If you’ve ever tried to learn a new skill, you’ll have experienced Hebbian Learning (although you may have just thought of it as ‘learning’). Take driving, for instance. When you start out, everything you do is incredibly deliberate. You remind yourself to turn on your indicator, to check your blind spot, and so on. However, after years of experience, these processes become so automatic that you perform them without even thinking.

The neuroscientific concept of Hebbian learning was introduced by Donald Hebb in his 1949 publication of The Organization of Behaviour. Also known as Hebb’s Rule or Cell Assembly Theory, Hebbian Learning attempts to connect the psychological and neurological underpinnings of learning.

The basis of the theory is when our brains learn something new, neurons are activated and connected with other neurons, forming a neural network. These connections start off weak, but each time the stimulus is repeated, the connections grow stronger and stronger, and the action becomes more intuitive.

A good example is the act of learning to drive. When you start out, everything you do is incredibly deliberate. You remind yourself to turn on your indicator, to check your blind spot, and so on. However, after years of experience, these processes become so automatic that you perform them without even thinking.

Neuron: The basic building block of the brain, composed of a cell body, axons, and dendrites. They transmit information to other neurons through electric currents.

Synapse: The contact point where one neuron meets another, allowing them to pass messages to each other.

Neural Network: A group of connected neurons.

Synaptic Plasticity: The brain’s ability to strengthen or weaken connections between neurons (called synapses) based on experience and activity. It’s how the brain learns, stores memories, and adapts over time.

Backpropagation: A common method in machine learning used to train artificial neural networks. It works by adjusting the network’s internal connections (called weights) to reduce the difference between the network’s prediction and the correct answer.

Artificial Neural Networks (ANNs): Computer systems inspired by the structure and function of the human brain.

Hopfield Network: A type of artificial neural network designed to function as a form of associative memory system. In turn, an associative memory system is a type of memory model that allows you to retrieve a complete memory or pattern based on a partial or related input. 

Canadian-born Donald Hebb originally wanted to be a writer, receiving a BA from Dalhousie University. He was a teacher until he finished his Master’s degree in psychology from McGill University. Intriguingly, Hebb sketched the basis of his idea of neural networks in his MA thesis at Dalhousie, although he described the paper as ‘nonsense’ in later years.¹

Showing an interest in the physiology of psychology, Hebb went on to do a PhD with Karl Lashley at the University of Chicago, a renowned behaviorist of the time. It was likely this period of his career that sharpened his ability to combine behaviorist theories, particularly reinforcement learning, with physiology and neuroscience. Ivan Pavlov’s influence is particularly apparent in Hebb’s research, and many parallels have been drawn between Hebbian Learning and Pavlov’s conditioning theory.²

When he finished his work with Lashley, Hebb returned to Montreal in 1937 to work with Wilder Penfield at the Montreal Neurological Institute. Influenced by Lashley and other psychologists who had observed how cognitive functions (particularly memory) were localized in specific regions of the brain, Hebb first outlined what would be labelled Hebbian Learning in his 1949 book, The Organization of Behaviour³. The idea that neurons are capable of forming networks that create and store memories, crucial to learning, was groundbreaking at the time, and remains hugely influential to this day.

Hebb’s combining of psychological and neuroscientific concepts earned him the title of the “father of neuropsychology.” Hebb is credited with connecting the abstract concept of ‘the mind’, to specific physiological and biological brain functions. Neuropsychology is studied by both psychologists and neuroscientists to this day, who consistently explore the connection between brain function and behavior.

Donald Hebb

Canadian psychologist and inventor of Hebbian Learning Donald Hebb is considered the ‘father of neuropsychology.’  His groundbreaking experiments linked conventional psychology with advances in physiology and biology.

Karl Lashley

American psychologist and behaviorist, Lashley is known for his many contributions to the study of learning and memory. Lashley was one of the first psychologists to perform experiments on the brains of rats; he studied the effects of lesions by removing certain areas of the rat’s cortex. Lashley made several groundbreaking discoveries on how the brain stores and processes information, which greatly influenced his PhD student, Donald Hebb.

Hebb’s theory of neural connections has had major implications for how neuroscientists and psychologists understand memory. One example is Long-Term Potentiation (LTP), a theory that emerged in the late 1960s, showing that synapses are strengthened by recent patterns of activity, therefore confirming the findings of Hebbian Learning.⁴ Today, LTP continues to be a central focus of neuroscience, especially in research into Alzheimer’s disease, dementia, addiction, and PTSD. In these areas, scientists are investigating how strengthening or restoring synaptic plasticity (basically how malleable and adaptable the connections in our brain are) might improve cognitive function.

When it was first introduced, Hebbian Learning was seen as one part of the puzzle that surrounded memory function and storage. Around the 1950’s, however, researchers grew  certain that neural networks were  responsible for storing and retrieving associations, which opened up a whole new area of neuroscientific discovery. Combined with Lashley’s findings in his memory localization experiments, it was starting to become clearer that human beings possess different types of memories, and that each one is located in a different part of the brain.

