Brain Physiology, Cognition and Consciousness group: topic
This is a public discussion board
From Neuron to Memory System: How Memory Might Work
Graeme Smith
Friday, 12 June 2009 04:34 UTC
How does memory work?
We think we know, based on thousands of years of research into it, but there are still indications that we might be completely wrong. Scientists created computers with the best scientific approach to memory based on current scientific theories, in the 1940’s, and they got it wrong. Most neuroscientists today will tell you that the computer is nowhere close to a good model of memory, yet many of the attitudes that informed the decisions on how to make computer memories remain unchallenged. Our popular theories of how memory works were laid down in the middle ages, by people who thought memory was a fluid, and that the brain was a reservoir. Somehow they thought, we opened valves, and let the memory flow from the brain to where it was needed. I call this a demand model of memory because you demand the memory by opening the valves. In computing terms we have replaced the fluid theory of electricity but retained the idea of current, and at one time tubes which were replaced with transistors were called Valves. So computer memories could be seen to be operating in a demand memory model as well.
As Jerry A. Fodor said in The Mind Doesn’t Work That Way!:The scope and limitations of Computational Psychology Neither Phenomenal approaches such as Neural Networks nor strict computational approaches such as Truth Preserving Functions, seem likely to achieve a suitable computational model of consciousness.
As a researcher in Artificial Consciousness, my main thrust is to eventually get support for my own Artificial Consciousness to be built. However to get there, I had to start with a model of memory. I chose to hedge my bets, to begin with a Neural Model, and add Functional support to that model, where it was needed. Further I wasn’t going to limit myself to Truth Preserving Functions, where Soft computing would be more effective. This type of approach to consciousness is called a Hybrid approach. To start off, however I wanted a model of memory, and one of the things I had to do, to get there was overcome Fodors assurance that no-one had yet developed a phenomenal version of a demand memory.
Those who are stuck on William James idea that phenomenal systems can’t be built based on what we know about the brain, have problems with my use of the word phenomenal in the way that Jerry Fodor used it in his book. They like to look for exotic things like QM entanglement to explain the fact that some things just seem indivisible. I however think that there are indivisible elements in the memory but that it doesn’t matter because the mind doesn’t try to divide them, it finesses the system, without that effect.
Fodor’s claim is not without merit, phenomenal systems are definitely easier to build out of neural networks, than demand memories. However, David Marr defined a type of memory he thought might help explain the cerebral neocortex back in 1970, and by understanding what his work uncovered, and brushing off some of the unfortunate assumptions of his day, I think I have undercovered one of the hidden biases that is keeping us from understanding the way the brain, or at least the memory works. And that bias is the assumption that explicit (demand Model) memory, is the natural state of the memory system, and that anything that doesn’t fit the model must be an add on to the basic explicit memory model. So we expect Implicit Memory to somehow be an add-on to the essential demand memory, and we are wrong.
What Marr described in 1970 is a type of memory more basic than explicit memory when implemented in neurons. To computer guys like me, this seems counterintuitive. How can anything be more basic than a good demand memory, we can implement dynamic ram with a capacitor, a resistor, and a transistor, how much more basic can you get? But what we keep forgetting is that neurons don’t work like transistors, there is no technological overlap. Within the logic that makes neural networks work, demand memory is much more expensive to build than a simple content addressable memory. That is what Marr claimed his 4 layer CODON was, was a self-classifying content addressable memory.
To understand why this might be, we need a little theory. Although in the 80’s and 90’s the connectionist school was over-run by the Parallel Distributed Processing guys from the A.I. discipline, and so we have to take some of its theory as being deliberately misleading to steer people away from trying to understand real neural systems, The basics are fairly equivalent. A neuron is first of all a cell, it has to survive like other cells, by absorbing nutrients and getting rid of wastes. However at some point in the evolution of animals, neurons gave up some of their survival functions to a helper cell called a glial cell, and converted those functions over into mechanisms for transferring information between cells.
The Parallel Distributed Processing guys figured what was important to this communication was the firing of the cell. Sorry that is a misconception, firing just speeds the process of transferring information up, it is not the only mode of transfer, nor is it the most important one for understanding natural systems based on nerve cells. However because the PDP boys wrote the manuals for the industry, they got to tell 20 years of modelers what to think. As a result of mistakes like this, they set the neural modeling of Natural Neural Networks back, to the point where Gary Marcus, has clearly stated that he feels it necessary to do a hostile takeover of the connectionist school.
