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
-
Ottmars BAAAAAAAK!
Robert, Lets start with your statement that there is short term memory associated with long-term memory. In the biochemical thread model of neural function ALL long-term memory must be formed in a cell capable of forming short-term memory. In other words the conversion happens inside the cell, and does not require a buffer, unless all neurons are considered buffers. In either case, the word buffer loses it’s meaning.
Secondly, if the only way to tell if a memory cell is a long-term memory cell or a short term memory cell is to determine if it has the biochemical thread that converts short term memories into long-term memories, and if the short term memories are sustained separately from the long-term memories, and so that cell can react with one short term result and another long-term result for the same stimulus. Classifying one area of the memory long-term memory and another area short-term memory seems rather counterproductive.
So what I am suggesting is a better idea, is to expect different specialized areas of the brain to have both long-term and short term memory, If we look at this, then we can clearly see that there are three main memory loops, that affect the storage of memory. There is the Cortico-Thalamic Loop or main memory loop, the Cortical-hippocampus loop, and the Cortico-basal loop, your terms might be different but the three neural loops are well documented.
Outside direct linkage to these loops themselves, but perhaps linked through the selective function, is the PFC, which is like the rest of the cortex common to all three loops.
One of the problems with the theory I have exposed in this article, is that so far although we have set up an addressing scheme that can select specific memories, we have not yet dealt with the way we demand them so that we can implement a demand memory system. Retreival has proven much more complex in the past, so perhaps we should not expect it to be as simple as just mapping the index to the memory. One of the problems with a neural network based system, is that there is no map in a naive system to use to map memory with.
There is that pesky Fodorian Phenomenality which I have been assured is not what others think phenomenality is at all.So lets start with a naive system based on my hypothetical model, and we see that it has no way of selecting objects, because it has no method of telling what is an object and what isn’t except saliency. What it needs to be able to do, is map stable zones of salience and then analyze them to determine if they are objects or background.
To do that, it must store the chunk for the salient zone somewhere and analyze it over time. Now remember, it doesn’t have to store the contents of the memory the data cloud, it has to store the address array that triggers the data cloud. But which chunk should it store and analyze first? and how does it choose that chunk?
Now we are getting into control/motivation factors. Given no information on which to prioritize, how do you make a selection? You guess. If all salience zones are equally prioritized, then you could do worse than picking one at random. Thus the easiest form of control system is random impulse where you select a chunk at random and analyze it. Obviously this is a poor control system, with low survivability in a complex world, but without a system to prioritize random impulse at least means you end up with some data processed.
This is where we get into the idea of the Limbic System as Drives and Biases. If you are going to be moving randomly like an eel out of water, then it makes sense, that for survival you should be oriented towards moving in a direction that increases your survival. This is what the instinctual level of control is all about. It is the body, teaching the brain how to survive. One way that it does this, is by motivating the brain towards actions that increase it’s survival potential, (as defined by DNA). So naturally we have pleasure and pain as the ultimate motivators. Pain acts to reduce the bias in favor of doing something, and Pleasure acts to increase the bias in favor of doing that something. By biasing the selection of chunks to analyze, the limbic system drives the body towards survival.
The drives, reinforce the salience priorities, making it likely that areas that are highly salient, are also high priorities for analysis.
Once we have analyzed a salience zone, and found which elements are objects and which are background, we have the problem of how do we store the knowledge for later reuse. We don’t have to store the contents of the salience zone, but what we need to store, is the meta-data that tells us what our analysis found when we analyzed it. One way of doing this is what I call a Meta-index, we separate the data into indexes that follow each different type of analysis, and create a sort of master index, that lets us search the other indexes somehow.
The fact that we can find one of those index mechanisms, helps us determine where in the brain this type of function is likely to be found. The Episodal Memory at least in the Rat, has been located in the hippocampus by the finding of Place-Cells in CA3. Now caution should be taken with this discovery, because we are not 100 percent sure that Rat Hippocampi and Human hippocampi work the same, but there is reason to believe that they come from the same evolutionary base, so perhaps they might.
