Thursday, March 31, 2005
The Color Graph Hypothesis
To celebrate my blog's first birthday, I'm going to speculate about a cool idea I've been mulling over. I've always greatly enjoyed my sensation of color. It's perfectly clear that this phenomenon - say for instance the hot pink that one sees from the Hydrogen Emission Spectrum - is a product of the patterns of firing neurons in the brain. But in particular, I've been thinking that it should be possible to discover precisely what types of patterns of neurons are responsible for the sensations of different colors. Say you are viewing a clear blue sky over a smooth field of fresh green grass (I'm ready for spring!). The 470 nm Rayleigh scattered photons entering the eye will preferentially activate the blue cone cells in the bottom half of the retina, while the 530 nm photons reflected off of the grass will trigger the green cone cells in the top half of the retina. After some basic preprocessing of the image by the retinal neurons (for instance find the location and orientation of the borders between objects), the encoded image is sent via electrical impulses along the optic nerve to the visual cortex. We effectively see about a million "pixels", with most of our resolution concentrated in the center of our visual field (this high resolution is produced by the fovea), and the optic nerve is composed of about 1 million individual fibers. The brain itself contains around 100 billion neurons, so if we estimate that the visual cortex gets 10 billion neurons in order to create our perception of the image, then we can have about 10,000 neurons dedicated to each effective pixel in our visual field.
So the visual field is mapped out by 1 million different 10,000 neuron pixel groups. The hypothesis is then essentially that these 1 million groups are wired together in a very similar way. Consider just the color blue that we are sensing at one specific point in our visual field, which was triggered by a region of blue cone cells in the retina, and processed by the corresponding pixel group in the cortex map. This sensation of blue must be equivalent to (that is, it actually is) some specific pattern of neurons firing within the group of 10,000, perhaps a particular subset of 1,000 neurons. Furthermore it isn't the organic basis of the neurons that is so important in producing the conscious phenomenon of seeing blue, rather it is the pattern formed out of the neurons, essentially a connected graph with the nodes being equivalent to the neuron bodies and the paths the dendritic connections between them (although the specific chemistry of the neurons is of course crucial in evolving the brain states forward in time...). It is the topology of this particular connected graph that is the conscious sensation of the color blue. Furthermore, this is why I posit that the structures of all the million different pixel groups must be very similar: the sensation of blue across the top half of the visual field is the same, so there must be the same connected graphs (composed of firing neurons) in each of the 10,000 neuron pixel groups. And within the 10,000 neuron groups there would be other structures that enable the sensation of other colors, and of light levels and depth and movement and so on (with intermediate colors achieved by activating several neural graph subsets simultaneously).
While currently it would probably be quite challenging (and somewhat dangerous) to actually do experiments on this, it seems that in principle you should be able to test the idea. That is you would cut a very small hole through the skull to the visual cortex of a volunteer, and then while the patient is awake you can show them a slideshow of pure colors, and one should be able to detect the same pattern of neurons being activated over and over again all across their visual field, then switching to another consistent pattern when another color is shown on the screen and so on. In fact, prehaps you could show, say, a small red circle moving slowly across a background of white, and then you should be able to track the connected graphs forming the sensation of red moving around in the visual cortex.
Unfortunately, I'm pretty sure we're still a ways away from having sensitive enough instrumentation to be able to do this. But once we do it would be immensely cool to be able examine the different connected graphs that are equivalent to the sensations of different colors. Of course, it may well be that the color graphs for different people are quite different, so that perhaps your sensation of the color orange is like my green, or (and I bet this is more likely) they are totally different, so that each of us can't even conceive of the other's (at least with our current, upgrade-incapable biological brains). Indeed this brings us to the question of how the topology these graphs is programmed in the first place. My bet is that the genetic code does not specify a particular connectivity for each different color, but rather gives a general algorithm that will converge on some solution. One imagines a young baby looking up at a colorful mobile while the crib, and as the distinct brightly colored objects rotate through the baby's visual field, the baby sees each object shimmer through a dizzying rainbow of different colors, but over the weeks each object will converge to one stable color, so that the red of a plastic toy boat looks the same anywhere in the visual field (the neural networks group at Texas has done some interesting work on self-organization algorithms in the visual cortex along these lines).
So the visual field is mapped out by 1 million different 10,000 neuron pixel groups. The hypothesis is then essentially that these 1 million groups are wired together in a very similar way. Consider just the color blue that we are sensing at one specific point in our visual field, which was triggered by a region of blue cone cells in the retina, and processed by the corresponding pixel group in the cortex map. This sensation of blue must be equivalent to (that is, it actually is) some specific pattern of neurons firing within the group of 10,000, perhaps a particular subset of 1,000 neurons. Furthermore it isn't the organic basis of the neurons that is so important in producing the conscious phenomenon of seeing blue, rather it is the pattern formed out of the neurons, essentially a connected graph with the nodes being equivalent to the neuron bodies and the paths the dendritic connections between them (although the specific chemistry of the neurons is of course crucial in evolving the brain states forward in time...). It is the topology of this particular connected graph that is the conscious sensation of the color blue. Furthermore, this is why I posit that the structures of all the million different pixel groups must be very similar: the sensation of blue across the top half of the visual field is the same, so there must be the same connected graphs (composed of firing neurons) in each of the 10,000 neuron pixel groups. And within the 10,000 neuron groups there would be other structures that enable the sensation of other colors, and of light levels and depth and movement and so on (with intermediate colors achieved by activating several neural graph subsets simultaneously).
While currently it would probably be quite challenging (and somewhat dangerous) to actually do experiments on this, it seems that in principle you should be able to test the idea. That is you would cut a very small hole through the skull to the visual cortex of a volunteer, and then while the patient is awake you can show them a slideshow of pure colors, and one should be able to detect the same pattern of neurons being activated over and over again all across their visual field, then switching to another consistent pattern when another color is shown on the screen and so on. In fact, prehaps you could show, say, a small red circle moving slowly across a background of white, and then you should be able to track the connected graphs forming the sensation of red moving around in the visual cortex.
Unfortunately, I'm pretty sure we're still a ways away from having sensitive enough instrumentation to be able to do this. But once we do it would be immensely cool to be able examine the different connected graphs that are equivalent to the sensations of different colors. Of course, it may well be that the color graphs for different people are quite different, so that perhaps your sensation of the color orange is like my green, or (and I bet this is more likely) they are totally different, so that each of us can't even conceive of the other's (at least with our current, upgrade-incapable biological brains). Indeed this brings us to the question of how the topology these graphs is programmed in the first place. My bet is that the genetic code does not specify a particular connectivity for each different color, but rather gives a general algorithm that will converge on some solution. One imagines a young baby looking up at a colorful mobile while the crib, and as the distinct brightly colored objects rotate through the baby's visual field, the baby sees each object shimmer through a dizzying rainbow of different colors, but over the weeks each object will converge to one stable color, so that the red of a plastic toy boat looks the same anywhere in the visual field (the neural networks group at Texas has done some interesting work on self-organization algorithms in the visual cortex along these lines).
// posted by Travis Garrett @ 3:00 PM 
