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
-
Sorry to come in late, but let me float a couple of general observations.
Firstly, in the introductory essay there appears to be an implicit assumption that all memory is the same. Perhaps you use different labels for other forms of memory. The forms that are most probably structurally different are:
1) longer term memory ~ more like computer flash memory;
2) working memory ~ a scratch pad where information is juxtaposed briefly in the process of decision making, recognition and other tasks associated with consciousness;
3) buffer memory ~ where brief delays in information are needed for timing or prioritisation eg in the process of binding diverse informational streams, for instance vision and hearing;
4) domain specific short term memory ~ the short term memory associated with particular senses or cognitive processes.Working memory differs from other forms of short duration memory in that it can handle any type of information that can be conscious (and most probably more than that). Domain specific short term memory is most probably only partially conscious and is important for the process of recognition. This form of memory is most probably a major contributor to illusions, such as optical illusions, as visual information must be held preconsciously whilst some recognition process occurs.
Buffer memory does not manipulate the information carried forward but merely holds it briefly.
The structural differences should be readily recognised by the distribution of these forms of memory – we would expect to find domain specific short term memory near to or integrated into the modules that process sensory information eg in or near the visual cortex for visual information. We would expect to find working memory reasonably modular in some part of the brain, say the frontal lobes. Buffers we would expect to find liberally distributed about the brain but concentrated at one end of major informational pathways.
Secondly, if neurons are responsible for memory then all the neurons which will ever be involved in memory work must be active throughout life regardless of whether they are contributing to salient information processing or not. That is because a neuron that is left idle will shrivel up and die, especially in the first year or so of life.
There is no blank slate that gradually fills up – the neurons are active all the time (to the degree that any other live neuron is active).
The question, then, is what are these neurons up to whilst they wait for their turn to hold some important information?
One possible answer is that important information is constantly shuffled around the brain, more dynamically when there is less to remember. Another solution is that the process of memory is entirely a function of patterns of synapse where numerous patterns can be found in the same population of neurons such that it is near impossible to ever run out of ways to add new synapses without losing the memory that is already stored in the same area of the brain.
The shuffling idea has merit from a couple of additional perspectives: we know that reconsolidation occurs whenever we recall an event ~ does this mean that a recollection is laid down elsewhere in the brain afresh, so to speak? Further, those memories that are not recalled tend to deteriorate as compared to those that are. Is this an indication that reconsolidation is the method the brain uses to make sure that the maximum amount of memory space is used and kept perpetually fresh ie by stimulating neurons through reconsolidation?
In the computer this would be akin to defragging the memory, except that the brain is doing almost the exact opposite – it takes contingent memory and puts the various bits and pieces elsewhere in order to give the maximum number of neurons something to do and therefore to not die.
Gives ‘nostalgia’ a useful evolutionary role ie it not only refreshes those recalled memories but it freshens up the memory generally, an essential process if the above conjecture proves to have merit :)
Kind Regards
Robert Karl Stonjek -
Hello Robert welcome to the thread.
Hmmm… How best to answer your comment.
Ok, lets start with the idea that I am somehow lumping all memory together in one lump and saying that it all works the same way. That is a tough one, I had no such intention of course, but, at some level virtually every form of memory depends on cortex tissues, and I am describing how we build from neurons of different types to an understanding of cortex tissues. Then I wanted to explain how cortex tissues interact with attention mechanisms which gets us sort of to where I am now. It is true that I could have introduced working memory earlier, but I haven’t, partly because it is an interpretation that has unfortunate implications. Don’t get me wrong, it is not that I doubt the existence of working memory, just the implied mechanism whereby it is supposed to work.
In actual fact when I talk about rehearsal of CHUNKS, I am talking about short term and working memory, the problem is that the memory I think needs storage is the CHUNKS and Sub-Chunk arrangements of Neural Group addresses, not as is usually implied the contents of the storage. Because of this shift of emphasis on what needs to be stored, I haven’t stressed the words Working Memory or Short Term memory as strongly as you might have liked. However if you look closely at what I am proposing you have a rationale for why there is a bottleneck, you have a search mechanism that explains its serial dependency, and you have a reason why accessing a specific memory is strongly limited by that serial dependency, which in turn explains Millers magic number for the contents of short term memory, and the later work done in the 90’s that showed that phonological effects changed the amount of storage in the short term memory.