It is therefore my theory that the cortico-hippocampal loop, is probably the basis of just such a Meta-index system. Without getting tied up in the detail of exactly how such a meta-index would work, what it allows is for us to declare memories by typical indexes. I use the word Declare advisably because I am now going to make the case that in fact the presence of the meta-index is critical for the development of Declarative Memory.
Essentially we can’t search our cortex for a specific memory unless we can search the index, for its storage chunk and then rehearse it. So having a Meta-index, allows us to declare memories and search for them. It is only once we can do that, that we have a true demand memory.
Now we get to Amnesia Models. It wasn’t until scientists learned that H.M. a classic Amnesia patient who had both anterograde and retrograde amnesia, could learn skills, and do things that required long-term storage without being able to remember that he had done those things, that people began to suspect that amnesia might not be always loss of storage, but also might be caused by loss of index information. Up until that time, the idea that you recalled a memory if you knew it, was accepted. But once the idea was put into scientists heads that you could know something and not be able to remember it, they began to look for implicit memory.
What if, some scientists suggested, H.M.‘s amnesia wasn’t about loss of memory but loss of index information? What if it was a declarative loss, rather than a content loss. If this model held true, scientists asked, then why was there also a retrograde amnesia?
It was hypothesized that the retrograde amnesia was really the loss of information involved in a process called consolidation that somehow caused old information to be archived in long-term memory. So one set of scientists were looking at amnesia in H.M.‘s case as being loss of index, and another set of scientists were looking at H.M.’s case as being loss of transfer to long-term memory. This meant the area of the brain that H.M. had lost in the operation that created his amnesia, must be involved in long term memory.
Because everyone expected there to be a location where long-term memory was stored, this approach seemed the most likely and so everyone started talking as if the hippocampus was part of the long-term memory. It should be noted that consensus by science does not mean it is right, just that most people agree that it is likely that this is the case.
Our understanding of the Bio-chemistry of long-term memory was started by the assumption that the hippocampus was involved in long-term storage.
It caused Eric R. Kandel, and his contemporaries at Columbia University to study the hippocampal pyramidal cells for evidence of long-term storage, and he found that evidence, and earned his Nobel prize.It is partly because of his work that we have the cellular model of long-term memory. And here we get into an interesting tangle, does the presence of long-term memory threads in the biochemistry of a CA3 cell, indicate that the hippocampus is involved in long-term memory, or does it indicate that long-term memory is developed at the cellular level?
Evidence for long-term implicit memory also came from H.M., so Cortical Plasticity is an important test case, for suggesting that any cell that has the long-term memory mechanism could implement long-term memory, regardless of location within the brain. If the hippocampus isn’t the source of all long-term memory then, what does H.M.’s retrograde amnesia where he forgets portions of his memory of events up to 2.5 years before his operation mean?
Lets go back to the indexing model, It might mean that consolidation is really some operation that moves an image of the index, out of the hippocampus and into the cortex. Since the index is stored in the hippocampus, if we lost the hippocampus the index would be gone, and he would have total amnesia. But since he remembers everything up to 2.5 years before his operation, there must be some form of indexing that does not include the hippocampus. It is this, that is incomplete until consolidation completes it. So areas that are thought to be long-term memory in the cerebral cortex, might be instead index data consolidated from the Declarative Memory Meta-Index.
This information is critical for retrieval of data from the Declarative Memory but not from the implicit memory since that memory has its own long-term storage.
If we accept this model, and it is merely the most complete model, that I can come up with, then what are the characteristics of the memory cells that make them candidates for long-term memory? Evidence seems to suggest that there are biochemical threads of memory that are triggered by certain complex types of Calcium ion based synapses. Of interest so far, are the cells with the NMDA ion channel, and cells with the Serotonin based S calcium channel that has been linked to Facilitation.
The fact that both of these ion channels pump Calcium ions, suggests a link to habituation a process whereby the gradual reduction in calcium in the presynaptic bud seems to attenuate the signal of the synapse. Opening of the S syanpse seems to immediately reverse this trend, restoring full signal strength. Both the NMDA synapse and the S synapse produce a secondary transmitter called cAMP which has been implicated in the creation of bio-chemical cascade reactions that are complex, and according to Kandel, trigger expression of DNA which is associated with fibril growth. What else these biochemical cascade reactions are doing, is still under investigation.
