Plato's Cave: Reification


Webster defines "re-ify" [L res thing - more at REAL]: to regard something abstract as a material thing.

I have adopted the word reification to indicate the inverse operation to abstraction. Abstraction implies a many-to-one transformation from the many possible variants to a single invariant form. Reification on the other hand implies not a one-to-many transformation, which would potentially produce an infinite variety of variants, but rather a one-to-one-of-many transformation, although the exact variant that is generated could be any one of the infinite variety of variant forms.

For example, imagine a "node" (cell? cell assembly? pattern of activation?) in your brain which represents the concept "square", and which "lights up" in the presence of any square at any rotation, translation, or scale. A "top-down" activation of that node would potentially be able to generate a low-level representation of any one of those squares. In the absence of a visual input, such a reification would produce any one of those possible squares. In the presence of a partial input on the other hand, the reification would produce the single square which most closely matches both the bottom-up partial input, and the top-down invariant square. The reification would therefore produce a figural completion, or filling-in of missing information which is specific to the partial input provided.

Reification implies not just a feed-forward processing through the hierarchy of visual representation, but both a bottom-up feed-forward, and a simultaneous top-down feed-back processing of information.

The reason, I propose, that reification has not been considered much in the neural computation literature is not for lack of evidence of reification in perception, but rather for lack of ideas as to how such functionality might be achieved using conventional neural elements without invoking combinatorial complexity. This suggests that our understanding of conventional elements is missing some fundamental properties.

Return to argument

Return to Steve Lehar