A computational approach to modeling the brain which relies on the interconnection of many simple units to produce complex behavior. See also history of connectionism, symbolicism, dynamical systems theory.

Connectionism has a number of important considerations for the philosophy of mind. By positing connectionist models as the best way to model human cognition, philosophers have begun to see high level mental properties as: "emergent properties... that depend on lower-level phenomena in some systematic way" (Churchland and Sejnowski, 1992, p.2). These commitments have redefined for many the best way to understand the nature of representation and computation in the human mind. In his well known paper 'On the proper treatment of connectionism', Paul Smolensky forwards a connectionist (or subsymbolic) hypothesis in order to capture these new ideas:
The intuitive processor is a subconceptual connectionst dynamical system that does not admit a complete, forma, and precise conceptual-level description.
This commitment to a 'subconceptual' level of description of cognitive processes is a direct rejection of the symbolicist or GOFAI approach to human cognition.
Connectionist models can be classified by representational commitments in two categories; distributed and localist. Distributed representations are vectors in a representational state space, and are processed simultaneously by many nodes in a connectionist network. Localist models use individual nodes to represent one entire concept (such as 'dog'). In general, distributed representations are more neurologically realistic that localist representations. However, distributed models are often far more complex and difficult to analyze than localist models.
Critiques of connectionism have been forwarded by dynamic systems theorists, symbolicists, and neuroscientists. In particular, dynamic systems theorists claim that connectionist models are unrealistically wedded to ideas of representation and computation. Neuroscientists often note the lack of neurological realism in connectionist networks. These networks often have too little recursion, far too much inhibition, unrealistic learning algorithms, simplistic transfer functions, and no analog to the large number of neurotransmitters and hormones which affect human cognition.
Symbolicists have taken a number of lines of argument. Fodor and Pylyshyn (1988) have criticized connectionism as not being able to support the systematic and productive natures of human thought. As well, it is thought that the only role for connectionist work is to provide a method for implementing a symbolicist system in a manner similar to the brain. Thus, the best level of description of human cognition remains at the symbolic level. In recent years, however, a number of connectionist models have been produced which shows these criticisms to be questionable.
Chris Eliasmith