The S synapse might associated with a less sophisticated form of long-term memory, or alternately we could have multiple mechanisms whereby the same biochemical cascade reactions might be triggered that have therefore different parametric considerations as to when long-term memory is formed.
Alternately, the difference between Facilitation with the S synapse and LTP with the NMDA synapse might indicate that the difference lies more in the short-midterm memory characteristics of the cell. This is confused by the presence of two types of LTP, S-LTP which lasts hours, and L-LTP which lasts weeks. The existence of these two types of LTP either indicates that different species of neuron have different cascade reactions, or that the conversion from one form of LTP to another, is optional within the greater cascade reaction schema. In any case long-term memory is more complex than we thought, and the short-term long-term model of memory does not hold up well to scrutiny at the neuron level.
-
Graeme,
functionally, there is short term and various forms of long term memory. We can be confident in this functional separation as these functions can be independently lost through brain insult.For instance anterograde and retrograde long term episodic memory can be independently lost with no effect on short term or non-declarative memory.
The question is: if functional and structural memory systems do not map directly on to each other, how should we proceed?
The first thing to not do is to claim that a functional or structural module does not exist simply because there is no separate brain area or function for that area. We would, however, expect that a brain insult in a particular area would result in a particular functional deficit. But this does not actually always follow. Some brain insults to particular areas result in multiple deficits as the area effected has several functional roles.
Information is either stored in memory in a unitary form or diverse form. If it is unitary then no STS or assembly area is needed. If it is diverse then such an area is needed. As memory degrades asymmetrically (not in discrete units) we can assume that diverse memory storage is the system in use. Thus if I know a person, for instance, and meet up with them after some period, I will recall their face, name, where I know them from, what my relationship to that person is and so on, in diverse chunks that take time to form a complete picture.
If all that information were stored at the same time, would it be stored together? I think not. But general short term memory could have the function of assembling of recollections and so there may be functional separation without physical separation.
In loss of face recognition (prosopagnosia) we know that all the details of a familiar person’s face can be recalled and repeated without recognition of that face occurring, say if a wife’s face, the Queen’s face and Dame Edna Everage’s face are shown to a sufferer of prosopagnosia and they are unable to pick their wife.
Although multiple systems may be involved in such a process as face recognition, it does show how particular functions can be knocked out.
The assumption that there is no specific short term store brain area may or may not be accurate, but the fact is that there is functional short term memory and that that functional area can be effected independently of other memory functions eg it can become very short or in some cases fail to function at all, leaving a sufferer perpetually disorientated.
As for object selection from long term store (semantic memory, in this case), the selective loss of semantic memory may be instructive when assembling the meta-categories ahead of specific selection. For instance a person can lose recognition of all living things, of all motor cars and so on. The specificity of loss has astonished some researchers and I don’t have a good general reference to hand that catalogues all of these specific deficits. Word loss appears to be just as narrow: verbs, proper nouns but not people’s names and visa versa and so on.
One would expect that storage and retrieval are intimately linked, but can one lose storage and not retrieval? The answer is yes. But we don’t know about the opposite for obvious reasons.
On your comments on retrieval, the hierarchy for at least some forms of memory is proximity ~ those memories laid down last are the freshest and most readily retrievable. To delve further one must follow the link between proximal and more distal recollections. This would be one strategy available for all long term memory types.
The second form is via matching of emotion. Those recollections that were laid down with a strong fear emotion, for instance, are more likely to be retrieved when the ambient emotion is also fear. This is a very good survival strategy. Combined with the proximal-distal memory prioritisation an animal would be able to pull up the last successful strategy for avoiding the current danger very quickly.
Association is the third obvious method, especially if we start with meta-categories eg ‘all the people I know’.
We now have three methods of tracking down a memory and survivability is built in. It is likely that emotion rules if the emotion is very strong eg fear, but other methods become more prominent if the emotion is less strong or ambiguous. The second priority is most likely to be the proximal-distal method, but if there is less urgency for a recollection then associative method may become more prominent.