It is true that I have not yet introduced Allen Baddeley’s most recent update of the Modal Model, but if you think about the topic I just introduced where I describe the core, belt, and associative areas in the various areas of the Cerebral Cortex, then you might see, that you have reminded me, in the nick of time, to do just that, explain Allen Baddeley’s Modal Model in terms of the core and belt areas in the brain, and how those react to the involvement of the prefrontal cortex. But before I get to that I need to do a little more explaining about the PFC, so please bear with me.
As for buffer memory, I must admit that I don’t see as much need for it with this cognitive architecture as with previous cognitive architectures With Dr. Laberges ability to extend the term of attention, and the involvement of facilitation and LTP in the memory mechanisms that extend short term memory, It is possible that buffer memory, at least in the cerebral cortex, is not going to be found. This is especially important because I have seen no evidence of a separate type of memory that can be called buffer memory in the architectonics of the telencephalic cortex, except possibly in those areas that are not yet dealt with, such as areas of the PFC that have yet to be functionally mapped. I suspect the presence of such a buffer in the PFC connecting the Anterior Cingulate Cortex, with the Corpus Collosum, but that is not where you claim it should be.
As for domain specific short term memory, what exactly do you think the Association areas are? We wouldn’t need complicit attention if they weren’t domain specific specialist modules.
As for Long-term memory, there are two types of long-term memory, long-term implicit memory which is just a special mode of implicit memory, and triggered possibly by the imposition of the 40 hertz GSO, and Consolidated Long-term memory, which do you wish me to describe first? I can tell you that if you are patient I will get to Consolidated Long-term memory, but if I have forgotten to cover implict long-term memory, my apologies, I mustn’t have organized my thoughts well enough to pry it into the discussion. It should have been part of the discussion of implicit memory.
As to your SECOND point, One of the most innane things I have ever heard is the statement that we do not use some portion of our brain, I have heard a number of different percentages, and heard it mis-quoted from a number of different sources.
In fact you are correct, if a neuron doesn’t fire every so often it shrivels up and dies. In fact one of the interesting things about neuroscience is that neurons spontaneously fire, even if they do not get signals for a while.
Whether that is enough to sustain them or not I suspect not, since neurons slough off of the brain at certain points in development, suggesting that they were not active enough, and they died. However that might be because they were genetically sensitive at that time, and did not get enough DNA activation to sustain their lives. Unfortunately or fortunately depending on your viewpoint I am not qualified to to research that point, so I will leave it as an exercise for someone else to do.What are the neurons up to… good question.
If you remember my description of implicit memory you remember that it is redundant. Further you might remember that Jerry Fodor has said that neural networks are phenomenal, OK, OK, I get into problems quoting fodor, but what he said is that the location of a memory within the network is not easily made, because the network defies placing a memory element in a specific memory location. So we can’t point to a particular neuron and say that is the red color of my sweater.
Instead what is stored is relationships between neurons at the synaptic level, and the synapses are more or less interchangeable as long as they reference the same neuron, and don’t have a difference in timing or something that makes them more sensitive or less sensitive to a relationship in the neurons that means something. So in actual fact, the synapses that are connected to other neurons are storing data even if it doesn’t mean anything. The neuron has to learn to increase the significance of what it stores over time, and this is the process that eventually results in long-term memory. So don’t worry no neuron is empty, and waiting to be filled if it has even one synapse it is already processing some signal. Most neurons in the cerebral cortex have thousands of synapses in the adult, and this does not take into account the thousands of synapses that have been created and destroyed because they had no impact on the neuron.
I like your “Fragging” Metaphor actually I have written a book called Datamining your intuition that mentions in passing the mechanism I think is involved in rearranging data, and that might be a precursor for intuition. But of course it is speculative and only lightly touches on my theory. I am waiting on Wikibooks to get their act a little further together before I try to publish it. However if anyone wants a copy, let me know, and I will brush it up and print a few off.
Part of the reason I haven’t got to the Declarative Memory, Skill Memory and Consolidation is because I have been concentrating on describing the processes of memory in the main CorticoCortical Loop because it is less complicated than the other loops and so might give us a background in how to interpret Declarative and Skill Memory. Since consolidation belongs in my model to the Declarative memory, I will leave it until later to describe.
