Val Dusek, The tangled roots of neural network theory: philosophical and psychological

The basic idea of neural networks has been around since the late 1940s. The neurophysiological model for the early version of neural networks was suggested by the work of psychologist Donald O. Hebb (1949). Hebb described “reverberating circuits” in the brain as the basis of explanation of thought. There was a less well known philosophical generalization of Hebb’s work by the otherwise famous, Austrian, free market, conservative economist-philosopher Friedrich Hayek (1952). Hayek attempted to eliminate phenomenal sensory qualities by reducing them to a system of relations, rather along the lines of Mach and Carnap.

Hayek conceived of the mind as without a central organizing principle, but made up of competing neural elements, rather the way Adam Smith’s (1776) economic “invisible hand” of the market emerges out of individual competition.

Warren S. McCulloch and Walter H. Pitts developed the logical theory of neural networks. Interestingly for philosophers, and perhaps surprisingly for cognitive scientists, McCulloch was a follower of the medieval scholastic philosopher John Duns Scotus (McCulloch, 1961, pp. 5–7). The realist theory of universals of Duns Scotus was followed by the American pragmatist philosopher Charles S. Peirce. (In a little-noticed turn of twentieth-century philosophy, Martin Heidegger (1916), the German hermeneutic phenomenologist, and Peirce (1869), the pioneer of symbolic logic, were both early influenced by a work on “speculative grammar” purporting to be by Duns Scotus (Bursill-Hall, 1972). The scholasticism of Duns Scotus’s seventeenth-century followers was so rejected and ridiculed by early modern philosophers such as Bacon that “Duns” was the origin of our term “dunce,” for fool, yet his views lie behind two major branches of philosophy of language in the twentieth century.) Walter Pitts, McCulloch’s mathematical amanuensis, was living as a 13-year-old runaway in Chicago when he ran into a library to hide from bullies. Next to his hiding place was Bertrand Russell’s three-volume Principia Mathematica, deriving mathematics from logic, which Pitts read in a few days. He wrote to Russell in England with criticisms and corrections. According to one account Pitts happened later to meet Russell in a Chicago park (where he initially took the then somewhat seedy looking Russell for a fellow street person), and attended a lecture of Russell’s in Chicago. Russell recommended Pitts to the Vienna Circle logical positivist Rudolf Carnap at the University of Chicago. Pitts then went to MIT and worked with Norbert Wiener (1894–1964), himself an ex-prodigy (Heims, 1980), on cybernetics feedback mathematics, as well as with Jerome Lettvin, who wrote, among many articles with Pitts and McCulloch, a brilliant analysis of the visual system of the frog (Lettvin et al., 1959). Pitts, looking for an absent father figure, was everyone’s son at MIT and did the detailed mathematical working out of everyone else’s ideas.