Neural representations: some old problems revisited.

Hebb's (1949) cell assembly, originally conceived as an explanation for stimulus equivalence, also serves as a neural representation of stimuli. Association between cell assemblies was a major theme of Hebb's book, but the state of physiological knowledge at the time was such that no satisfactory basis for it could he devised. Subsequent theory has been more concerned with the recognition and "attractor" features of the cell assembly than its capacity to represent and associate concepts. This is unfortunate because while generalization is important, so is discrimination, which is not well served by an attractor model. This dilemma is avoided by postulating that stimulus representation and stimulus equivalence involve different neural circuits. Human beings can instantly form and use associations between many more concepts than there are synapses on the average cortical neuron, indicating that the associative links between engrams are sparse. The connections within an engram, on the other hand. must be dense to ensure that a weak input can activate all its neurons. It would appear that two processes are anatomically and physiologically different, which may account for the fact that engrams remain distinct in spite of being associated with each other. The fact that a single concept may have very many associations puts a heavy demand on the process of selective attention to avert complete chaos. I propose that attention is a manifestation of motivation. Motivation facilitates responses, which in turn facilitate engrams of associated stimuli. The enhanced engram activity is fed back through centrifugal paths to intensify sensory input that has previously played a part in executing the planned responses. Attention may also contribute to a mechanism that prevents the engrams of component parts of an object from being assimilated into the engram of the whole.