Please check back on my past postings especially where two or three come together as those are usually where I have advanced the original essay a bit more. Some of the intervening articles might be a bit off topic, because the respondents at the time, couldn’t grasp the issues, or wanted to draw me into their own pet theories.
-
I was reviewing your comment, and my own, and it came to me that you had defined domain specific memory differently than I originally thought.
To answer with a better answer, Domain specific memory was first suggested I believe by Jerry Fodor in Modularity of Mind, although I might be wrong on the source of the idea. What he described as I have been told, not having had my hands on the actual book, was that there were these domain specific memory areas he called Modules, that didn’t require attention in order to operate, and processed shallow processing schemes autonomically.
Since then there has been a number of people who suggest that he was too cautious in his description, and have suggested massive modularity.
My own personal view is linked the the core/belt/Associative area structure, I propose three layers of modularity, the core layer is Fodors Modules, the Belt area has modules of a different sort, indirectly selected by Attention, and the Association areas have directly selected by attention modules. Now each of these modules can be seen to be a separate domain specific memory area, but because they are part of the core/belt/associative mapping, they do not need to have separate structures, except for the connections.
In fact Dispersion Spectrum Imaging of the connections between areas in the cerebral cortex, and the main network of connections down the center line of the brain, suggests that the connections are oriented towards just such a modularity throughout the cerebral cortex.
Now the problem might be simply that Architectonics at the neuron level requires multiple staining techniques to resolve to any level of sophistication, Neurons being generally too small to detect even with the MRI, so while some researchers have claimed fine grained resolution of domain specific memory areas, others haven’t been able to find the same level of difference, suggesting either a variation in technique or
over-reporting of fine structure. So the existence of a fine structure of Architectonics is currently as far as I know not accepted, while at a slightly higher level of organization the Core/Belt/Association Area level, there is agreement of the existence of significant differences in the structure of the brain. This could all be out of date, since the books I recently read on the subject were written in the 80’s.So if this is true, the Domain Specific Memories may not be noticeably different at the architectonic level, but still probably exist as parts of the core/belt/association areas.
-
As far as domain-specific short term memory goes, I was talking about domains such as vision, hearing etc for which the lengths of these short term stores has been measured.
In order to recognise something (anything) at the domain level (eg vision) the visual information must be held briefly while comparisons are made with known objects until a match is made. This requires both short term store and a form of long term semantic memory. It is well known that this type of semantic visual memory can be selectively lost indicating the existence of physically separate storage areas in the brain.
Domain specificity is much clearer for language. We know that language can be lost and even particular types of words can be lost after brain insult eg stroke. One can lose recollection of all names, or verbs, or particular categories of objects. Thus language must have its own short term and general long term forms of memory. Short term verbal memory is needed for the assembly of sentences ~ syntax, especially in that language that seems to me to be spoken almost entirely backwards (German) requires the storage of key words ahead of sentence structure.
Note that short term memory is often tested with the use of words alone. Surely it is obvious that what is being tested is not working memory or any global short term memory but the domain specific verbal memory.
A simple test of this observation is as follows. remember that around five to seven items is claimed to be the extent of short term memory. Take a subject into a room they have not been in before, ask them to look around and note as much of the environment and events leading up to their blindfolding as they can.
Now have another subject enter the room and test their short term memory. How many individual items and events do they recall from the few minutes leading up to their blindfolding? I suspect that the average recollection will include between 250 and 500 items but with priming, between 500 and 1,000.
People will remember the approximate size of the room, colours, floor coverings, light fittings, shelves, some of the contents of the shelves, the placement of doors and windows, conversations leading up to the blind folding, details of the person leading them into the room including details of their clothes, their facial features, their voice and at least some of their words, they will remember the events that occurred to them and so on. As they forget most of these details after a couple of minutes then, by definition, those details must have been held in short term memory.
If your general short term memory were interrupted just at the point where the blindfold were placed on you, say by a cleverly administered electric shock, then the following deficits would be noted:
The subject would not know where they were or how they got there, would not know what time of day it is, would not know why they were there, would not know why they were blindfolded, would not know where they were supposed to be, where they planned to be, or where they were going to next. Indeed, the individual would be totally disorientated.Orientation in time, space and daily schedule requires a quite prodigious short term memory, a lot more than five or six recallable words. We carry forward huge amounts of information, as is demonstrated by the room (thought) experiment which is a little tricky to carry out in practice because the list of contents of the general short term memory takes so long to list that the information is long gone before the list is even 10% complete, assuming a short term store of less than a minute.
