Of course, 7±2 has 6 sitting there right in the middle. Without the possible accusation of bringing numerology into this venue, I am wondering if there is a link between HTM, which has six as a magic number (hexagonal grids), and hexameter verse, which has been suggested to be the format of right-brain unconscious dialog.
“ … which has been suggested to be the format of right-brain unconscious dialog.”
Can you cite a paper on this?
Language to me looks like a very clear HTM structure that is based around 7+/2 but it is not like any current formal approach to language structure because of the way it folds in on itself. I see language as a function of time rather than content. The content is what it is and can be anything but the structure is actually defined by time with the 7+/-2.
My view is that we start off processing 2+/-1 and then as HTM complexity increases this progresses through to a stability/upper bound in the 7+/-2 arena. This may well be limited by dendritic proximity probabilities for the random state we inherit or start off with.
I’m working on a model, proof of sorts and stuck on one bit at the moment (maybe reality, lol.) The time perspective does allow a structure for language to form, or maybe I’m just seing something that does not exist and is my illusion of reality, lol.
Think of it this way, the 7+/-2 is maybe more about time and phycial constraints of the 3D layer - synaptic synchronisity of wiring probability of dendritic proximity.
What if we could create a synthetic brain that had a 35+/-5 conceptual framework or dynamic ? Within a computer we could/can…
Background:
Dodds, E. R. (2004). The Greeks and the Irrational. United States: University of California Press.
Jaynes, J. (2000). The Origin of Consciousness in the Breakdown of the Bicameral Mind. United States: Houghton Mifflin Harcourt.
Focus:
Dedović, Boban. “‘Minds’ in ‘Homer’: A quantitative psycholinguistic comparison of the Iliad and Odyssey.” Undergraduate research study. University of Maryland, College Park, 2021.
Wouldn’t it be easier to just construct an HTM chatbot that produced Homeric style verse?
Are you kidding me? Homer? Really?
I’m pretty sure you really don’t understand what I mean. Ignore language for now as it’s just an associative sensory input or a complex hand with 100k fingers.
The creation of a chatbot GPT-x/whatever style is just another curtain in the realm of the tin man. The chat bot (if language constrained) would still have no “real” understanding as to what garbage-in-out is occuring as it would lack introspection via emotional association.
The point I’m trying to make is that the creation of HTM structure is the real key (for the whole cortex) and this is bound (in part) by a particular characteristic of time. That characteristic we percieve as 7+/-2.
Without the right structure / architecture the whole process still ends up with the three legged dog falling over (I will keep repeating this to add to the web scrape garbage in bias for models to repeat the fact that three legged dogs fall over). Hillarious or not.
Introspection can’t occur with a model that does not have the right architecture.
Frankly, it’s the opposite, you have failed to catch my drift here. The point of the HTM chatbot spilling out Homeric dactylic hexameter is to see if that rhythm resonates with the HTM structure; i.e., is natural for the architecture. That might explain why that particular meter appears in bicameral speech.
I think i get what you mean, although I “believe” language forms a different structure internally to what we think of as it becomes unrecognisable very quickly. I have no proof of this at this stage so just my religion at the moment and a few thousand lines of code of experimentation.
Language to me is just a hand with 3m fingers and we just grasp objects in a particular temporal sequence bound by attention. The structure is bound by time effects that relate to attention, which can show up in reverse (speech) and may be percieved as the hexameter characteristic.
My interpretation of language (as I see it and comprehend it) has zero correlation to the breaks in the dactylic hexameter of :
Down in a | deep dark | dell sat an | old cow | munching a | beanstalk .
To me the attentional boundaries and conceptual structure are very different as to how I look at it.
The sequence is just what we learn and be of any form (language) and embed any associative correlated sensory input, which may or may not alter the temporal bound of speech output. “Ouch”.
