How do many agencies communicate about places and shapes? Minsky's hypothesis is that "many agencies inside our brains use frames whose terminals are controlled by interaction-square arrays. Only now we'll use those square arrays not to represent the interactions of different causes, but to describe the relations between closely related locations." These are agents called "direction-nemes".
NB: The basic idea here is that different agents are laid out in a grid, and the activation of one agent in the grid will represent a direction relative to the center of the grid. I wonder if this kind of scheme is practical, though. Lots of the work in visual cognition argues that there are many, many different spatial relations that could be computed -- too many to compute automatically. This is why Ullman and Pylyshyn appeal to a level of processing called visual cognition. Wouldn't it be better in Minsky's framework to let different agencies develop for computing these different relationships when required? (Or am I missing the point -- do the direction-nemes work automatically or not??)