The priming mentioned above, by the way, is related to questions the researcher might ask eg “do you remember anything about the light fitting?” “Yes, it was a long double fluorescent light with a diamond patterned covering.” (five items recalled: the type of light, the size of the light, the number of tubes, the fact of the covering, the detail of the covering).
It is really quite obvious just how persistent an incorrect method of investigation can be ~ verbal short term memory is tested and assumed to be general short term store when any doctor who administered or witnessed the administration of ECT or even general aesthetic will have noted how disorientated people are when they come to, having had their short term memory wiped out ~ “where am I”, “who are you”, “what am I doing here” ~ these are the pieces of information that the short term memory seeks immediately. It doesn’t seek phone numbers or lists of words…
Robert
-
Actually we are talking about the same thing, I just didn’t explain myself fully.
The core/belt/Associative areas are arranged in the brain according to sensory modality. For instance the occipital cortex has a core/belt/associative block associated with the visual modality. The Core/Belt/Associative block in the upper temporal lobe is associated with the Auditory Modality, the Parietal Block is associated with the prioperception modality etc.
There is of course a need for a two-source memory, but that is well covered by the interface between the core and the belt. I was just working on that, at the architectonic level, today, and described a connection via Laminae IIIc that might explain how such a connection is possible. It however does not require an actual buffer, just a capability to address both belt and core at the same time, and an interface in IIIc that transfers the information from the core to the belt via local fibers. It has the added effect of suggesting that the size of pyramidal neurons is affected by the distance they must transmit data, since the size of the pyramids on the core or transmitting side of the fibers is larger than the size on the belt or receiving side.
The interesting thing is that this model works with the idea that the formation of neural groups is controlled to some extent by the lower layers of the cortex, as is suggested in my model. To see the difference that it might make we just need to look at the difference in architectonics between the core and the belt in the visual cortex, where the superficial layers are very similar but from IIIc downward the relationships between the other layers are shifted as if IIIc was added into the core, and then some deeper layers changed. It is quite an abrupt transition by the way. If you want to look at it, look at figure 22, the bottom plate, on page 71 of Heiko Braaks Architectonics of the Human Telencephalic Cortex, Springer Verlag (1980)
ISBN 3-540-10312-0 or 0-387-10312-0 depending I would guess on where it was published, Germany or New York.The reason we don’t need a buffer, like I have said, is because Dr. Laberges Triangular Circuit of Attention, has been shown to work not only with short-term attention but with extended attention, and since the data is held in two locations at the same time, it can be interpreted by the second if the second also includes feedback from the first. (via IIIc)
On the subject of short term memory, My version of short term memory is quite capable of storing large numbers of objects, as long as only 7-13 of them are accessed within a 3 second window. The exact number being a function of the time it takes to refresh them by rehearsal. In fact, the whole idea of short term memory has been mismanaged because of our assumption of serial input. The brain is massively parallel, using Dr. Laberges Triangular circuit of memory, it is possible to access in parallel a zone of environmental stimuli, rather than a single element.
During the Second or belt stage of analysis, this zone can be separated into elements according to salience. In other words it only takes one of the spots in short term memory to store hundreds of elements, if we extract them from the basic salience zone. It then can take time, but not all that much space to select elements and extract them from the larger data cloud. Your objections do not fit the model I am suggesting.But then by your bitter comments at the end, there I think you haven’t really understood the model I am trying to present here, and so you have assumed that somehow it is the same as similar previous models, I really urge you to go back and look at the double postings, there is more to this model than I managed to put down in the original essay. This model is about how to build a memory system from the nerve up, the memory system is much larger than I could have possibly explained in the original essay.
-
Graeme,
Your comments on short term memory appear to be a little contradictory and confused.You say that there is no limit to the number of items that can be stored in short term memory, but then make the arbitrary restriction of 7~13 items and the refresh rate of three seconds. And further, you mention the massively parallel nature of the brain and yet appear to forget that parallel nature when recollection from short term memory is considered.
A person exposed to thousands of bits of information, such as a furnished room, recalls thousands of bits of information with recollection of the room. The image is made up of numerous bits of information which can be catalogued by the individual, but in practice there is a limitation because short term memory is short term and the process of cataloguing items will clear the STS fairly quickly.
