Donald Hebb and the Origins of Learning in Neural Networks

·

In his unpublished manuscript Scientific Methods in Psychology: A Theory of Epistemology Based on Objective Methods in Psychology, Hebb (1934) explored the philosophical concept of mind and critically examined the role of introspection in psychological inquiry. He rejected introspection as a reliable scientific method and sought instead to integrate a theory of mind with behaviorist principles and the physiological foundations of mental activity. In Chapter 2, titled “Idealism,” he reflected on how both the brain and the mind contribute to perception. Hebb argued that the mind can be understood as a theory of behavior. His discussion shows clear parallels with classical empiricism, particularly in the works of Bacon, Locke, and Hume, who emphasized observation and experience over introspection as the foundation of understanding mental processes (Brown, 2020, p. 27).

Central Processes and Intelligence

Hebbian learning is often summarized by the phrase: “neurons that fire together wire together.” While this principle may seem counterintuitive, Hebb described it in physiological terms: when one neuron repeatedly contributes to the firing of another, structural or metabolic changes occur that increase the efficacy of the connection between them (Kozma, 2019, p. 2).

According to Hebb (1949), behavior in relatively simple mammals, such as rats, involves more than a direct interaction between sensory input and motor output.Some form of internal processing – analogous to thought – appears to intervene. While the term “thought” suggests human-level complexity, observations of rat behavior demonstrate that the brain generates actions beyond immediate sensory inputs.Instead, central neural processes influence behavior independently of direct sensory stimulation.

What, then, is the nature of these relatively autonomous processes within the cerebrum? Hebb (1949, Preface xvi) drew on clinical observations to explore this question. For example, some individuals maintain high intellectual performance even after major cortical damage. One man retained a very high IQ after removal of a prefrontal lobe (Hebb, 1939), and a woman described by Rowe (1937) achieved an above-average IQ despite losing the right hemisphere of her cortex (Hebb, 1949, p. 2). Hebb personally evaluated such patients and observed minimal changes on the standardized tests available at the time. He concluded either that extensive frontal lobe removal had limited effect on mental abilities or that the tests were insufficiently sensitive to detect subtle deficits. These experiences at the Montreal Neurological Institute shaped his emerging ideas about intelligence and its assessment.

Hebb (1949, p. 7) proposed that there is a rational basis for a central neural factor that modulates the effect of a stimulus. While the initial problem of discovering its rules seems purely neurophysiological, much of the evidence comes from psychological or behavioral data. Attention plays a central role in this issue, and it becomes most apparent in the activity of the whole organism. Hebb emphasized that understanding these central processes requires combining insights from both physiological and behavioral evidence.

Perception: Growth of the Assembly

One of Hebb’s most important contributions to psychological theory is the concept of the cell assembly, a framework linking perceptual generalization, the permanence of learning, and attention to coordinated neural networks (Hebb, 1949, pp. 69–74).

Hebb proposed several key points regarding synaptic function:

  1. Structural connections exist between individual neurons, but a single neuron is not the sole unit of transmission; connections contribute to the direction of signaling but do not fully determine it.
  2. In the adult brain, neurons generally do not establish direct sensory-motor connections in this manner.
  3. Reverberatory activity, the ongoing excitation within a neural circuit, drives structural changes at synapses, forming a dual-trace mechanism that supports memory and learning.

A likely mechanism for strengthening one neuron’s influence over another involves the growth of synaptic knobs. When a neuron repeatedly contributes to activating another, its axon terminals can enlarge or develop additional contact points on the receiving neuron’s soma. This structural change produces a lasting effect of reverberatory activity and embodies the principle that repeated co-activation strengthens neural connections. In essence, any two neurons or groups of neurons that are consistently active together tend to become associated, so that activity in one facilitates activity in the other (Hebb, 1949, p. 70).

A cell assembly is a network of neurons and their pathways that function as a coordinated unit. Activating part of an assembly can trigger a self-sustaining, reverberating circuit involving multiple interconnected pathways. This persistence allows the organism to bridge temporal gaps between stimulus and response.

When multiple cell assemblies are linked in a temporally organized pattern, they form a phase sequence. Phase sequences are chains of assemblies that influence each other over time, providing the neural substrate for a train of thought. This sequential activation enables associative learning, imagination, and the combination of independent concepts into new ideas (Hebb, 1949, pp. 79–106).

The cell assembly can also be viewed as a set of neurons forming closed loops, connected through perceptual learning. These self-exciting loops can maintain activity even in the absence of external stimuli, providing the neural basis for thought and internal mental activity (Hebb, 1949, p. 103).

Modern learning algorithms, such as the LMS rule, echo Hebb’s principle: connection strengths are adjusted in proportion to error signals, mirroring Hebb’s idea that repeated co-activation modifies neural connections and supports learning (Kozma, 2019, p. 4).

References

Hebb, D. O. (1934). Scientific Methods in Psychology: A Theory of Epistemology Based on Objective Methods in Psychology. Unpublished manuscript.

Hebb, D. O. (1939). The Effects of Prefrontal Lobotomy on Intellectual Functioning. Montreal Neurological Institute.

Hebb, D. O. (1949). The Organization of Behavior. New York: Wiley.

Kozma, R. (2019). Artificial Intelligence in the Age of Neural Networks and Brain Computing.

Langille, J., & Brown, R. (n.d.). The Synaptic Theory of Memory: A Historical Survey and Reconciliation of Recent Opposition.

Brown, R. (2020). Donald O. Hebb and the Organization of Behavior: 17 years in the writing. Molecular Brain.

Rowe, D. (1937). Case study of cortical lesions and intelligence.