One approach that the PDP guys thought was dangerous was David Marr’s attempt to use probability Mathematics to define a circuit that was made up of heterogeneous neurons. Despite the fact that they did not have the ability to model heterogeneous groups of neurons, a direct attack on his claiming that the heterogeneous group he called a codon was a content addressable memory, was made pointing out that the model did not exibit wave-forms similar to the real cortex.
This attack, assumed that Marr’s circuit had to have frequency artifacts similar to the real cortex in order to suggest the role of content addressable memory for the cerebral neocortex. Well Marr was a pioneer in the field and he got some things wrong, and others even more wrong, but it was a seminal stage in the science, and what he got right is more important.
Marr was able to use probability mathematics to analyze heterogeneous networks of neurons and predict their function. Nobody before that even tried, and after the way he was treated few after him had the nerve to try again.
I am not going to start spouting probability equations, if only because I don’t follow his math. But the PDP guys called a neuron a processor that had input, process and output capabilities. I however look at it from a different perspective I see a neuron that has storage, processing, and transfer capabilities. And when I look at the 4 layer cortex model that Marr worked with, I also see a content addressable memory, but I think that perhaps he expected too much equivalency between neurons in the connections department, and was sadly dissapointed before his death in 1980. So I wonder at how much self-classification the system is doing.
If you look at neurons from my viewpoint you begin to see that the shape of the neuron has a logic, that memory neurons have lots of synapses, and opportunistic process growth, that processing neurons tend to have complex and often bushy or mossy dendrites, and that transport neurons tend to have virtually no division at the dendrite level and form long thin neurons.
If you look at the type of organs that are produced, Memory Neurons tend to form tissues, Processing Neurons tend to form globular organs, and transport neurons tend to form fibrous bundles. In other words the shape of the neuron, indicates it’s function, and the shape of the organ might indicate to some small extent what type of function it performs.
Looking at Marr’s 4 layer model, we see that the first layer is processing neurons, the second and third layer are memory neurons, and the fourth layer is dominated by processing neurons again. The whole structure is probably a memory role because it is a tissue.
Stimuli coming in at the first layer, are processed affecting their storage in the second and third layers, and the 4th layer probably is mostly control neurons, and in some cases inputs of senses that have been directed through the thalamus. These latter signals condition the general inhibitive environment caused by the Martinotti cells and thus reduce the inhibition at the first layer encouraging the layer 2/3 neurons to fire.
This simple memory circuit can be seen to be designed to respond to either signals from layer 4 or layer 1, and the first layer is used to train the second and third layers to respond to patterns of stimulus. Given a stimulus each pyramidal cell in layer 2/3 decides whether or not to fire, and this type of system is exactly what is needed for a content addressable memory.
It is also exactly wrong for an explicit memory suggesting that it must be the explicit memory that is the add on. Indeed when we look at the micro-architecture we see that a type of tissue called Allocortical tissue makes up the tissues at the bottom of the Sulcys and Divides of the brain. and since Sulcys and Divides are a later architectural configuration there might be an indication that the six layer tissue called isocortical tissue, that pushes the gyruses up away from the sulcys and divides, was a specialization of the brain.
-
Replies
Jump to resultsResults
-
Well if I read the last few postings right, what we are dealing with, in Sossin’s work, is the interface between the dynamic synaptic storage mode, and the short-term biochemical pathway that converts signals into changes in weight in the Neural Model. The missing piece I think, is some link between the protein that the synapse generates, and the membrane replacement mechanism. If someone gets around to proving a bichemical link, then we will understand how signals reinforce the weights of synapses.
The fluorescent dye method of biochemical tracing is a powerful tool because it allows us to mark the presence of chemical reactions that otherwise would hide in the general chemical soup, and be undetectable.
An interesting question based on my neural model, is how many different types of proteins are synthesized, and are they stripped from the synapse during the process of membrane replacement, or do they survive the replacement process. My model suggests we need both types of protein tags.
Ones that are stripped away, for general adjustment of weights, and ones that are more permanent for longer term memory processes within the cell. -
Well Ottmar you have some interesting questions there, and I doubt I can answer them all to your satisfaction, if only because we are still learning the answers.
As to hard evidence that memories are stored in the brain, I personally have none, except the memories that are stored in my own brain. However, other scientists have shown that mice learning in a particular rich environment actually have heavier brains right after learning, than the ones that learn in a deprived environment. Now this seems to suggest that there are actual physical changes in the brain as part of learning. Unfortunately these physical changes that can be detected by weight do not always remain detectable for the life of the brain, suggesting that there are also mechanisms that absorb such weight, possibly as a process of consolidating multiple memories into a smaller number of memories.