But why would a massively parallel brain be restricted to just ten items when there are around ten domains tracked in STS alone, regardless of the number of items in each domain. In the room example, a person will be able to recall something seen, something heard (probably the fact of a conversation if this was important at the time), possibly smells, probably something about their own body eg an itch, their position in space and time, the place where they came from “how did I get here”, the purpose of their most recent actions, the goal or direction to which they were heading at the time they were interrupted (with the blindfold and having their STS queried) and so on.
All of these domains are presented, recalled and refreshed in parallel. And that is one point on which we fiercely agree. The received view of cognitive science is that at the peripherals the brain is a parallel processor but consciousness is serial. And with that I disagree ~ consciousness is very parallel.
I didn’t think I was being ‘bitter’ in my comments. Pointing out that the commonly held view that only a few items are held in STS memory when in fact hundreds actually are is merely an easily provable fact, as I have outlined. And it is something I stumbled upon quite recently, when reading up on current memory research. I have been a little perplexed at why some of the other commentators I have discussed this issue with hold to the low number of items and yet can not explain the room example I gave, as if STS in its natural working environment is irrelevant to the clinical procedure of presenting several words or numbers to see how many can be retained is somehow a more valid measure of STS ~ it isn’t.
The first and foremost function of short term memory for all animals that have a brain and a STS is to orientate them and to maintain orientation. We can catalogue what needs to be know, things like “why am I here”, “how did I get here”, “why did I come here”, to “am I tired” and so on. One can observe this process even in such simple activities as anesthetising animals and watching them as they awake, completely disorientated ie there is nothing in their STS. Once information relating to basic orientation is in STS the animal begins to relax – before then it will be looking around (to see where it is ~ this information is not in STS) and so on.
STS also performs the functional role of a buffer, holding information until conscious attention can be paid to it (or dropping the information after some interval). A buffer is a functional role that may or may not require dedicated brain areas to achieve the required holding or delaying of information.
Robert
-
No I am not confused, I meant what I said which is not what you thought I said at all. However perhaps I didn’t express it as well as I could have, and thus triggered your expectations that I was saying something else.
The problem lies in the fact that you have missed much of the discussion on the nature of implicit memory. Implicit memory outputs a data cloud of parallel and redundant data, and what passes into the bottleneck, is while only a fraction of that, still a data-cloud rather than a single memory. So within the data cloud there can be references to multiple objects, but only 7-13 data clouds can be accessed at a particular time (within 3 seconds) The exact amount seems dependent on the search step in the bottleneck.
Once the data is converted to a CHUNK, it can be recovered by rehearsal. This means that the 7-13 value is a limit to the number of chunks that can be rehearsed at any one time.
To extract an element from the data-cloud we need to edit the chunk to eliminate other data, which requires that we edit the chunk list during rehearsal. Thus we must store the original chunk, plus rehearse it with editing which gives us a new chunk that is smaller, if that chunk only contains one element, then we can say that we can retrieve 7-13 objects at a time, but that potentially we have access to hundreds of objects in the original chunk. So your confusion happens because you don’t understand this new form of chunking.
The reason why we have limited retrieval, is because of the nature of the bottleneck conversion search, which imposes a serial dependency on the data, that limits how many rehearsals can be done in 3 seconds, which is the approximate time it takes for the data to begin to degrade. Probably due to Habituation.
Believe me, I do not impose this bottleneck myself, it is a measured thing, and turns up on many different psychological tests.
The trick is to understand what it means, and in this case why I can agree with you that you observe multiple objects and list them beyond the so called limits of short term memory. The difference is that I am discussing a mechanism that explains both understandings but negates them at the same time.
The potential for storage in a CHUNK is unmeasured, however when we try to retrieve from that chunk, the number of elements that we can retrieve depends on rehearsal periods, we must rehearse them if we want to retrieve them, and we must rehearse the chunk if we want to analyze it for contents.
Miller back in 1956 attempted to find the size of short term memory thinking like you that it acted as a buffer, and failed. Like you he tried to use information theory to define it in terms of bits, and failed. What he did find was that there was a limit, his magic number 7 plus or minus 2 that seemed to indicate an upper limit to how many objects could be referenced.