These hypotheses were confirmed by  the fascinating case of Henry Molaison, who was, and still is, famously referred to as HM. HM suffered from epilepsy, and in 1953, his surgeon recommended that he have the medial temporal lobes of his brain removed. The surgery was successful in curing his epilepsy, but it appeared to severely damage his explicit, long-term memory recall (ie. his memory for factual information). HM could easily deploy his short-term memory, and had no trouble acquiring new motor skills or using what psychologists call his ‘implicit’ long-term memory (ie. ability to complete procedural tasks like riding a bike). HM wasn’t able to recall, however, that he had ridden a bike the day before since this was a “fact.” As a result of HM’s experience, medial temporal lobes are considered responsible for the formation of long-term explicit memories. Hebb’s theory was used to support this approach to the study of memory, since the idea that memories are formed and stored through neural connections was supported by the case of HM.

Since then, neuroscience has significantly advanced our understanding of memory systems. Technologies such as fMRI, optogenetics, and real-time brain mapping have allowed researchers to observe synaptic changes as they happen. More refined models, such as Spike-Timing-Dependent Plasticity (STDP), have built on Hebb’s ideas by showing that not just co-activation, but precise timing between neuron firings, determines whether a connection is strengthened or weakened.

Current research is focused on how we can make AI more like the human brain, and Hebbian learning plays an important role in this quest. A recent study explored how Hebbian learning could be used to train Convolutional Neural Networks (CNNs)—a type of AI model often used for image recognition—without relying on traditional, resource-heavy training methods like backpropagation.¹¹ By mimicking how real neurons strengthen their connections through repeated activity, the researchers built a more brain-like, energy-efficient system that achieved nearly the same accuracy as standard CNNs. Their findings suggest that Hebbian learning can help create AI that is not only powerful but also more biologically realistic and interpretable.

In general, Hebbian Learning has been well-accepted across neuroscience and psychology. However, the theory is not without its limitations and controversies. Starting with the basic premise—“neurons that fire together, wire together”—some have argued that this is a highly simplified model of how learning actually happens in the brain. While it’s an important component of the brain’s plasticity, it’s only part of the story.

Let’s think about what that famous phrase actually means. Hebb’s original idea is often interpreted as referring to neurons that are synchronized, but it’s actually more general than that. He meant that if one neuron repeatedly helps activate another—say, A helps B fire—then the connection between them gets stronger. It's about consistent co-activation over time, not necessarily precise timing.

Researchers from the Ernst Strüngmann Institute argue that what really matters is whether the neurons are firing in sync.⁸ Specifically, they looked at the brain's activity when synchronized with strong gamma waves in the lower spectrum (about 20-48 Hz), and found that certain neurons became more responsive to specific stimuli. In the study, when cats were shown images while their brains were stimulated in sync with gamma rhythms, the neurons in their visual cortex adjusted to better respond to the image. But when there was no rhythmic syncing, the neurons didn’t change much. This suggests that synchronized brain activity is key to long-term learning, helping the brain decide which information is important enough to remember.

Criticisms also arise when Hebbian learning is used as a basis for artificial intelligence or algorithm development. The primary issue is that incorrect or inappropriate neural connections can be made, and worse, strengthened, under Hebbian Learning. In some cases, this could lead to bad habits, poor learning, and potentially destructive behavior. Another issue is that there is no upper limit for how strong a connection can become, so from a computational perspective, some type of control is required to prevent ‘over-learning.’ Finally, hundreds of experiments have shown that human learning is hugely influenced by feedback, from other people as well as the stimuli we encounter in our daily lives. Hebbian Learning does not fully account for feedback mediation, although this area has received greater attention in more recent decades.⁵

Hebbian Learning & Artificial Intelligence

Hebbian Learning has also been the basis for several advances in computer science, especially artificial intelligence and artificial neural networks (ANNs). Hebb’s original idea—“neurons that fire together, wire together”—proposed that the strength of connections between neurons increases when they are activated simultaneously. This principle became foundational for unsupervised learning in ANNs.

The Hopfield Network, an ANN introduced by John Hopfield in 1982, is based on rules stipulated under Hebbian Learning.⁶ By creating an artificial neural network, Hopfield found that information can be stored and retrieved in similar ways to the human brain. Through repetition and continuous learning, artificial intelligence can strengthen certain connections that speed its processing of certain situations, helping it ‘learn’ faster. This has proven especially useful in the case of pattern recognition and the development of AI algorithms.

Hebbian learning has also been the focus of research in terms of AI applied to medical imaging.⁷ Diffusion Tensor Imaging (DTI) and fiber tractography are advanced medical imaging techniques used to map the brain’s white matter pathways—basically, the “wiring” that connects different parts of the brain. These techniques rely on tracking the movement of water molecules along nerve fibers, which helps doctors and researchers see how different brain regions are connected.

In the context of AI, especially using Hebbian learning, these imaging techniques can be enhanced. In AI, the basic principle of Hebbian learning means that when certain patterns or pathways are activated together repeatedly—such as specific structures in brain scans—the system learns to strengthen those associations.