This appears to be how recollection operates functionally, and it is hard to see how a species could survive if the order of recollection was other than the above (proximal, emotional, associative, random). The random process I would relate to lateral thinking and place that in a fourth category, perhaps along with meditation and contemplation. It may also be the first but I would estimate that such a method would be a transient or fall back method where no other method was available. But then we could say that if the first three methods gave no guidance then the fourth would be used anyway, so some recollection always occurs when the memory is paged.
As for the reuse you mention, ‘reconsolidation’ is a well known phenomena and is generally thought to occur for all recalled episodic information and possibly other forms of memory as well. Thus when you recall that day you went to granny’s house 25 years ago, what you are actually recalling is the last recalled version and not the memory as originally laid down.
As for the discovery in rats and I would the discovery of some ‘mother’ cells in humans, I would point out that this is a fast index of often recalled information and not a general index. Losing these specialised cells is unlikely to result in loss of the memory, just the loss of the fast index reference.
I think we have consider the brain tissue and how it must operate organically. An index system is compelling until you realise that all the cells that may be ever used in the index must be occupied from day one, so to speak. The number of ‘mother cells’ probably changes very slowly through life, and the change is most probably a loss of cells in that index despite a rise in the total recallable information in the brain.
H.M. had very little retrograde amnesia and much of it returned anyway. Another famous amnesiac, patient N.A., suffered no retrograde amnesia (he remembers the incident that caused his amnesia, being struck in the nose by a toy foil by a colleague who was just horsing around) but almost total anterograde amnesia. In both cases non-declarative (symmetric) memory was discovered to still be present.
This appears to be an indication of at least two different memory systems.
Note that the short-term ~ long term memory model is not in dispute from a functional perspective. As for structural models, the selective loss of each function appears to confirm at least some associated structures. My reading indicated that whilst the location of memories is not yet understood, we are much closer to understanding the storage and retrieval mechanisms.
As for function not necessarily mapping onto structure, consider a form of short term memory or buffer responsible for assembling long term memory into a particular ‘information cloud’. This can be achieved without any separate structure if the process that locates the elements of the information cloud merely ‘tags’ them eg by making some key neuron/s fire continuously at some rate. The process of retrieving the memory then only has to locate the ‘tags’ and connect to them. Thus the actual information may remain in place even though it is located, assembled as part of a cloud and then made available, all functionally separate modules.
Robert
-
Robert says that “Functionally there are short term and long-term memory”
Yes, But it would be a mistake to assume that this is the way the brain is organized.
Robert then goes on to say, “The problem is, if Functional and Physical systems do not map onto each other how should we proceed”
Well part of the problem is that I am trying to show you that the physical systems are a superset, of the functional systems, and therefore that the functional systems are not adequate to model them. Either we need a more physically oriented model, which I am working on, or we need to rethink our functional model to make it more closely describe the physical systems.
Either way, I am not happy with the state of the current functional model.
I have not so much said that functional areas do not exist, as I have challenged them to fit more closely with the physical system, or have offered alternate interpretations of the data they are based on.
On unitary or diverse forms of storage:
There is no doubt that memory is stored in a diverse form. However the assumption that a separate stage of gathering that information is required, speaks to a misapprehension about how memory works. Despite it’s diverse nature, memory can be accessed in parallel in such a way as to capture a Unitary image of a diversely stored cluster of data, without requiring a separate location for storage. The resulting cluster of data, is called a Functional cluster, as introduced to me by Dr. Edelman in his book “The Remembered Present”.
The diversely stored data, is bound by selection according to GSO data added on top of the original signal. The main problem I have with the Functional Model here is it’s assumption that what is stored is information about objects such as foreground and background, instead of information about zones of salience within the raw data. Part of the problem is that recognition of objects cannot happen until there is recognition which lies in the secondary perception areas not in the primary perception areas, as I hope I have recently shown in a paper I am just completing.
So what do we do with a functional model that looks for a buffer, where there is no buffer needed? and mistakenly looks for information on objects before objects can be recognized? Call for it’s revision certainly, if only that!
