Connectionism: Modeling the mind with neural networks

A lucid primer on connectionism as a model of mind. Starts with spreading activation (the Genghis-Khan node example, the Deese-Roediger-McDermott false-memory effect) and its failures, then builds full connectionism — units, weighted links, training a chair/not-chair classifier, hidden nodes — and explains why neural nets dissolve the ancient category-boundary problem (Plato's featherless biped vs Diogenes' plucked chicken) by reasoning over 'cluster structures in thingspace.' Catalogs brain-like properties: no grandmother cell (Lashley's nonlocalized memories), graceful degradation, realistic remembering/forgetting, soft-constraint satisfaction, context-sensitivity. Closes by integrating his behaviorism sequence: cognition updates on surprise, motivation updates on reinforcement, and behaviorism is just the simple straight-link special case of connectionism. Notably prescient for 2011, pre-deep-learning.

Quality 73 · Strong. Strong-ish explainer — substantive, well-organized, and impressively prescient, doing real integrative work (tying associationism, behaviorism, thingspace, and reinforcement into one picture). Pure exposition of established PDP/connectionism, so it stays low-Strong.

Claude’s paradigm shift 38 · Slight. Moderate — connectionism is 1980s Rumelhart-McClelland; the value is accessible exposition plus the behaviorism-as-special-case framing.

Real-world impact 2 · Minor. A lucid, prescient primer on connectionism that does real integrative work (tying associationism, behaviorism, and thingspace together; neural nets dissolving the category-boundary problem). Conceptual/pedagogical influence within rationalist discourse, exposition of established PDP, no material change — low RWI.