Not so long ago, I listened to a 99% Invisible podcast episode "The Country Of The Blind" which was an interview with Andrew Leland who wrote a book of the same name, based on his experience with gradually going blind due to a medical condition.
The inevitability and permanence of his upcoming full blindness was never in question.
It's amazing to see that there's a non-trivial chance that blindness may be curable in the future.
Eye movement photography during reading has been used for years to understand the relationship between word recognition processes and cognitive control of tracking lines of print. For example, the number of milliseconds the eyes fixate on a section of a line before moving to the next section in a saccadic jump, how often and in what context the eyes regress to revisit sections processed seconds ago, etc. Theoretically, the eyes can process a very tiny portion of foveal vision capturing 7-12 character sections; so letter strings are perceived not as words but as a set of characters to be discernible as words later. The perception of characters happens in 50 ms and then pyscholinguistic loops take over automatically to push the data into another layer of processing until lexical access in long term memory is achieved at 250 ms. It’s intriguing that preprocessing of frequencies could be done at the eye and automatic cognitive processes are activated at 50 ms. Word perception is different from raw perception of light waves because as far as I know the image of a dog or a tree takes much less than 1/4 of a second to register in consciousness. We know that color, for example, interferes with processing words. The Stroop effect named after the researcher who discovered it tells us that lexical access for the word “green” printed in red letters slows lexical access. Do you know if retraining the ganglia cells to be sensitive to light frequencies in the way you described using goggles to fine tune the firings has any effect on ability to read? Is there something peculiar about visual word perception of phonetic script that sets it apart from learning to perceive objects and shapes showing up in this new research?
A difference between human reading and AI reading at least processually is the use of the eyes to process words through psycholinguistic activity below the threshold of consciousness in humans. I’m not sure what AI does with letter strings, but I’m thinking it’s probably not aimed at lexical access in the same way.
I enjoy your writing and always wake up and wonder:) Thanks!!!
The short answer is that with optogenetics, it's too early to say for sure. So far, the focus has been on basic visual perception, such as detecting light and dark or recognising simple shapes and movements, rather than the detailed processing required for reading.
But if I were to speculate, I would guess that with higher doses and improved technology, reading might happen fairly normally. Initially, the text might need to be large, with high contrast between the letters and the background, to account for the reduction in the resolution of the input from the eyes. My reasoning is that neurons in V1—the first brain area to process visual information—respond strongly to specific oriented stimuli with high contrast, but the area of space they represent is small. So V1 neurons are responsible for giving us fine details. If the incoming information from the eyes is low resolution, V1 might require larger, high-contrast text to ensure that the visual system can adequately process and recognise the letters.
Reading comprehension and word perception happens later on in the processing stream (in language areas like Wernicke's area). If the early stages of the visual system are not able to process the fine detail in letters, they can't pass them onto these later stages.
Regarding the difference between human reading and AI reading, I think many would agree. Humans process words through psycholinguistic activity—most of which happens unconsciously. While AI processes letter strings using algorithms and pattern recognition techniques, which are fundamentally different from human reading processes. But some might argue that the brain is also engaging in a form of pattern recognition. Humans and AI seem to do it in different ways, but the question many cognitive sciences have is -- how different?
What are your thoughts about this. After the signal has reached V1 in the brain, the brain begins to send data to auditory processes, linking graphemes to phonemes. Shortly after, the brain signals the speech muscular to pronounce a word. When we read silently, there is no vocalizing, but physical indices tell us that some degree of subvocalization occurs (tiny movements of the tongue). In theory, when we learn vocabulary in early childhood, we store it phonologically. Later, visual perception builds upon these traces, and we don’t really read silently ever—our brain can’t make lexical access without shuttling through sound. Monitoring of the incoming letter stream causes readers to regress when something makes no sense—and this regression can happen long before semantics have been activated. Moreover, many readers reveal changes in semantics even when reading silently based on how words “sound” when reading silently. There’s also a kinetic element that lingers from early childhood—a slight swaying or rocking when reading a word like “bicycle.” Language and reading are linked to the physical body where meaning can be stored in the musculature. I really am agnostic in regard to how this matters in terms of human vs machine reading but it really interests me for some reason
Brilliant article. I hadn’t realised such progress had occurred. Thank you.
Thanks John! I'm glad you enjoyed it.