I am amused at your sophistry to imply a buffer by using the same binding model I suggested to refute it. Obviously we think of different things when we say the word buffer, I might be counter indicating buffers because I understand how they work in a computer, and can’t find evidence of that type of connection in the Architectonics of the brain, while you have a less concrete idea perhaps based on the role that the buffer takes that lets you see a buffer, where the function is supported by other means.
I am amused at your assumption that the STS is an assembly area, rather than a short-term store for processing. This is certainly not the assumption of Miller who back in the 1950’s first attempted to measure it. However I will grant that those unfortunate enough to buy into Workspace theories will be tempted to look at it in that light. Quite simply, the STS is defined by the bottleneck, and is critical as a cache for short-term storage of explicitly addressable memory that has not yet been indexed. However it is not an assembly area, it is simply too small for that.
Trying to derive information about the main memory from episodic memory is a bad choice, I must warn you. Episodic memory is based as near as we can tell in some rather unique circuitry in the area of the hippocampus. Evidence shows that this area contains immature forms of cortex that have been coopted to be used for increasing the sophistication of the mature Isocortex or neocortex function.
The cortico-hippocampus loop, is critical for Declarative Memory, but it is the last place I would claim to understand. Different layering schemes mix and mingle in strange ways in that area, and I would need much more complete studies of it, to be able to analyze what the neurons are doing, than were available in Heiko Braaks book.
I think I have seen evidence of an immature form of explicit memory in the Entorhinal cortex, but then I am expecting such a form in that area. And would not make the mistake that my freind Dr. LaBerge did of assuming that just because it is explicit it must be addressed by the thalamus. Or be connected in the exact same manner to the PFC.
If my hypothesis is correct, about how the layers in the allocortical areas and isocortical layers outside the hippocampal areas work, then we can see Implicit and explicit memory configurations in areas of the cortex that are not the neocortex, and these areas may have also an equivalent to short term and long term memory that is not directly detectable by sampling the main memory. It would be a mistake to call them short term memory even though their term is short, if only because there would then be the assumption that you could somehow sample them from main memory. This is why I think we need to overhaul our definitions of short term and long term memory.
Evidence is that memory in the declarative index, contributes to retrieval of long-term memory but not to its content. However loss of specific index classes could result in loss of specific types of memories just as loss of content could. In H.M.’s case he lost only one hippocampal organ out of two so eventually by repatterning he could recover some of his function. This does not mean that the original injury did not result in retrograde amnesia.
What is important is not how much retrograde amnesia he suffered, but, the fact that there was any memory, not stored in the hippocampal area to use.
-
“I am amused at your assumption that the STS is an assembly area, rather than a short-term store for processing. This is certainly not the assumption of Miller who back in the 1950’s first attempted to measure it. However I will grant that those unfortunate enough to buy into Workspace theories will be tempted to look at it in that light. Quite simply, the STS is defined by the bottleneck, and is critical as a cache for short-term storage of explicitly addressable memory that has not yet been indexed. However it is not an assembly area, it is simply too small for that.”
I was considering an additional domain specific STS associated with the retrieval of memory items.
Note that the concept of STS and ‘Working Memory’ are not the same. Working memory is a temporary cache for information gathered or generated by a decision making process whereas STS is the last stage of perception, where information is briefly held ahead of processing. The two areas may be physically the same but functionally distinct.
Note also that all the memory items recently recalled remain fresh and readily retrievable. This is common when one is engaged in a discussion on some topic ~ ever more salient information becomes available. The ‘information cloud’ must have some way of readying the information to be retrieved.
It is my understanding that Henry Gustav Molaison (H.M.) suffered bilateral loss of hippocampal tissue
HereBTW I do not necessarily subscribe to all or any functional models ~ I have chosen this perspective, that is all.
As for avoiding confusion with ‘THE’ short term memory when discussing other short term stores one can use equivalent but non-confusing labels such as ‘Transient Store/Memory’. I like the term ‘perishable’ to denote information that is likely to be lost. One can also use more grandiose titles such as ‘Temporally Unstable’ or ‘Interval Dependant Stability’ and so on. There is no shortage of workaround for your other STS :)
Robert
-
Robert said: “I was considering an additional domain-specific STS associated with the retrieval of memory items.”
