Grid Cells and the 'Where' and 'What' Pathways in the Brain?

I recently read “Vision and Art: The Biology of Seeing” by Margaret Livingstone, and I’ve just started diving into “Conscious MIND Resonant BRAiN” by Stephen Grossberg. As I was reading, my thoughts turned to Numenta’s Thousand Brains Theory, specifically focusing on grid cells within each mini-column of the neocortex.

Livingstone’s book highlights how visual information from the eyes is processed in distinct pathways: the “where” and “what” parts of the brain. The “where” pathway is responsible for spatial localization, helping us determine where objects are in our environment. Meanwhile, the “what” pathway processes information related to object recognition and identification.

Interestingly, Livingstone demonstrates how color perception plays a role in this process. In certain paintings, where colors have equal brightness (luminance), the brain perceives a subtle wiggle effect. This occurs because the brain relies on dark/bright contrast to figure out the “where” in an image. Color information, on the other hand, is primarily sent to the “what” area of the brain.

Now, let’s consider the role of grid cells. These specialized neurons are thought to play a crucial role in spatial navigation and mapping. They create a grid-like representation of our environment, helping us understand our position relative to objects around us.

Could grid cells be treated as data sent directly to the “where” part of the brain, rather than being distributed throughout every mini-column in the neocortex? This concept seems more logical, especially in light of Livingstone’s discussion on brain injuries. She notes that when the “where”region is compromised, individuals retain the ability to identify objects (“what”) yet struggle with locating them (“where”) or coordinating actions to interact with them.


The way it looks to me, “where” pathway represents spatio-temporal frame of reference, in which the original input flow is ordered. Yes, that would be “distributed” grid cells. “What” pathway is secondary, it reframes the input flow in a taxonomy of recognized objects. These are formed through some kind of clustering in primary areas, they don’t exist in original input flow. So both would perfom connectivity clustering through laminar connections, which don’t really exist in typical ANN. The difference is that “what” pathway requires a form of Spectral clustering - Wikipedia before it can do this laminar connectivity clustering. So they are not exactly parallel, “what” is a higher-order or derived frame of reference. And I think if we look closer, there are even higher-order what-of-what frames of reference in higher association areas.

1 Like