User:Graeme E. Smith/Multi-Phase Attention

Multi-Phase Attention Some phases of attention and how they might work  Graeme E. Smith, GreySmith Institute of Advanced Studies  http://en.wikiversity.org/wiki/Portal:GreySmith_Institute   http://en.wikiversity.org/wiki/User:GraySmith_Institute  grysmith#telus.net

David LaBerge has been roundly criticized for the lack of depth to his triangular theory of attention, Care should be taken not to throw out the importance of his contributions, in the zeal to capture more of the attention system than his work could possibly achieve. His work is limited only because he attempts too hard to force it to account for awareness. Attention is not primarily about awareness, although it contributes to it, and so by focussing his interests so tightly, he ignores significant contributions by others. This work is based heavily on my own unique interpretation of what is happening in the memory system, and so it combines many different models of attention into a single attention system having multiple phases. 

When David LaBerge, designed the theory of the Triangular Circuit of Attention, his concern was how attention affected awareness. While this is no doubt important, Attention is by no means all about awareness, in fact awareness is merely a side-effect of attention. Attentions main purpose is to direct information between different areas of the brain so that it can be properly managed. In this article I will introduce a Snail Model of Orienting, the Functional Cluster model of Implicit Attention, three different forms of Dr. LaBerges Triangular Attention for different phases of explicit memory, an inductive attention system for skill memory, a parallel system for the meta-index that looks like LaBerges Triangular circuit but involves different organs, and a multi-pass and sleep mode model for implicit loading of the meta-index, and for consolidation of the meta-index back into the cerebral cortex. All in all this model suggests that there may be as many as nine separate attention phases in the brain, or possibly more that I don't know about yet.

The Snail Model of Orienting
In Memories Voice: Deciphering the Mind/Brain code Daniel L. Arkon made some staggering assumptions. First he assumed that he could learn from snails lessons about how humans thought, and secondly, he assumed that by deciphering the biological chemistry of the neuron, he could learn more about how it worked. Both worked out, and he was able to show parallels between cells in the snail ganglia and cells in the rabbit hippocampus. As well, he was able to show conditioned responses to stimuli that showed that snails learned to associate cross modal stimuli through activation of ganglionic neurons. It might be pushing things to assume that the ganglia in a human do a similar job to those in the snail, but one way of looking at evolution is that evolution conserves patterns that work well, and builds on them, while suppressing other patterns that don't work as well. If so, it might be possible to trace the evolution of the brain from the basic snail brain to human brains even if the original branch that connects them has long since become extinct. Because I haven't been trained to do that work, I will take it as sufficient that the human brain might have evolved from something a lot like a snail brain. With this in mind, I see the modern brain as simply being a system that added sophistication in areas where it was warranted, and did not affect the operation of the brain negatively. Now that I have gone that far, let me go even further. The snail brain consists of essentially three types of cells, the input or sensory cells, the output or motor cells, and the processing or ganglionic cells. Dr. Alkon's research showed how the snail could associate bright light, with a tilting of the table it was on, causing it to reflexively increase the suction to its foot, in order to hold on tighter. This effect called focusing, makes the foot smaller a visible reaction. By shining a bright light, when the table tilted, the snail was conditioned to expect that the table would tilt when the light went on. As a result the conditioned response was to focus it's foot when the light went on. Other similar experiments with snails cause them to retract their syphons, or their gills when some other conditioning stimulus is triggered. Since all these conditioned responses seem to be triggered by cells in the ganglia, it is reasonable to assume that the ganglia are where the conditioned responses meet the reflexive responses, and where sensory data of different types are combined. Since snail brains don't include much more than these three types of cells, we can say that the ganglia is where the response to stimuli is first to be found. In other words, it can be said that the ganglia is the conditioned and reflexive first response mechanism, at least in snail brains. So let us look at a specific brand of Attention, called orienting. Orienting is the tendency for an organism to turn it's sensory systems to point towards a salient stimulus. We see this effect when a sound of broken glass is heard at a party. Almost everyone turns towards the sound. This is like the snail turning it's head toward a flash of bright light. If the snail can do it with only the ganglia, perhaps humans can do it as well using primarily their ganglia. One way of looking at this, is the narrowing of information concept suggested by Dr. Igor Aleksander. In his work on Machine Consciousness he has implied that attention is a search required whenever the datapath between the environment and the processing system narrows or widens. Essentially a way of remapping a limited channel. A similar concept is explored in more depth by Naotsugu Tsuchiya and Christopher Koch in Attention and Consciousness If we think about it, the sensory system is a narrowing of possible data. If our body was made up completely of light sensors, then we wouldn't need an attention system to focus on light stimuli. However we get our knowledge from a maximum of two eyes, and so we have a narrowing of the data channel between the Universe of light stimuli, and the brain that must somehow make sense of the light stimuli. So our attention system is designed to do a search for the most salient visual sensory data, and focus our eyes on that source. Because orienting is cross modal, we focus our eyes on the source of the broken glass sound, as well as on flashing lights. While this movement is instinctive, it can however be suppressed by higher attention levels. I call this the "Snail Model" of Orienting.