So lets look a little closer to see what science has been able to find out about the processing of memories. I have suggested that there are actual physical changes in the nerve cells associated with learning, at the synapse level, and these changes have to do with an increase in the number of ion channels that each synapse has. So how would we go about testing such a suggestion? We can’t see ion channels, they are too small, the only way we know about them are by measuring the ion currents they form, and by biochemical analysis of the synaptic sensitive patch.
So, to prove that there are more ion channels, we can only do a couple of things, we can measure an increase in the size of the ion current from a cell that has learned something, we can measure the fluorescence of a cell that has a fluorescent marker associated with a protein used in ion channels, or we can measure the chemical contents of a puree of sensitive patch, to determine if there is a higher concentration of ion channel proteins in the sample of a cell that has learned something or not.
Two of the three of these experiments have been done with positive results, we know that the ion current of a cell increases when it has learned something, and we know that the proportion by weight of ion channels has increased with cells that learn something. Someone might be doing the middle experiment somewhere in some laboratory as we speak, but as far as I know it has not yet been done in a way that obviously shows ion channel concentration being larger in the more active synapses of the cell.
Now it is important to note that synapses are only one way that nerves store information, another way is by opportunistic connections between neurons. We can actually see the fibrils of a neuron grow, and attempt to connect between two neurons that are active at about the same time. there is no doubt that there are actual physical connections between neurons. Actually the problem is that there are too many of them. There are billions of neurons and many of them have thousands of connections, multiplying the number of connections into the trillions. The number of fibrils that grow, are actually much larger than the number that are needed within the brain for storage, and so there is also a mechanism by which some of the fibrils are absorbed back into the cells they grew from. It is partially because so many fibrils grown during learning, and so many are reabsorbed that the weight of the brain measure of learning is so useless at proving that memories stay in the brain.
However there is no doubt in my mind, that memories are stored in the brain and in a number of different ways. Sossin’s earlier work suggests that there is actually a number of biochemical pathways by which one form of memory in the brain is transformed into another form of memory whether biochemical, or physical connection. We simply have not yet spent the time necessary to track them all down and explain them to each other.
When you ask me to prove memories are stored in the brain by regurgitating them in some manner other than the time honored way of taking someone’s word for the fact that they remember, you bring me up against one of the most frustrating parts of working with the human brain, it’s complexity. Quite simply, we haven’t yet learned how to translate the signals from the brain back into the memories that produced them. There is a chaotic aspect to neural connections that defies mapping. Then there is a dynamic aspect to the way that neurons store data, that also defies mapping. between them they pretty well eliminate the opportunity to “Detect specific Memories” at the cellular level. But this doesn’t mean we can’t detect them at higher levels of organization, in fact recent work with arrays of contacts has shown that we can detect activation at the neural group level, unfortunately research has also shown that neural groups are interchangeable, and so the same neural group in one brain does not store the same data as the same neural group in the next brain. The brain is wonderfully unstandardized, and as a result, is much more difficult to translate back into data, than an equivalent circuit in an electronic device would be. However do not despair, electronic devices are getting more complex, and learning how to take a page from the brain, in order to make the same circuit work differently in different devices as well. For instance it is now possible to design an electronic memory that substitutes memory cells that are working for memory cells that have failed, thus changing the location in memory of a particular byte, without changing the address that the computer expects to find it at.
How are we conscious of our memories? Ah, well that is the million dollar question, I don’t know how you expect me to be able to answer it now when we probably won’t have the answer for 10 years or so, but let me begin to pave the way.
We aren’t conscious of our memories until we tell ourselves we are conscious of our memories. Consciousness involves feedback. A memory passing through our implicit memory system goes unremarked. We can navigate from work to home almost instinctively, and only when we get home realize that we have no memory of how we got there. So obviously not every memory we have, gets to be consciously accessible.
Memories are also either stimuli from outside the brain, or generated within the brain, including imagination and derivative forms of memory such as associations and interpretations of original data. An example is the bunny rabbit/duck illusion in which you see either a bunny rabbit or a duck depending on which interpretation/orientation the picture is presented as.
Another example is the short/long line illusion where depending on which way the angled lines run into the horizontal line, it seems either shorter or longer. Obviously it is not the stimulus that makes the line longer or shorter but the interpretation of the stimuli, by the brain.Again we are not conscious about the derived portions of the data until our attention is drawn to the fact that they are derived, when someone tells us our first assumption that the one line is longer than the other is wrong.