Since then, there have been numerous studies that showed the limit was actually smaller than he suggested, and a few that proved that the limit was larger if you for instance did a digit span test of university grads in Chinese.
In the 1990’s it was shown that the extra digits that Chinese students achieved were the result of some link between phonological factors such as the number of syllables in the name of the digit, and a 3 second window for rehearsal.
Now this was all done assuming that the STS was some sort of buffer, like you seem to think. In fact we have developed workspace theories under just that assumption, that the brain chalks things up in a separate memory space while it is working on them.
Problem is that no one has found these purported buffers.
On the other hand, they have found evidence for extended attention. Now the effect might be similar but the mechanism is different, and that is important to know because we can quit looking for something that isn’t there, and start looking for something that is.
I was amused at your description of the early waking state, but question your assurance that it is the STS that is the holder of the references needed to build the gestalt of location. If the STS was all that was needed it would have been filled in the first 3 seconds. Look deeper into the memory for these functions please, do not expect the STS to do everything.
-
Graeme,
You have done nothing to dismiss the observation that orientation is a function of short term memory except to state that it does not fit your theory. All the information needed for orientation is perishable and relevant only for a very short interval. They are the essential criteria for short term memory. Where else do you suggest one looks? Long term memory perhaps?It is a fact that:
orientation information is essential;
orientation information must be continually accessible;
orientation information must be constantly updated.How, then, is orientation information not held in short term memory??
Your description of information clouds is vaguely similar to what I have described elsewhere as mnemonic memory strings. I have written (in informal essays to the various discussion groups that I own) that we have access to around ten unique strings. I estimate these unique strings would have evolved from sensory modalities eg hearing, vision etc but in humans they are no longer fixed to any one domain.
If short term memory is holding information in lieu of processing then it is performing the function of buffering. A buffer differs from a memory in the following way:
Memory has the function of storage and retrieval;
A Buffer merely delays or holds information between two domains eg information from the senses, information from various long term memory stores.If the consumer of Short term memory is not the agent that deposits information for later recall then that Short term memory is performing a buffering function. Working memory, on the other hand, is where information used in a decision making process is briefly stored for later retrieval by the same process (the decision making process).
As short term memory holds information from a number of domains to be retrieved by a domain other than the contributor of that information then it is performing a buffering function. I assume that it performs some genuine memory role as well.
As for the information cloud, this appears to be an attempt to resolve the many pieces of information that is available verses the smaller information handling capability of the process that makes decisions. These sorts of theories make the implicit assumption that sensed information remains pristine until it is utilised. The reality is that information does not merely degrade but is processed into ever more general summaries. These summaries are not necessarily information clouds but are genuine summaries that may or may not lead back to the originally sensed information in a less degraded or processed form.
The most common form of top level summaries are in the form of emotions, feelings, moods and so on. Thus your information cloud is subjectively experienced as a kind of feeling or emotion. For instance one may read a novel and then, at some later date, be reminded of it, say a friend asks if we have read it.
Now you might say that our first recollection is an information cloud that represents the total recollection of the book. What the average person is more likely to remember is a combination of emotions that act as the summary of the book ~ how that book made them feel. They like it/don’t like it, it made them feel this or that. As one recalls more one remembers characters and how they made you feel, events and how they made you feel. Eventually one end up with an informational list such as a list of characters and the events that occurred to them.
If we only recalled an information cloud (the facts that made it into long term memory) then we would recite the list of characters and events much like an autistic individual would and we would have great difficulty summarising events.
On the other hand, in an emotion only model or an emotion only individual we would be able to relate only what we feel about the book and would be unable to give any details.
Clearly we do both, but what is the natural mix and which one is primary? We know that in high functioning autistic people that appear to have an absence of social emotions they are unable to summarise (which would require emotional evaluation) but are good at listing facts. Emotional people have the opposite profile. But is what forgotten last as memory fades, for instance? Do we recall the name of a book and author but can’t recall what we thought of it or do we recall the feeling associated with a book but can’t remember the title?
The answer is that both are true, but vary with individuals and even with various subjects with which one may have a more emotional or informational attitude.
