Source Summary: Hebb 1949 — The Organization of Behavior
Full title: The Organization of Behavior: A Neuropsychological Theory
Authors: Donald O. Hebb
Affiliation: McGill University
Venue: John Wiley & Sons, New York (1949)
PDF: papers/hebb-1949-the-organization-of-behavior-a.pdf
Contributions
- Hebbian learning rule — the postulate that synaptic strength increases when pre- and post-synaptic neurons fire together; the biological mechanism for learning and memory.
- Cell assembly — a reverberating circuit of neurons that forms a functional unit representing a concept or percept; the neural substrate of a memory or idea.
- Phase sequence — ordered sequences of cell assembly activations representing temporal patterns of thought and perception.
- Perceptual learning — complex visual perceptions are learned, not innate; the visual system requires experience to develop proper response to shapes (supported by studies of cataract patients gaining sight).
The Neurophysiological Postulate (Chapter 4)
This is the key passage from the book:
“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 process or metabolic change takes place in one or both cells such that A’s efficiency, as one of the cells firing B, is increased.”
More specifically, Hebb proposed that synaptic knobs (the terminal endings of axons) grow larger and develop new contacts when an axon repeatedly helps fire the postsynaptic neuron. The structural change increases the probability and strength of transmission across that synapse.
Summarized as: Cells that fire together, wire together.
(This paraphrase is by Siegrid Löwel, 1992 — not Hebb himself)
The postulate in modern notation
If neuron A and neuron B fire simultaneously (or A fires just before B), the weight w_{AB} increases:
Δw_{AB} = η · xₐ · xᵦ
where xₐ, xᵦ are the activations and η is the learning rate.
This is unsupervised: no error signal, no teacher. Weight changes depend only on co-activation.
The Cell Assembly (Chapter 4)
A cell assembly is a group of neurons that become strongly interconnected through Hebbian learning after being repeatedly co-activated by the same stimulus. Once formed:
- Activation of any subset of the assembly tends to activate the whole
- The assembly can sustain reverberant activity briefly after the stimulus ends (short-term memory trace)
- Repeated reverberant activity leads to permanent structural changes (long-term memory)
Cell assemblies are the proposed neural substrate for concepts, percepts, and memories. The word “apple” activates an assembly; the sight of an apple activates an assembly; learning associates these assemblies.
Phase sequences
Assemblies activate each other in learned sequences — the neural basis of sequential thought, planning, and expectation.
Key Empirical Evidence Cited
- Cataract patients (Senden, 1932): Adults who gain vision after congenital cataracts cannot immediately perceive shapes — they must learn to see. The visual system is not prewired for form recognition.
- Rats reared in darkness (Hebb’s own experiments): Show deficits in visual discrimination. Early experience is critical.
- Frontal lobe lesions: Patients show specific deficits in tasks requiring temporal integration and working memory — consistent with disruption of phase sequences.
- EEG spontaneous activity: The brain is persistently active, suggesting ongoing reverberatory processes (cell assemblies reverberating).
Limitations Acknowledged
- The postulate was speculative in 1949 — direct evidence for synaptic knob growth was not available
- The model makes no specific predictions about weight magnitudes or timing requirements
- No explicit mechanism for synaptic depression (forgetting); pure Hebbian learning is unstable — weights grow without bound
Influence
Hebb’s rule became the foundation for:
- Hopfield networks (1982): content-addressable memory using Hebbian weights
- Perceptron learning (Rosenblatt 1958): supervised version of weight update based on co-activity
- Backpropagation (1986): generalization to multilayer networks with error signal
- Long-term potentiation (LTP) discovery (Bliss & Lømo, 1973): experimental confirmation that high-frequency stimulation of a synapse strengthens it — the biological substrate Hebb predicted
- Spike-timing dependent plasticity (STDP): refined Hebbian rule where exact timing of pre/post firing matters
Key References
- McCulloch & Pitts 1943 — the neuron model Hebb built on
- Lorente de Nó 1938 — reverberatory circuits in the nervous system
- Senden 1932 — cataract patients and visual learning