We have to literally struggle with our minds to accept that the two lines are the same length, at that point their length becomes conscious. So somewhere between the time the lengths were derived and the time we struggled with the impression we had that one line was longer than the other, we became conscious of the lines. What defined the difference? I suspect we will find that it was a feedback loop, that projected the struggle back into our minds so that we could tell ourselves that we were aware of it.
As to emotions, emotions are two things, signals from our limbic system to our bodies that prepare the bodies for particular events, and the impressions on our minds that those preparations create.
-
To recap, the memory that comes after the bottleneck, which I call explicit memory and some others still call implicit memory has the following characteristics.
1. The Content is stored in implicit memory and output is via data-cloud, What I used to call phenomenal, but have been told means something else.
2. The representation of the data after the bottleneck is a CHUNK, essentially a list of Place-codes, or addresses of Neural Groups, that are defined by the Layer 5 Pyramidal Neurons which form the AXIS of mini-columns according to Dr. LaBerge, and which have the ability to preactivate the neural groups in short or longer term states depending on the length of the Layer 5 Apical Dendrites.
3. A Chunk, is equivalent to a Functional Cluster in the implicit memory which has been determined represents a zone of salience within the input stimuli, which the body thinks is important, and has linked using a GSO.
4. The only way to recover the actual memory associated with the chunk, is to rehearse it, which means presenting the chunk to the Thalamus, and having the thalamus pre-activate the neural groups in the cerebral cortex.
5. The output of which is still a data-cloud and thus cannot be divided.
I used to call this Phenomenally Explicit Memory, because of its dependency on the implicit memory for output. But right now I am a little shy about using that term for it, since phenomenal doesn’t seem too popular a description.
So the next problem, is, if you wanted to make a demand memory, and you can’t demand content, because the output is what I used to call a Quale, because it is so indivisible, what could you possibly do with this type of memory?
The answer as I have alluded to before, is to change the CHUNK rather than changing the Data-Cloud. The chunk address it seems is routed from the PFC through the Nucleus Accumbens in the Striate Cortex, which in turn, can be linked to the Nucleus Reticularis Thalami that surrounds the thalamus, and has an inhibiting effect on signals from the thalamus. What this means is that an arbitrary number of CHUNK mini-column addresses, can be suppressed, and thus the data-cloud created can be smaller than the original chunk addressed Functional Cluster.
This has to be an arbitrary decision at first, if only because there is no way to predict whether or not the sub-chunk arrangement will be more or less salient. The Striate Cortex can build up facility with addressing important sub-elements of a recognized chunk, only if it knows which sub-chunk arrangements are more useful. The salience measure gives us this measure of usefulness, and in doing so, encourages the thalamus to give the sub-chunk array of addresses it’s own GSO, so that the ACC can select between them.
By passing the Sub-Chunk back through the bottleneck depending on its salience, the brain can narrow it’s focus to a specific memory eventually, and thus begin to build an index of where the memories are stored. The combination of knowing the salience of each CHUNK, and an index to the location of memories using chunk based addressing, gives us an opportunity to process specific memories.
-
Dear Graeme,
I have been trying to follow your memory scheme here and on your web site and I don’t really see just what insight it is that you think you are adding here. Your recent posts seem to claim that individual memories are held in individual cortical columns which are directly addressed by some cloud scheme which accesses them as data structures as in discrete state machines.
Are you saying that memories are stored in individual columns and then projected across specialized topical regions?
What about the subtemporal region which virtually all of the literature I have seen proposes as the long term memory storage region. How does this fit into you scheme?
Quite frankly, I think you need to first explain your scheme in the abstract, i.e., without all these direct references to cortical structures. IOW, how could it be built?
Michael Baggot
-
To get back to the abstract, this is meant to be implemented as a Hypbrid Scheme, which means the simulation starts out as a neural network. The problem is that we have to implement the network so that it takes into account Heterogeneous groups, (Groups of different types of cells) rather than the standard homogeneous group neural networks that are part of PDP. To help with this, I have defined different types of neurons by the way the dendrites, synapses and axon fibrils form. These characteristics then cluster the types of neurons into neurons that are specialized for one of three roles, Storage, Processing, and transport. Much of the rest of the structure of the neuron may have to do with the form the brain organs take, which tends to be tissues for memory, globular elements for processing and fibers for transport. Some forms of neurons lend themselves better to tissues than others. For instance in a book I am recently reading about Architectronics, a class of storage neurons called Modified Pyramidal Neurons exist, (Some of which may predate the pyramid) that often do not have a distinct basal plane to their Dendrites, but may for instance produce multiple planes of dendrites at the soma. Instead of creating a natural disk of radial fibers, such neurons would produce a sphere of radial fibers, which would complicate the formation of a Tissue with them, but not seriously impede the implementation of a Neural Network Simulation. So by ignoring the details about the connection architecture, and how it changes the shape of the neuron, we could simulate both modified, and unmodified Pyramidal Neurons in much the same simulation.