Not so long ago, I listened to a 99% Invisible podcast episode "The Country Of The Blind" which was an interview with Andrew Leland who wrote a book of the same name, based on his experience with gradually going blind due to a medical condition.
The inevitability and permanence of his upcoming full blindness was never in question.
It's amazing to see that there's a non-trivial chance that blindness may be curable in the future.
Thanks, Daniel! Yes, it's pretty amazing what might be possible. It's an area of research to keep your eye on.
That podcast episode sounds fascinating, I'm adding it to my list.
Good article. Impressive research!
Eye movement photography during reading has been used for years to understand the relationship between word recognition processes and cognitive control of tracking lines of print. For example, the number of milliseconds the eyes fixate on a section of a line before moving to the next section in a saccadic jump, how often and in what context the eyes regress to revisit sections processed seconds ago, etc. Theoretically, the eyes can process a very tiny portion of foveal vision capturing 7-12 character sections; so letter strings are perceived not as words but as a set of characters to be discernible as words later. The perception of characters happens in 50 ms and then pyscholinguistic loops take over automatically to push the data into another layer of processing until lexical access in long term memory is achieved at 250 ms. It’s intriguing that preprocessing of frequencies could be done at the eye and automatic cognitive processes are activated at 50 ms. Word perception is different from raw perception of light waves because as far as I know the image of a dog or a tree takes much less than 1/4 of a second to register in consciousness. We know that color, for example, interferes with processing words. The Stroop effect named after the researcher who discovered it tells us that lexical access for the word “green” printed in red letters slows lexical access. Do you know if retraining the ganglia cells to be sensitive to light frequencies in the way you described using goggles to fine tune the firings has any effect on ability to read? Is there something peculiar about visual word perception of phonetic script that sets it apart from learning to perceive objects and shapes showing up in this new research?
A difference between human reading and AI reading at least processually is the use of the eyes to process words through psycholinguistic activity below the threshold of consciousness in humans. I’m not sure what AI does with letter strings, but I’m thinking it’s probably not aimed at lexical access in the same way.
I enjoy your writing and always wake up and wonder:) Thanks!!!
What a fascinating comment and questions!
The short answer is that with optogenetics, it's too early to say for sure. So far, the focus has been on basic visual perception, such as detecting light and dark or recognising simple shapes and movements, rather than the detailed processing required for reading.
But if I were to speculate, I would guess that with higher doses and improved technology, reading might happen fairly normally. Initially, the text might need to be large, with high contrast between the letters and the background, to account for the reduction in the resolution of the input from the eyes. My reasoning is that neurons in V1—the first brain area to process visual information—respond strongly to specific oriented stimuli with high contrast, but the area of space they represent is small. So V1 neurons are responsible for giving us fine details. If the incoming information from the eyes is low resolution, V1 might require larger, high-contrast text to ensure that the visual system can adequately process and recognise the letters.
Reading comprehension and word perception happens later on in the processing stream (in language areas like Wernicke's area). If the early stages of the visual system are not able to process the fine detail in letters, they can't pass them onto these later stages.
Regarding the difference between human reading and AI reading, I think many would agree. Humans process words through psycholinguistic activity—most of which happens unconsciously. While AI processes letter strings using algorithms and pattern recognition techniques, which are fundamentally different from human reading processes. But some might argue that the brain is also engaging in a form of pattern recognition. Humans and AI seem to do it in different ways, but the question many cognitive sciences have is -- how different?
What are your thoughts about this. After the signal has reached V1 in the brain, the brain begins to send data to auditory processes, linking graphemes to phonemes. Shortly after, the brain signals the speech muscular to pronounce a word. When we read silently, there is no vocalizing, but physical indices tell us that some degree of subvocalization occurs (tiny movements of the tongue). In theory, when we learn vocabulary in early childhood, we store it phonologically. Later, visual perception builds upon these traces, and we don’t really read silently ever—our brain can’t make lexical access without shuttling through sound. Monitoring of the incoming letter stream causes readers to regress when something makes no sense—and this regression can happen long before semantics have been activated. Moreover, many readers reveal changes in semantics even when reading silently based on how words “sound” when reading silently. There’s also a kinetic element that lingers from early childhood—a slight swaying or rocking when reading a word like “bicycle.” Language and reading are linked to the physical body where meaning can be stored in the musculature. I really am agnostic in regard to how this matters in terms of human vs machine reading but it really interests me for some reason