connectionism
/ kuh-NEK-shun-iz-um /
Connectionism is the approach to AI that takes inspiration from the brain: instead of hand-writing rules, you build a network of many simple units (loosely modeled on neurons) wired together, and let it learn by adjusting the strengths of those connections from examples. The intelligence isn't written down anywhere in particular — it emerges from the pattern of millions of tiny connection weights, the way a wave emerges from countless water molecules rather than from any single one.
It is the philosophical opposite of symbolic AI. Symbolic AI says: knowledge is explicit rules and symbols, and reasoning is logic. Connectionism says: knowledge is distributed across a web of connections, and learning is gradually tuning those connections so the network produces the right outputs. Where symbolic systems are readable but brittle, connectionist ones are flexible and great at messy perception — recognizing faces, sounds, handwriting — but their reasoning is opaque, buried in numbers no human can easily interpret.
Connectionism's fortunes rose and fell and rose again. Early excitement around the perceptron in the late 1950s faded after critics highlighted its limits, contributing to a quiet spell. It revived in the 1980s when backpropagation gave practical ways to train multi-layer networks. Then, from around 2012, with vast data and powerful hardware, it exploded into today's deep learning — the engine behind modern image recognition, speech, translation, and large language models. Almost everything people now call 'AI' is connectionist at its core.
Ask a connectionist network 'where is the rule that recognizes the letter A?' and there is no answer — no single line says 'an A has two slanted strokes and a bar.' The ability is spread across thousands of connection strengths tuned by training. That distributed, emergent quality is both why it handles messy real-world input so well and why it is so hard to explain.
In a connectionist network, knowledge lives in the connections, not in any readable rule.
Connectionism and symbolic AI are two answers to the same question. Connectionism (neural networks, deep learning) won the modern era for perception and language, but it inherits real weaknesses too: opacity, hunger for data, and shaky logical reasoning.