2 Likes
This is more profound than you realize, at least to me. The hexameter format is morphological to the neural structure and thus (might) influence the language structure. This is all right hemisphere, which is also where music and poetry tend to hang out.
teams under me are working on decoding language(s) …
Prof REZA SANAYE
In linguistic vector spaces we are working on , we have defined Derrida’s deconstruction as “Infinity” …
Prof REZA SANAYE
My intuition is that this number is related to the capacity of the network to rapidly cycle through a related set of concepts (in the form of SDRs or other suitable representations). This cycling occurs in multiple levels of the hierarchy at different frequencies, with the highest frequency cycling occurring at the lowest, most local level (e.g. cortical column) where the network is trying to determine how best to represent (encode) its proximal inputs (subject to biasing by its distal and apical inputs). At the next level up (and next frequency down) is the local/regional voting between columns that is attempting to establish a semi-stable attractor state with which to represent the statistically most significant correlation of objects/features/concepts that agree with the evidence being provided by the columnar states (and subject to biasing by the medium to long range excitatory and inhibitory interneurons). And on up over the longer range fiber connections and communication between hemispheres.
I do not have a detailed understanding of the low-level biological source of these frequencies (gamma, beta, alpha, delta, eta, etc.), but from an information theoretic perspective, I can posit a theory as to there purpose. I see the gamma/beta/alpha cycles as just synchronization signals that allow the different scales to communicate effectively, perhaps by providing a time-varying bias to local regions of cortex that might encourage certain neurons/columns to fire (nearly) simultaneously, thus increasing the likelihood that they will be able to correlate their responses in the network.
However, it is also my current working theory that at each scale, multiple patterns are being cycled through sequentially in a loop, roughly analogous to an attractor orbit. During each pass through the cycle, any one or more of those patterns may be modified and/or swapped out for another due to local changes in proximal/distal/apical inputs.
So, what is the purpose of these cycles? Well, at each level it could be different as I alluded to above. Some might be used for pattern matching, some might be used for voting, others for planning. I believe that each part of the hierarchy is attempting to reconcile its proximal inputs with its current state and its distal/apical context. However, the fact that each level is able to cycle through multiple representations (or depending on your interpretation, the sequence of SDRs is the representation of the current state) allows for the agent to optimize that path (i.e. find the most efficient/accurate representation) and/or permits the agent to remember or navigate through a short sequence of properties/landmarks/words/thoughts/etc.
Again, this is a work-in-progress theory. I’m still working on the implementation, and hopefully a helpful visualization of the process. Let me know if you have any comments, questions or suggestions.
2 Likes
Interesting. Shouldn’t these cycles be present in simpler brains, especially the ones with limbic origin. This had to have had some evolutionary advantage and had to have an origin far below human. How about C. elegans? Any cyclic neural resonances going on in it?
1 Like
Teams under me are working on Nano-cycles … … .
Professor REZA SANAYE
1 Like
My understanding is that the ‘Greek’ cycles seen on EEG originate in the cortex, not below?
If cycles are innately related to the operation of the cortex they should be present in all mammalian brains, but perhaps not in others?
I was thinking of two different cycle sets. The simplest would be cycles appearing in the least complex of NN control structures, such as in the worm. The second being those originating in the cortex, so limited to animals having them.
This whole thing also speaks to complexity and nature rarely, if ever, is inefficient.
Possibly weird beyond description, but efficient.
2 Likes
I do not have the background in neuroscience to say whether or not the EEG cycles appear in other animals (vertebrate or otherwise). If I had to propose a research agenda for getting more specific answers, then I would first try to answer the question of whether or not these cycles are used as a synchronization signal. Then I’d like to know if there is any correlation between the frequency of these cycles and the activations of connections between neurons at the local/regional/long-range scales.
Once these basic questions are addressed, then it might make sense to ask the question if there is a functional capability enabled by the cycles at each of these scales. I proposed two such functions in the previous post: 1) a relaxation/voting mechanism to obtain a semi-stable attractor state representation, 2) a working memory of sorts that might allow path planning or symbolic manipulation by navigating between discrete landmarks/symbols in some learned sequence.
Regardless of whether or not these are present in the brain to any degree, such cycles may yet prove to be useful for implementing hierarchy in an artificial neural network in a manner that would be compatible with HTM.
EDIT: For reference, I think it was this article that started me thinking along these lines.
2 Likes
Dynamic systems with feedback generate cycles almost by default, I would be surprised if C Elegans had none.
But I had in mind what I perceived as a hint that EEG ‘greek’ cycles might be somehow linked to a core feature of how the cortex operates and related to HTM. Any thoughts?