Once we can implement heterogeneous groups, the next step is to implement a basic Implicit Memory using a three layer architecture. This is where the data cloud comes in. The output of the implicit memory is a data cloud, because of the nature of Content Addressable Memory which I maintain is the mode of address of the implicit memory. This three layer structure is found in most of the allocortical areas, albeit in the earliest forms of cortex structure closest to the brain stem, such as the structures surrounding the hippocampus, it is not always obvious that an implicit memory is being formed.
Once we have our basic Allocortical Tissue, we can play around with improving on it, and in Heiko Braak’s Architectronics of the Telencephalic Cortex, we can clearly see that the subiculum and entorhinal cortex phase between different architectures, often overlapping as if one layer of function was one organ and another the second type, including a possible 4 layer implementation of an explicit access scheme that seems to be used to store access information on concepts. Eventually however, this intermediate and often evolutionarily distinct set of tissues stabilizes to form Isocortical Tissues which form the Gyrus level strucutures in the Gyrus/sulcys/divide architecture that breaks up our cortex storage into Sensory Modalities, and specialized areas.
It is in fact at this level I maintain and only above this level that we are able to map the function of the brain by the location of the data, because at the gyrus/sulcys/divide level, there is enough simularities in the brain to allow mapping within a species, and to some extent between species. Once we have isolated a BrodMann area, for a specific functional role in the brain, that area if projected onto a different animal should have about the same role, the animals brain.
Brodmann areas are not mapped specifically by the gyrus/sulcys/divide system, they are mapped acording to cytoArchitectonic zones where the characteristics of the zone are similar to each other over an extent. However in general while the Architectonic details of the layers are different, we can class them easily into Allocortical tissue, and isocortical tissue of a number of general subtypes.
When you claim a specific function for a specific part of the brain, such as speach for the Broca/Wernicke areas, and long-term memory for the Sub-Temporal Area, you are supporting the view that at these Brodmann areas specific functions have specific architectures. I haven’t got quite that far with my theory, if only because I am still looking for information on the specific architectures of the specific brodmann areas. I have a hope that the companion book, I just got from the library, Architectronics of the Cerebral Cortex will help with this part of my model.
My analysis in general of Isocortical Tissue is that it’s role is to provide a stabilized explicit access interface to the general Allocortical memory, by adding a second interface so that memory can be addressed in two ways.
(ref my Dual Mode Cortex article). Of interest here is the way that the attention system suggested by Dr. LaBerge interfaces with the Isocortical Tissue architecture. I extend this with some trepidation to include two types of implicit memory based attention what I call Orienting and Implicit Attention, which definitely put me in the weak attention school as suggested in the thread on Consciousness and Attention here on Nature Network. If you want a bit of an overview of my attention theory, look at my article on Attention under the list of stuff being reviewed by Dr. LaBerge.Ok, so now let us look closer at the Isocortical Structure. The confusion around the Columns and Mini-columns comes from mixing Dr. Laberges Apical Dendrite information with Eccles 1983 article The Horizontal (Tangential Fiber) System of Laminae I of the Cerebral Neocortex, Acta Morph Hung 31, 261-284. Essentially what Eccles is saying is that the architecture of the cortex favors cylindrical arrangements called Columns, and what LaBerge is saying is that within these cylindrical arrangements are structures called Mini-columns also cylindrical but which have axis neurons formed from Layer 5 Pyramidal Neurons, and which, I have suggested correspond closely in size to Neural Groups, and therefore might actually be a way of pre-activating specific neural groups.
The Bottom-up Attention system, I suggest is therefore possibly implicated in the “Place-Code Addressing” required to explicitly access a specific memory. Which in turn is needed to rehearse it, or to recall it at a later date. This often causes problems during explanations because place-code addressing is a term based on information theory, It just means that there is an address for a specific memory element, in this case a Neural Group.
So here we are, with the ability to address a specific memory, and why you wonder do I complicate it with talking about a data-cloud? I mean from the tone of your question I assume that you think a neural group is a state output.
Ok, so lets be clear here. The problem isn’t with the statehood of the neuron, the problem is with the fact that the output of all these states, is in parallel, and forms an unstructured data-cloud because it is built on Implicit memory and that is the nature of the output of implicit memory.