Functional Cluster Model of Implicit Attention
In The Remebered Present: A Biological Theory of Consciousness Gerald M. Edelman a leading researcher in Phenomenal Artificial Consciousness noted that the concept of a Functional Cluster bound neurons from across the cerebral cortex together and that we could detect this because the clusters resonated at the same frequency. Later research indicated that the signals involved were Gamma range (40 hertz) synchronous Oscillations (GSOs). Later research indicated that GSOs were implicated in long-term implicit memory, and that they did not act like a radio frequency band, nor did they represent the difference between foreground and background elements. These later discoveries seem to fly in the face of the idea that GSOs are used by the Prefrontal cortex, in order to select memories from the implicit memory. Yet there is also clear evidence of the involvement of the Anterior Cingulate Cortex (ACC) in the selection from among different GSOs. So why is the GSO a factor in the selection, but not a frequency band, and what is it selecting, if not foreground and background? To answer this question I need to go back to Jerry Fodor's book The Mind doesn't Work That Way!: The Scope and Limits of Computational Psychology in this book, Fodor makes the point that Phenomenal Memory can't be isolated into distinct memory references. If this is true, and I maintain that it is, at least until some conditioning is done to the network, then, we shouldn't expect implicit memory to pick out foreground elements anyway. What I think is happening instead, is that the implicit attention system is isolating zones of salience, which harks back to my snail model, and suggests involvement by the basal ganglia, in determining the zones of salience from which to select. If this interpretation is true, what a functional cluster is, is an area in the environment that has salience that is isolated from other areas of salience by the particular frequency of the GSO. The role of the ACC becomes merely filtering the output probably by suppressing everything that is not tagged with a specific GSO associated with a specific salience value. In other words not only does the snail brain orient the body towards salient stimuli, but it tags the environmental zones of our perception with a tag associated with a salience value, that allows us to prioritize our attention to the areas of greatest salience. This is all we can expect from implicit attention. To get further meaning out of the data we would have to break down the information in that zone into sub-elements and analyze the sub-elements, which Fodor assures us can't be done in a phenomenal system. In Annette Karmilloff-Smith's book Beyond Modularity: A developmental perspective on Cognitive Science she notes that a process called Representational Redescription seems to happen between implicit and explicit memory.

The Bottleneck and mapping clusters to chunks
In my article The Dual Mode Cortex, I note a method of addressing that allows demand type memory to access phenomenally implicit memory. In my article The Bottleneck: a New Perspective I explore how top-down attention selects via GSOs for individual Functional Clusters and converts them to Chunks which are lists of mini-column addresses a la 5th Layer Apical Dendrite Connections to the Thalamus as suggested in Attentional Control: Brief and Prolonged by David LaBerge. The basic concept here, is that the Chunk, is a demand type memory address that defines a Functional Cluster previously selected by implicit attention. To retrieve the data field that the functional cluster represents, you simply present the chunk array to the thalamus, and trigger the output of a similar data-field. I have chosen to call this data-field a Quale, since it answers the questions that are asked about Qualia, by creating an opportunity for a single Chunk to access a broad rich data source. In fact the main problem with this Quale definition is that there is a need for it to be broken down, but because it is produced in a phenomenal mechanism and forms an unstructured cloud of data, it cannot be broken down AFTER rehearsal.

