Oliver Selfridge Proposes Pandemonium

Back at MIT, Oliver Selfridge of its Lincoln Laboratory, presented "Pandemonium: A Paradigm for Learning" at the Symposium on the Mechanisation of Thought Processes at the National Physical Laboratory in Teddington, England, in November 1958; the proceedings appeared in 1959. It was an exciting time for computing, before it had really been pigeonholed into less experimental roles from the late '60s onwards (until maybe around 2010 with the advent of deep learning proper).

Selfridge's model for learning to recognize patterns had four levels of "demons." Data demons stored the image and computational demons computed features from it. Cognitive demons weighed that evidence and shouted; a decision demon picked the loudest. It learned by adjusting the weight given to each feature and by culling and breeding its feature demons through a kind of natural selection.

Not neural networks but "demons," one can still see how prevalent statistical weights are in machine learning.