Three Schools of AI

A classification of AI research, common in Chinese textbooks, into logic-symbol, neural-network, and interact-with-environment approaches.

This is a classification of AI research commonly taught in Chinese AI textbooks, dividing approaches by “where intelligence comes from” into three schools. Symbolism (also called logicism) holds that intelligence is logical reasoning over symbols; its landmarks are Allen Newell and Herbert Simon's Logic Theorist program from the 1950s and the expert systems that followed. Connectionism (also called the bionic school) holds that intelligence comes from the connections and learning of large numbers of simple neurons; today's deep learning and large models fall under this school. Behaviorism (also called the cybernetics school) holds that intelligence is expressed in a perception-action loop and does not require a complete world model to be built first; its landmark is Rodney Brooks's subsumption-architecture robots at MIT in the 1980s. Embodied AI is philosophically closest to behaviorism but methodologically relies heavily on connectionist neural networks; in practice, the three schools have largely merged today.

ExampleTake the task “put the apple in the bowl”: a symbolist approach would hand-write a rule such as “if an apple is seen and the hand is empty, grasp it”; a connectionist approach would train a neural network on many demonstrations to map images directly to actions; a behaviorist approach would wire up a set of “see-reach-adjust” reflex loops and let the behavior take shape through interaction with the environment. Today's VLA models mainly follow the connectionist path.

Also called

Symbolism, Connectionism, and Behaviorism, Three Paradigms of AI

Related

Subsumption Architecture · Symbol Grounding Problem · Embodied Cognition · Perception-Action Loop · Neural Network · Moravec's Paradox

Sources

Symbolic artificial intelligence - Wikipedia
联结主义 - 维基百科 (Chinese)
Behavior-based robotics - Wikipedia