So, each loop in the brain would have a parallel STS? even though they all start at the cerebral cortex?
In my model there is one STS for all three loops, but there is short term storage outside the actual STS. When you talk about it being the last step of perception before we process the data, you miss the implications about the Core, Belt, and Association Areas, and their relationship to perception. All three of these areas are domain specific, all are short term memory, as well as long-term memory, and all are one step before processing even though there is a natural progression through the areas that starts at the core, goes through the belt, and ends at one of the Association Areas.
Now I admit that I am somewhat confused about the difference between STS and Working Memory, Baddely’s Modal Model, strongly suggests that working memory is domain specific. But his more recent versions have included an Episodal Buffer as one of the domains. In other words working memory has an interface to the Declarative Meta-index. In the physical system this interface seems to lie in the Dorsal Medial section of the PFC, which is not usually thought of as part of the Working Memory, although some people have been looking for buffers in the PFC to take the role of Working Memory.
I have tenatively suggested that the role of the Dorsal Medial PFC is to select data from multiple different areas of the brain, and project that information onto the Thalamus, so it can pre-activate the common cerebral cortex areas that all three loops share. In other words the Declarative Memory Meta-Index, outputs via the Dorsal Medial PFC to the Thalamus, where it pre-activates memories in the cerebral cortex, in much the same method as Rehearsal, but using the meta-index chunks instead of short term memory chunks. There may be an intervening step where a symbolic value in the Declarative Memory is converted to a chunk, but the effect is to direct the attention to a specific memory in the Cerebral Cortex.
This places the short term memory in the hippocampal area out of band for the main STS, while allowing it to select from the main STS.
-
Now if I am going to suggest that the role of the Medial Doral PFC is to redirect signals from other parts of the brain, to the thalamus, in order to pre-activate areas in the prefrontal cortex, And I am going to hint at there being three loops of memory, then it makes sense that I must suggest that the third loop which connects to the brain stem/Cerebellar area, might also be connected indirectly through the thalamus.
To understand what this means however we have to go back to Marr’s 1969 theory of the Cerebellar Cortex, where he discusses how it might work. In this article he suggested that purkinje cells are similar to pyramidal cells, except that they have much more complex apical dendrites. His interpretation of this, was that instead of each cell holding a specific signal, each purkinje cell held a number of signals. The idea being that the cerebellum produced sequences of actions, instead of a single action, and these actions were memorized by rote, and then served up according to context signals that entered via the molecular layer.
He went on to suggest a direct mapping from the inferior olive, but didn’t go into detail on how the system described could result in action sequences.
If we think about this, what we are describing is a sort of historical action sequence server, that is triggered by context, and probably due to the close connection between the thalamus and the Cerebellar Cortex, serves up the action sequences via pre-activation of cerebral cortex neurons such as the motor cortex neurons that drive the muscles.
Now I would like to throw some doubt on Marr’s model, because he assumed a 1 to 1 mapping of Motor Cortex area to Inferior Olive to Purkinje cell. I on the other hand would like to suggest the possibility that it is not a 1 to 1 mapping, but that the activation areas for modules in the Associative areas of the cerebral cortex might also be included in this mapping. Other than that part of the theory, The organ would operate as advertised.
What this addition means is that pre-activation of sequences of processing commands can be done as easily as pre-activation of sequences of actions.
How does this connect to the thalamus? well, my idea is that the sequences of action/processing commands, are equivalent to neural group activations, and can therefore be substituted directly for them. In essence what the cerebellum is serving up is sequences of chunks.
Like the implicit memory however what we get is a redundant cloud of activated areas of the cerebral cortex. It is useless unless we can also select from among the sequences, in a manner similar to that done by the ACC in the formation of CHUNKS in the first place.
It seems necessary to have some GSO-like signal to separate different sequences from the cluster. I believe that it might be an Alpha range frequency that is used, so instead of GSO I call these tags, ASO’s.
A similar effect might be required for the declarative memory and this area is thought to be the source of beta frequencies, so I would expect BSO’s from it.