So, you might ask me why don’t we just address things at the explicit level and ignore the implicit memory if it is so hard to deal with. the short answer is that you can’t because the Implicit memory contains the analysis you need of the stimuli coming in, You need that information, but it is in a form that can’t be easily digested. And this is what my contribution is, is an understanding of how it all fits together via the Bottleneck, where the output of the Implicit Attention System, is converted into an equivalent Explicit Address by which it can be retrieved.
But in figuring out the system I have also figured out some caveats on the nature of data being converted, which has a bearing on what we can do with it, and demands further attention phases to resolve specific memories out of the Data-Cloud. It is these later forms of attention, that allow me to extend my theory beyond just a memory theory and begin to discuss higher processing models, control systems, and a Consciousness Model based on a specific Cognitive Architecture. But what I can’t seem to get across to people is the caveats on implicit memory and this intermediate form of memory that lies between implicit memory and Declarative Memory, and therefore does not easily fall within the so called Explicit Range, even if I call it that.
-
Dear All
I am Kamal, a new member just signed up today. i am sorry if i interrupt with an irrelevant question, but i see from your discussion that u may find the answer for my question or even guide me where to go. i am a researcher in pediatric neuro oncology and i am facing a challenging issue every day: in pediatric patients with brain tumors doing a neuro surgery for the excision of the tumor. at our institution we do a post op mri within 24 hrs post surgery. my question is how can i differentiate between residual tumor and post op changes? -
Sorry Kamal, I can be little help to you I have never worked with an MRI.
From what little I know of them, the primary problem is to create a contrast between the tumor cells and the post-op changes, which probably requires that you detect some chemical change that is unique to the tumor, and use a display mode that gives that chemical change more contrast.
Since I am not an Oncologist I don’t know what that chemical characteristic would be, nor what standard modes of Visualization your MRI might have. Recently, researchers have taken to doing an extra step where they take the base information from the MRI and massage it outside the MRI to do more processing on it, than the MRI would do on its own. DSI comes to mind. Ideally you would simply come up with a new visualization technique that so contrasted the Tumor against post-op changes that it was obvious which was which. I have no way of knowing if someone has already done that since I am not up on the Journals of your Discipline, but if not, you might, if you can figure it out better than anyone else, have an MRI mode that people will buy from you for its utility, thus at least offsetting the cost of the MRI time you would need to develop it. Good Luck.
-
To get back on topic, One of the reasons that I have not yet implemented a simulation of my memory system, using neural networks, is because the Neural Network Theory I am familiar with, does not lend itself to variation of the types of neurons that it is based on.
When I have attempted to define a new neural model in Java, I have come up against an explosive increase in threads when I considered creating threads for different learning phases within the cell. Not being a sophisticated enough programmer, I had to abandon my model temporarily until I can think of a way of synchronizing multiple neurons each with multiple phases of learning going on in parallel over multiple activations. The brain does this in parallel with every cell doing its own processing of multiple learning processes biochemically, but I am worried that in the limited environment of a Cluster Computer, where I hope to simulate this process, internal sequentiality within the computer will result in synchronization problems that will expand with the number of threads being implemented within the model. If the thread count gets too high, I am worried the simulation will founder.
When I add to this complexity, the scheme of adjusting the number and distribution of connections that form as part of the learning processes, according to some inheritance scenerio, my mind founders, not the simulation.
An alternative I am considering is to use polling instead of threads to implement the parallel biochemical learning mechanisms, and to use a parametric data-file ala XML to hold the characteristics of the learning processes, but the exact implementation eludes me. I am simply not that sophisticated yet in Java, that I can be sure to pull off that complex a simulation.I had hoped that by publishing some of my ideas on the Internet I could attract others that would be willing to work with me to flesh out my theories, and do the necessary research to begin to implement my Artificial Consciousness, but explaining my ideas, is much more difficult than I had hoped, and the Virtual Institute I am building on the english Wikiversity,
http://en.wikiversity.org/wiki/Portal:GreySmith_Institute has turned out to suggest such a far ranging research project, that just mapping it, out has created too large a site for easy reference. I need to consolidate the site somehow, but have not yet arrived at a scheme that will allow me to do so without losing the quality of research I hope to achieve. -
Ok, The lack of response lately means that I have belabored explicit memory enough, or I have chased off the hecklers again.
The next topic I would like to discuss is the Necessary Attention Phases a system like I describe would have to go through.