As for finding buffers, not finding the neural mechanism is no evidence of non-existence. If it was then we would have to doubt the existence of long term memory as no-one has even managed to find one of those either. Nor consciousness (think of the hunt for NCCs). Nor the self. etc
On the subject of informational clouds, let’s take the example of recollection of a non-fiction book, say a book on motorcycle maintenance, upon gaining the access to the cloud, how is the information prioritised? I’m not familiar with any person who can remember such things in a logical order (page one followed by page two etc). The only form of prioritisation is related to the feelings one had at the time of reading with those sections having the greatest emotional impact being recalled first eg those passages which one found interesting (a feeling of interest was stimulated).
I also note that if there are bottlenecks in the brain the statement ‘the brain is massively parallel’ is not a valid generalisation. Some parts of the brain are, some aren’t (those areas where bottle necks are found are not massively parallel). One might say that ‘the brain as a whole is massively parallel’.
Robert
-
It is a fallacy that there is only short term and long-term memory.
In fact the drive to find short term and long term memory has shown that the main difference is which chemical circuit in the nerve, is activated by which synapse.
While there is undoubtedly short-term memory in locations that are claimed to be long-term memory and long-term memory in places that are claimed to be short term memory, the whole concept needs to be revisited.
Orienting, in the sense of finding your references so you know where you are, instead of Orienting as in the sense of turning to face the most salient zones for perception, is probably a function of the PFC, not the main memory loop. Thus it would fit in a specialized memory type, not in general short term memory.
Recently it has been suggested that the PFC is a storage place for memory related to how to deal with novel situations. If this is true, then it seems a likely place for orienting information to be stored. Certainly damage to this location will result in loss of orientation.
This has the added cachet of being involved in the regulation of action, and so is involved in the short term memory as well.
-
Graeme,
you are showing up the difference between functional and physical models. A functional model catalogues various functions regardless of whether separate physical brain areas are required for them whereas physical models catalogue various brain areas even if they share the same function.No doubt a complete model of the brain includes both.
Functionally, any information held transiently is short term. One considers this transient memory associated with various functional modules that are identified by unique informational types. The evidence of these types can be measured by varying short term store for different modalities eg for sounds heard verses things seen verses words and so on.
To establish a functional module for orientation we could measure the duration of orientation information. Further, can the same physical area store types of information other than orientation?
I think that there must be domain specific short term memory associated with long term memory and this is partially what causes confusion. For storage of information into long term memory it must be held for some period whilst long term memory is addressed. This would make that form of memory more rightly designated a buffer. A similar functional module is required for retrieval. The so called hippocampal circuit may also function as a buffer where information is circulated for some period of time as it is stored or retrieved.
On the data cloud, there may be some subjective experience of such a cloud. The mnemonic string I mentioned is essentially a two dimensional version of a three dimensional cloud. The mnemonic string is conceptually modelled on the mnemonic method of memory storage and retrieval with the assumption being that the learned technique is merely a conscious version of the common subconscious technique used for memory storage ie one thing recalled is related to some other thing, so ‘pulling’ on the string, so to speak, can pull up numerous related bits of information.
You can ‘make’ a three dimensional cloud from a two dimensional string by realising that numerous strings can be attached to any data point and that the resulting array of strings can form structures so that pulling up any point can lead to an entire informational ‘cloud’. An example may be a discipline one has learned eg if we have learned mathematics and we think ‘mathematics’ we gain access to a cloud, from any point we can find our way to any other point without having to pass through every intervening point (as one would with a string).
On the other hand, a string is the logical form of recollection of events and for recollections of newly learned information (before a structure or cloud has formed).
In my model, still very much in its construction phases, there is a one dimensional and four dimensional version as well, making up a total of four. The one dimensional version is any trivially recognised thing, eg a word, or any point on a two, three or four dimensional recollection. The fourth dimension is any recognised thing that immediately leads to some number of ‘clouds’. The recognition of a close or well known person, for instance, may recall many such clouds, but for most recognised people we recall a 3D cloud. The difference can be better understood when people demand to know what you are: are you a scientist? a gardener? A liberal? These a three dimensional structures that can serve as a template for a particular person. Those you know well are many things ~ you know them by their career, by their friendship, by your shared experiences and so on, all clouds in their own right.
The emotion side is just as rich in theoretical structure and probably has more in common with non-human species than the simple informational models. Emotion is able to summarise, simplify, and aid rapid response. But what is an emotion, a feeling and so on is where we are at ~ very crude beginnings indeed.
There is no point in bickering over details when there is so much to do :)
Robert
Results
-