Once the clusters are tagged with a frequency, there is no reason why the ACC couldn’t sort them just as easily as it does GSO’s. So the mechanism used by the cerebellum to serve processing/action sequences would use primarily the same mechanism of attention as that of the bottleneck attention systems, except that because the CHUNK is built at the cerebellum instead of by the Bottleneck it wouldn’t have to go through the bottleneck and become serially dependent, and thus limited.
Now that I have introduced the hypothesis on which I am basing my assumptions about skill memory, lets go on to deal with the higher order processing that it makes possible. Essentially I have told you that The data-function tuple is equivalent to a computer command, and thus that there must be a command language of some sort. Now I have described a program store, that stores sequences of processing/actions, which look at least to me like small automations. What I think I am describing therefore is the Central processor that has been the holy grail of functional psychology, but instead of being a section of cortex, where most people have searched for it, we find it takes up at least 4 separate organs which each have their own tasks assigned to them.
Having found the programming language the next problem is to determine how the body programs it. Essentially this is where I see the Dorsal Medial PFC coming into play, If as I have said it’s role is to select from outputs of other areas those that will be directed to the thalamus, then what it could be used for, is to select sequences from the cerebellum. These snippets of code that would be called Macro’s in computerese could then be formed into slightly higher order processes for automation and intention. But to do that we would need to string the macros together into larger macros.
This is where I think the SMA comes into play, essentially the SMA has been likened to a process that collects sequences, and a process that selects from among them, just the sort of interface you need to build a macro out of snippets of code. During intention the SMA is probably instrumental in helping the ACC select from among processing/action sequences from the Cerebellar Cortex. The SMA of course is directly interfaced with the Ventro-lateral PFC. So the circuit of SMA-Dorsolateral PFC-Cerebellar Cortex-Thalamus-ACC-Ventro-Lateral PFC-SMA-ACC would control the higher order processing at the intention level.
-
Graeme,
If your hypothesis is correct, then the programming language and macros (I understand that earlier computer models called them ‘scripts’) must have evolved from simpler forms. What should we look for in, for instance, Aplysia brains?Second, if there are four physically separate contributors (organs) to a process (the central processor) then you should be able to predict the outcome of a deficit to one of those contributors. There have been strokes and other brain damage (from accidents, cancer, operations to correct focal epilepsy) that effect just about every part of the brain. What behavioural indications are there of a deficit or even a malfunction of one of these organs?
Another place to look when nutting out these models is the process of maturation ~ do all four organs operate from birth and in the same way? Are there any innate macros that get the ball rolling? I think we can safely say that there must be at least some.
Thus to arrive at your model of a normal working brain we must evolve, we must find vestiges in extant species eg Aplysia, we must mature from a quiescent brain to a fully functioning one, and we should see signs of a brain with various modules or organs no longer functioning as they should.
And we should also be able to identify any signs of these macros and the three memory loops as they present consciously. In that we are well advanced as far declarative and non-declarative memory is concerned, and episodic and semantic memory also appear to be quite distinct when paged subjectively. But what of the Marcos? Are these different from skill memory, where we practice, say, a golf or tennis stroke until the body can do it all on its own (with the mind observing at a safe distance :)
Robert
-
Robert asked “If your hypothesis is correct, then the commands and macros must have evolved from simpler forms what should we look for in Aplyssia Brains.”
First of all, I have to say that the programming language and macros, depend on the cognitive architecture, so any previous cognitive architecture might be seen to be less sophisticated, and therefore earlier architectures might not support a programming language and macros. Aplyssia, for instance is too early a brain to support this type of cognitive architecture. Instead we see that the Applyssia brain has only three types of neurons, Sensory Neurons, Motor Neurons and Ganglia neurons.
In my model, therefore the Aplyssia snail, is a Quick response network, such as is required for the Orienting phase of attention. In other words we should not expect to find a programming language and macros in its brain.
There is a reason that I am using the word Macro instead of script. Scripts imply a mature programming language, such as the Command Line Interpreter of an Operating System. Macros on the other hand are often written in a low level language such as assembler. The Command Language I have suggested might exist, is by no means a high-level language. Hence the term macro for programs written in it.
Also the technique for generating a macro, and compiling it are different than they are for script languages, and so I am hinting at mechanisms needed to achieve the effect.
