Connectionism / Parallel Distributed Processing (PDP) / Neural Networks: a few milestones backward in time, showing that "new" is relative. (Note that page references are currently listed only for the FIFTH edition of Hergenhahn's An Introduction To The History Of Psychology, though the references are easily identifiable from the index of the sixth edition.)

  • David Rumelhart and James McClelland (1986): Connectionism re-emerges in a more complex form and becomes psychology's dominant model of the mind, with parallel processing by many simple neuron-like units instead of the serial step-by-step following of programmed instructions (as implied by traditional artifical intelligence and cognitive science research). (p. 584)
  • Marvin Minsky and Seymour Papert (1969): Connectionism in the form of "perceptrons" was mathematically demolished as a possible model of the working of the mind. (pp. 583-584)
  • Frank Rosenblatt (1958): The "perceptron" was proposed as a model of the mind based on the architecture of the brain, with an "input" and an "output" set of simple neuron-like units, whose interconnections determine the output (or result of processing) for a given input. (p. 583)
  • Donald Hebb (1949): proposed that the connection between two neurons is strengthened when they become active at the same time -- the basis of the Hebb Rule which is still the fundamental mechanism for learning in connectionist networks. (pp. 581-582, 559-560)
  • Warren McCulloch and Walter Pitts (1943): noted that some features of networks of neurons in the brain -- their arrangements of excitatory and inhibitory connections -- could be the basis of computation and mental processing (p. 582)
  • Edwin Guthrie (1935): argued that the "stimulus" of behavioral psychology is actually a practically infinite multitude of micro-stimuli, representing every possible sensation and bodily state impinging on an organism at a given moment -- thus implying for future theorists that an "input" actually consists of a pattern of a vast number of detectable but unspecifiable inputs, just as connectionism will eventually make use of.
  • Edward Thorndike (1898): coined the term "connectionism" to describe his theory in which associationism is first applied to behavior in terms of connections between stimuli and responses, instead of associations among ideas in the mind. The "connections" he literally had in mind were between stimulus and response centers in the brain. (p. 343)
  • David Hartley (1749): related sensations to physical vibrations in the nervous system, giving a material basis to the associaton of ideas and claiming that associations were formed when such vibrations occurred at the same time (thus anticipating the Hebb rule by 200 years). (p. 136)
  • David Hume (1748/1739): claimed that the mind itself was an unnecessary element of theories of mental life, since if sensations and ideas acted in accord with some simple laws of association (such as resemblance and contiguity), they would combine in such a way as to build our concepts and thoughts without some other entity like "the mind" having to play a role (i.e., the ideas would think for themselves!); this anticipated the neural network theorist's goal of computing using simple rules for modulating connections among simple units, instead of using explicit sets of program instructions. (pp. 132, 133-134)
  • Aristotle (c. 350 BC): conceived of associations among ideas (based on their simliarity, contrast, and contiguity) as the basis of all complex thought, thus setting the terms of the whole psychological enterprise for the next 2300 years (and counting...) (p. 49)