that strengthened with use.

The rule is often shortened to "neurons that fire together, wire together." The slogan is handy but slightly off. Hebb said cell A must take part in firing cell B, which implies A fires just before B, not merely at the same moment. Timing turns out to matter in real brains.

## How is Hebb's rule used in artificial neural networks?

In an artificial neural network, each connection between two units has a number called a weight, which says how strongly one unit influences the other. Hebb's rule becomes a recipe for changing that number: when two connected units are active together, increase the weight between them. In its simplest form the change is just the product of the two activities, so it grows only when both are "on."

This kind of learning needs no teacher and no right answers. The network simply strengthens whatever patterns occur together often, which is why Hebbian learning is often described as a basis for unsupervised learning. Networks built on this idea can store a pattern and later recall the whole thing from a fragment, a little like recognizing a song from its first few notes.

Most of the large networks in the news in 2019, such as image recognizers and language models, are trained a different way, by measuring their errors and adjusting every weight to reduce them. Hebb's rule is still the starting point for understanding why connection strengths are where a network's knowledge lives.

## Sources

- The Organization of Behavior: A Neuropsychological Theory: https://archive.org/details/in.ernet.dli.2015.97836
- Hebbian theory: https://en.wikipedia.org/w/index.php?title=Hebbian_theory&oldid=920489553
- Donald Olding Hebb: https://www.thecanadianencyclopedia.ca/en/article/donald-olding-hebb