As far as I know this is the first time this theory has been publicly articulated, so there are no supporting lesion studies. Further, I know of no research that has been done to determine exactly what the modules in the Associative areas do, so determining the actual language has not yet been achieved. AFIK.
In this model there is bound to be some developmental delays, because the system starts out operationally naive, and has to learn to use the various mechanisms before it can activate them at will. However skill at moving and focusing the eyes seems to come fairly early, while skill at moving the extremities , and recognition of the mothers face takes longer. Since recognition of the mothers face is thought to be the function of a module in the Associative visual areas called the Fusiform Facial Area, we can assume that by the time this skill is achieved, the child has already developed the ability to program commands in the base language, and is starting to build skills by storing those programs into the cerebellum. According to Alicia Karmiloff-Smith, Children are born with an instinctive feeling for physics, so that once their eyes have begun to track, they begin to predict the location of things that are moving, and can signal their feelings about the way that the objects are manipulated. Of course in this system just getting the eyes to track, takes a significant number of hours of operation during which the brain is soaking in information like a sponge, so innateness is difficult to determine.
On the Macro Front, it should be noted that Macro’s are a skill memory, application in this model, and therefore we would see their operation in the building of skill in processing. Tests to dissociate skill in processing from programming via consciousness could be developed, I suppose. Lesion studies of the cerebellum, might indicate loss of skill in certain types of processing for instance, we could look at the learning curve for tested processing skills, and see how it changes without the cerebellum being active, that sort of thing. Some ground work has probably been done in the field but I wouldn’t know what it is yet.
Otmar, I have to say I haven’t looked into the new thread, it didn’t interest me.
-
Graeme,
note that the term ‘macro’ is also used for higher level programming such as those used in applications such MSWord, MSAccess, MSExcel and other such applications. That was the first usage that occurred to me when you mentioned ‘macro’. Assembler language is way down the other end of the programming scale.You may have to think through presentation of your ideas to keep these wide apart concepts separate.
Robert
-
On High level Application Macros, I have to say that the mechanism for forming macros in high level applications used to be much the same as the mechanisms for forming macros in assembly language. It is only in Microsoft products that the VB language was substituted in for what was essentially lists of button pushes, and thus macros became scripts. The problem isn’t with my nomenclature as far as I know but with the fact that Microsoft didn’t change the name of the function when they substituted a programming language for the original macro definition technique.
From my perspective, there should be no difference between application macros and assembler macros, with the exception of products that don’t follow the rules of macro assembly. Unfortunately Microsoft products are widespread and macro assemblers have lost a lot of popularity so the roots of the words are hidden, and the errors have become the norm.
There is an equivalent concept in C define statements, but that is a high level language and so you won’t see the parallels.
Essentially to build a macro, you take a list of commands, and equate it to a symbol. Then when-ever you insert the symbol, the list of commands is unwound to get the process. It is a very simple process, and has a minimalist interpreter having only two loops to implement it. The inner loop follows the list of commands and calls them, the outer loop accepts the symbol and unwinds the list of commands. The Forth language was built on such a scheme.
In my model, the inner loop is the cerebellum, and the outer loop is the SMA. Communication between them is achieved via the Dorsal Medial PFC. which selects commands from the set contained in the cerebellum and uses the cerebellum-thalamus link to preactivate those commands. The interesting thing is that once a macro is defined, the cerebellum begins to memorize it by rote, and so the number of commands that potentially can be addressed via the dorsal medial PFC increases. Because this does not go through the bottleneck, macros often run faster than the rote programs they copy.
One of the reasons that intention is possible is the link between the sequences or macros and the contexts that trigger them. In essence every time a new context is dealt with, the cerebellum contains a trace of how it was dealt with, and eventually, the cerebellum learns to output the sequence needed to deal with the situation automatically. Thus the mechanism for intention is essentially one of evaluation and selection. Somehow the activated sequences need to be evaluated for relevance, and one selected that is more salient than the others.
Salience of a sequence is more complicated than salience of a zone of interest in perception so there is likely more circuitry needed than just the basal ganglia and limbic system. However I only have some ideas of what is involved, and little information on how it might be achieved.
Results
-
