Connectionism is the approach to understanding the mind that holds that mental phenomena arise from the collective activity of large numbers of simple, neuron-like units linked by weighted connections. Knowledge is not stored in discrete symbols at specific locations; it is encoded in the pattern of connection strengths spread across the whole network, and learning consists of adjusting those strengths. Concepts are represented as distributed patterns of activity rather than as single entries.
The modern connectionist program was consolidated by the 1986 two-volume work “Parallel Distributed Processing” from David Rumelhart, James McClelland, and their research group. That work argued that many cognitive abilities, including perception, memory, and language, could be modeled by networks that learn from examples, and it helped revive neural networks as a scientific enterprise after years of neglect.
Connectionism has long stood in tension with the symbolic tradition associated with Jerry Fodor’s language of thought, which insists that thinking requires structured, rule-governed symbol manipulation. The debate centers on whether distributed networks can capture the systematic, compositional character of human thought. It is a live question again, because today’s large neural networks are the most powerful connectionist systems ever built, and people disagree about how much genuine compositional structure they possess.
For a general reader, connectionism is the intellectual lineage of deep learning: nearly every modern AI system is a connectionist model at heart, so the framework’s strengths and its long-debated limits bear directly on what these systems can do.