When applied to DTI and tractography, AI models can use Hebbian-like rules to improve how they identify and segment fiber pathways, especially in complex or noisy brain data. For example, if certain diffusion patterns (the way water molecules move through tissues in the body) often appear together in healthy brains, the AI can learn this association and use it to more accurately detect abnormalities in patient scans. Over time, the AI becomes better at recognizing brain structure patterns, helping with tasks like tumor mapping, surgical planning, or understanding neurological diseases.

Hebbian Learning & Deep Brain Stimulation (DBS)

Deep brain stimulation (DBS) is a well-established treatment for several neurological conditions, including Parkinson’s disease, essential tremor, epilepsy, and Tourette syndrome. The procedure involves implanting electrodes into specific areas of the brain to deliver controlled electrical impulses. However, despite working for many conditions, the underlying neural mechanisms of DBS and its long-term effects on brain connectivity are still unclear. One rare case study, however, suggests that it may have something to do with Hebbian principles.¹⁰ 

Researchers from various universities in the UK, Denmark, and Spain conducted a study on a patient with Parkinson’s disease, analyzing the brain’s white matter connections before and after extended DBS treatment. Parkinson’s disease is a progressive neurological disorder that affects movement, caused by the degeneration of dopamine-producing neurons in a brain region called the substantia nigra. Over time, this disruption affects the brain’s neural connectivity. 

The patient received DBS for seven years. Using computational modeling, they assessed the impact of DBS on both the structural and functional connectivity within the patient’s brain. The findings suggest that DBS can lead to brain changes similar to Hebbian learning, where connections between certain brain areas become stronger with repeated activity. In this case, DBS strengthened connections between regions of the patient’s brain that are involved in movement control. These changes also affected how different parts of the brain communicate at rest, making the brain’s activity patterns look more like those of healthy individuals. This study shows that DBS might not only help with symptoms in the short term but also help reorganize and repair brain networks over time, giving new insights into how the brain can adapt and how to make DBS treatments even more effective.

Algorithms for Simpler Decision-Making (1/2): The Case for Cognitive Prosthetics

Jason Burton explores how algorithms can and should be used to optimize our lives, but with a warning label attached.

Why do we retain information better when we learn it over a long time period?

The spacing effect occurs when information is repeatedly learned over a spaced-out period of time, resulting in an individual’s increased ability  to remember the information.

Tempting the Creation of Habits

Yasmine Kalkstein explores how we can create habits that become automatic behaviors. Much of this  thinking stems from ideas that are linked to Hebbian Learning and reinforcement theory.

1. Milner, P. (2003). A brief history of the Hebbian learning rule. Canadian Psychology, 44, 5-9.
2. Langille, J. J., & Brown, R. E. (2018). The synaptic theory of memory: a historical survey and reconciliation of recent opposition. Frontiers in systems neuroscience, 12, 52.
3. Hebb, D. O. (1949). The organization of behavior; a neuropsycholocigal theory. A Wiley Book in Clinical Psychology, 62, 78.
4. Purves, D., Augustine, G., Fitzpatrick, D., Katz, L., LaMantia, A., McNamara, J., & Williams, S. (2001). Neuroscience 2nd edition. sunderland (ma) sinauer associates. Types of Eye Movements and Their Functions.
5. McClelland, J. L. (2006). How far can you go with Hebbian learning, and when does it lead you astray. Processes of change in brain and cognitive development: Attention and performance xxi, 21, 33-69.
6. Sathasivam, Saratha (2008). “Logic Learning in Hopfield Networks”. arXiv:0804.4075 [cs.LO].
7. Walizai, T. (2024, July 9). Diffusion tensor imaging and fiber tractography. Radiopaedia. https://radiopaedia.org/articles/diffusion-tensor-imaging-and-fiber-tractography-1?embed_domain=hackmd.io%25252f%252540yipuafecsl2jsu8smr5njq%25252fbnjhjgjghjghjgh&lang=us
8. Galuske RAW, Munk HJM, Singer W (2019). Relation between gamma oscillations and neuronal plasticity in visual cortex. Proc Natl Acad Sci USA 116(46), 23317-23325. https://doi.org/10.1073/pnas.1901277116
9. Caporale, N., & Dan, Y. (2008). Spike timing–dependent plasticity: a Hebbian learning rule. Annual Review of Neuroscience, 31, 25–46.
10. Van Hartevelt, T. J. et al. (2015). Evidence from a rare case study for Hebbian-like changes in structural connectivity induced by long-term deep brain stimulation. Frontiers in Behavioural Neuroscience, 9. https://doi.org/10.3389/fnbeh.2015.00167
11. Nimmo, J. J. & Mondragon, E. (2025). Advancing the Biological Plausibility and Efficacy of Hebbian Convolutional Neural Networks. arXiv:2501.17266v1 [cs.NE]. https://arxiv.org/pdf/2501.17266
12. Shatz CJ. The developing brain. Sci Am. 1992 Sep;267(3):60-7. doi: 10.1038/scientificamerican0992-60. PMID: 1502524.

Squire L. R. (2009). The legacy of patient H.M. for neuroscience. Neuron, 61(1), 6–9. https://doi.org/10.1016/j.neuron.2008.12.023