Hi everyone, I’ve been learning about HTM the past couple months and this is my first post here.
I’m reading “How the Mind Works” by Steven Pinker and his section on stereo vision raised questions for me about the assumption behind HTM that the neocortex is doing basically the same thing everywhere.
Pinker explains that stereo vision (the ability to see depth based on the disparity between images on the two eyes) is genetic based on a few lines of evidence:
- 2% of the population can see fine out of each eye, but can’t see in stereo (stereograms don’t work for them)
- There are different types of stereo blindness: one for pairs of points that coincide almost exactly for fine-grained depth perception, one for closer objects and one for farther objects. Whitman Richards, who discovered these types of stereo blindness, proposed that that there are three pools of neurons, one for each type. Neurons of these types have been found in brains of monkeys and cats.
- “The different kinds of stereoblindess appear to be genetically determined, suggesting that each pool of neurons is installed by a different combination of genes.”
Newborn babies can’t see in stereo, however he contends that babies do not “learn to see in stereo.” He prefers the explanation of psychologist Richard Held: initially the neurons in the visual cortex respond to inputs from either eye. Over the first few months of life, a competitive process leads the neurons to respond only to input from one eye or the other. Then:
“Around the three-month mark each neuron settles on a favorite eye to respond to. The neurons lying one connection downstream can now know when a mark falls on one spot in one eye and on the same spot, or a slightly shifted-over spot, in the other eye–the grist for stereo vision. In cats and monkeys, whose brains have been studied directly, this is indeed what happens. As soon as the animal’s cortex can tell the eyes apart, the animal sees stereograms in depth. That suggests that when the inputs are first tagged “left eye” or “right eye,” the circuitry for stereo computation on layer downstream is already installed and functioning.”
The idea that so much of the machinery for stereo vision is genetically programmed would seem to contradict the idea that the neocortex does pretty much the same thing everywhere, at least a strong version of it which claims any part of the neocortex could really do anything if you switch the inputs around (Which Jeff Hawkins seems to imply when he cites the ferret study in which they rewired ferret’s optic nerve to also stimulate the auditory cortex. I don’t know if he mentions this, but the ferrets didn’t see as well with their auditory cortex as with the visual cortex).
Pinker’s book is from 1997, so it could be the info I’m getting from his book is outdated. Also, I may be taking the homogeneous neocortex assumption too literally; that is, maybe people behind HTM agree that some functions can’t be explained by an initially fully homogeneous neocortex that gets differentiated solely by different inputs. It could still be the case that the basic algorithm is the same everywhere in the neocortex with genetically determined pre-wirings that differ across regions of the brain, but the neocortex is still flexible enough to adapt brain regions to new functions if the inputs changes, especially early in life.
Curious to hear people’s thoughts on this. Also, I know almost nothing about vision, so I apologize if this is a trivial question I could learn from any current textbook.