The first phase, I call my Snail Model of attention. In this phase, information from the primary perception centers of the brain is telegraphed to the Basal Ganglia where a quick response network mixes instinctive and conditioned responses to do immediate responses to stimuli.
I base this phase on Dr. Arkon’s work on Snails. Essentially a snail brain consists primarily of three types of neurons, Sensory Neurons, Ganglia Neurons and Motor Neurons. Conditioning experiments ahve shown that this simple brain is able to learn conditioned responses that involve more than one sensory modality despite the segregation of the neurons at the sensory level. For instance, a snail can be conditioned to withdraw it’s syphon or gills according to a conditioned stimulus that is not within the instinctual pathway. Also a snail can be conditioned to “Focus” it’s foot, a mechanism that usually results in increased suction, via a different sensory modality than the balance organ that usually triggers the effect.
One of the conditioning experiments is to put the snail on a tilting table and tilt the table while flashing a bright light. Eventually the snail will be conditioned to focus it’s foot without tilting the table whenever a bright light is flashed. Research into which cells are involved, has implicated cells in the ganglia in the triggering of the motor response needed to focus the foot.
This means that the ganglia somehow binds the cross modal signals in order to trigger the conditioned response. This early binding effect is part of the reason I started responding to this group in the Binding Problem thread, however I found that thread hostile to my work, which is why I started this thread. The reason I blame the ganglia in the snail on the binding effect, is simple up until that point the signals for the various senses were not combined and therefore defined a stovepipe of analysis based only on the single modality.
If this is true, then looking at the Quantum Mechanical level for signs of mixing of signals is counterproductive. However that was not a popular point of view on the thread in question.
What I suspect the basal ganglia do in the human brain, is mediate instinctive movements such as orienting. They do this by somehow priorizing possible movement according to salience, and triggering somehow the motor neurons that activate movement. There is no reason to assume conscious involvement since the only input conscious involvement seems to have is some input on the priority of movements, allowing it to somehow suppress the reaction if there is a reason.
How Ottmar, in his zeal to protect himself from any animal model of consciousness, has claimed that humans are not instinctive in their actions but are deliberate. While that might be true some of the time, All it takes is to drop a piece of glass at a party and you will see many heads turn towards the sound, suggesting that this instinctive reaction is not all that optional. It is mostly those that know that the sound will happen that do not turn to look.
The fact that we turn our eyes towards a sound, is a sign that this instinctive reaction is cross-modal.
Now another important distinction has been made by Hans in the Consciousness and Attention thread, where he has suggested that turning you head and failing to attend to a stimulus, is not the same as rigid attention to the stimulus. What he is saying is that somehow this weaker form of attention is less validly called attention because it does not succeed at catching the attention at higher levels.
But let us look at what is involved, if my model of implicit memory is correct. Somehow the brain is connecting a stimulus with a specific zone or area of the environment, well enough to turn in the general direction, well enough that if we are interested in the sound our further attentions can direct our senses to the actual location of the sound, and we can detect the source of the sound, and then, if we want to dismiss it. Let’s face it broken glass behind the bar, is a lot less threatening than broken glass on the dance floor where someone might step on it.
Now You would think that with my indivisible Data Cloud that I thought that interpretation of the data cloud was impossible. Not so, division of the data cloud is difficult, interpretation just requires a neural network, and time for it to learn what signals from what parts of the brain, require what responses. The important thing to understand is that there is no segregation of memories involved, just a reaction to the content of the data-cloud.
However there is segregation of the reaction zones. Somehow we need to be able to locate the sound, and turn our eyes in that direction, which means we need to know information gathered in the primary perception areas of the brain. There is also some form of salience ranknig to prioritize which areas of the party we turn our heads to first. Which is more important the sound of our loves voice, or the sound of glass breaking?
This probably involves the limbic system in the determination of which stimuli are more salient. One way of looking at Orienting is that the basal ganglia filters the data-cloud for salient zones, and automatically orients the body towards the most salient zone.
Now we get to an idea of what the nature of implicit attention might be. If we have information about which zones are most salient, then we can label the senses from those zones, and thus filter the conversion from implicit to explicit memory so that the memories that are the most salient get converted first.
This explains by the way, why foreground and background were not found to be significant in GSO formation, in essence foreground and background are too specific for direct conversion to happen. However if we link each salience zone to a different GSO, then sampling salience zones according to a priority becomes easier, and Orienting merely sets us up for further attention modes.
It also explains why Functional Clusters, specified by GSO frequency, are cross-modal in nature, the ganglia have linked the salience to the external environment zone, not the sensory modality. What confuses me about this model is simply that fact that the Thalamus is thought to be the source of GSO signals because of its connections in the Reticular Activation System.
How then does it activate zones determined by the Basal ganglia?
One possible mechanism is by modulating the strength of the sensory inputs that are routed through the Thalamus in higher mammals. It also seems unlikely since that would probably be a detectable variation in the inputs to the cerebral cortex.
Another possible mechanism is by modulating the Layer 5 pyramidal neurons, but how the basal ganglia addresses those neurons is problematic.
The final possibility is that the thalamus supplies the base frequencies and the Basal Ganglia distributes them via its projections returning to the cerebral cortex.
-
Now I get to a part of the model that I have just found some support for in Architectonics. The idea is that the explicit form of memory, because it uses the implicit memory for its storage, needs to be redescribed into distinct and separately addressed memories. The first step is to redescribe the functional cluster into a CHUNK, The next step is to rehearse the chunk but edit the address list.
Thus we have in the first case our primary or core areas associated with a specific sensory modality, and in the second case we need someplace to store an image of the partial data cloud. This secondary area is called a BELT in the architectonic documentation. Now what is important here is that we don’t need to store the data, since that is still in the primary core, what we need is to store an interpretation of the data, as we segregate it from the original core perception. As a result the Belt areas are actually larger than the core areas because they store interpretations for more and smaller elements. Finally we have the so called Association areas, which I will get to later.
The mechanism I suggest for the attention system that controls the movement of data from the core areas to the belt areas, is the Triangular Circuit Theory of attention by Dr. Laberge. In this theory, we have bottom-up attention thought to be controlled by the thalamus, Then we have top down attention thought to be controlled by the prefrontal cortex. As well we have signals passing from the prefrontal cortex to the Thalamus through the Nucleus Accumbens to the Nucleus Reticularis Thalami. It is this pathway that I think is used for the editing of the CHUNK, in order to get sub-chunk arrangements of bottom-up addresses.
So the Triangular circuit not only pre-activates the functional clusters, and acts as an addressing scheme for the explicit memory, it also directs information from the primary or cores memories to the secondary or belt areas, and in doing so, allows for interpretation of the elements hidden within the original data cloud.
The important thing to realize here, is that we didn’t edit the data cloud, we editted the address list.
The next step is very speculative, what I think happens is that the same addressing scheme is applied a third time, to recruit association areas.
To understand this hypothetical scheme we have to accept that memory and processing cannot be separated, we can only optimize cells for one or the other. In fact, the idea of content addressable memory, is that each memory element samples the content, and decides for itself if it will respond by firing. One of the aspects of the cerebral cortex, is that while we can see similar layered structures in the Isocortex, each area of the brain has a slightly different flavor of Isocortical Tissue, and this is partly because each area of the cortex is processing a slightly different type of signal.UP until now we have been talking about autonomous calculations that we guide simply by directing data from one area of the memory to another. We select which memory is moved, but where it goes is hardwired. In this stage, we want to not only select which memory to move, but we want to recruit a specific specialized processing center to deal with it.
For an example let us consider the Fusiform Facial Gyrus, which it is thought is recruited in order to process faces. Now obviously for this to work we first need to be able to select the face, and then recruit the specialized module. I call this type of attention Complicit attention because it requires that two types of attention be able to be done simultaneously, a Data fetch, and a process fetch. Sorry if these terms are not familiar they are computer terms, used primarily because the concept is familiar to me via my background in computers. The term is obviously coined from Comp short form for Computational and licit the last part of Implicit and explicit which I have used to describe other types of attention. However it means a being mixed up with another in an action. Which sort of describes the way the two types of attention combine.
To make this work properly, without mixing the attention modes and thus mixing up the data clouds, the data has to be squeezed through a buffer into a network, which will present it simultaneously with the pre-activation of the processing center. It is thought that the corpus collossum takes on this role. Other similar commissures may have a similar role where the corpus collosum fails to connect directly to the associative areas involved.
There is evidence of address buffers, that feed the attention system, in both the frontal lobes and the hippocampus area, an interesting note is that if either of these centers is disabled somehow the brain recovers, but if both are damaged short term memory is lost.
If we accept the presence of these buffers, then we can say that the combination of data in both of the two buffers constitutes a data-function tuple that is equivalent to a command to take a particular data element and process it in a particular function. In other words recruitment of specialized centers in the cerebral cortex is equivalent to commands in a processing system. Once the data is processed it becomes just another zone of salience to be evaluated and either selected or not.
Results
-