Sub-Qualia by Sub-Chunk Arrangements
Dr. LaBerge noted a connection between the PFC and the Thalamus in his Triangular Circuit Theory. Later researchers noted that the connection passed through the Nucleus Accumbens and the Nucleus Reticularis Thalami, a sort of inhibitive layer that surround the thalamus. What I have interpreted this to mean, is that the PFC can choose to rearrange the Chunk addresses, by suppression of specific mini-column activations that would normally be part of the Chunk. This would be relatively useless, if the resulting functional cluster were not analyzed for salience and given its own unique GSO. However by doing so, the brain assures that such sub-chunk arrangements can be evaluated for whether they increase the salience of their new functional clusters over the original chunk or not. The Quales they produce are therefore already tagged for greater or lesser attention. this ability to isolate sub-quales that have more salience from their larger salient zones, is critical for picking out foreground from background and associating specific memories with specific objects. It is only once that we can do this, that our perception becomes specific enough to pick out individuals, or individual objects. It is partly because of this secondary perceptual analysis, that I have suggested there must be more than one phase of modularity in the brain. The secondary perceptual analysis field must only be practical after sub-chunks have formed sub-qualar data elements. Since this is dependent on the suppression of mini-column addresses done by the attention system, secondary perception must be passively guided by attention.

Complicit Attention, Access and Recruitment
Up until this point in the memory system we really haven't been able to process the information so much as analyze it, and lump all the analysis steps together into a Quale. Ideally at this point, we would want to segregate out specific analysis results and direct them into specific responses or functions that might further process them. or make our body move in a some manner. For instance, secondary perception areas might be capable of detecting the likely presence of a face. this works so well that we even can detect analogous relationships between windows and doors so that houses look like faces. If so, experiments have shown that there is an area of the brain called the fusiform facial area, that actually goes the next step and tries to recognize the face. This space does not light up, except when the brain might be looking at a face. It is therefore Recruited or accessed in a content dependent manner. This step can't be done until the face can be detected which means that the memory has to have already gone through the step that breaks down into sub-qualia, the functional clusters quale. To make this work, first we must rehearse the sub-quale, we want, and then we must somehow direct it's content to a particular module such as the fusiform facial area, that specializes for that particular type of data. To achieve this we need two registers, one for the address of the data to be analyzed, and one for the address of the function to be used for further analysis. This forms something I call a value-function tuple, which is equivalent to a command. The collection of all such functions and actions that might be triggered, would define a language. One way of doing this, is to put the data into a short term memory element, and then feed that short term memory element through some sort of network to the function we want. A possible location for such a network might be the corpus collossum which is associated with the Prefrontal cortex, where we think short-term working memory might reside. I have determined that processing in a neural network is actually automatic, given data, it will be processed, and an output will be formed. However the pre-activation of an area makes it more likely that the data will be given a high priority when it is added to the inputs of the area, and therefore that the outputs will be significant. Thus the Thalamic connections actually act to allow a range of processes to be activated with a single data quale. The Qualar output of these connections is again tested for salience by the basal ganglia, and because it allows GSOs to be attached to each area and supplies a salience value, it can guide the further processing of the outputs to favor the most salient responses.

Inductive Skill Memory Attention
In order to understand this attention system, we have to start with a tautology, that the brain sees no difference between triggering a motor neuron, and triggering a function as described in the previous section. In both cases it triggers a data rehearsal, and then triggers a specific pattern of mini-column addresses to pre-activate either a motor neuron, or a function. This in turn, triggers a connection to the cerebellum, where the context of the choice is fed into the mossy fibers of the cerebellar cortex, and signals from the inferior olive are used to trigger Purkinje cells via the climbing fibers, thus causing them to connect the context to the action chosen. At some point, merely the presence of the context will cause the action sequence to be triggered. Another circuit creates a learned-conditioned response that allows maintenance reflexes such as dynamic balancing. One of the interesting things about the cerebellar cortex, is that the same purkinje cells can absorb a number of contexts allowing them to release different sequences when they detect different contexts. this means that similar contexts might trigger multiple action sequences that if they all were expressed at the same time would cause confusion instead of skill. This is where Top down Attention needs to be applied. If each different sequence is given it's own GSO, then perhaps selection from among the sequences could be achieved by simply using the ACC to suppress all the GSOs that were not selected. There is a role here for the Ventro Lateral PFC that has been traced to connect to the Secondary Memory Area, an area that consists of two sections, one that detects new sequences being produced by the cerebellum, and symbolizes them, and one that models the likely effects of allowing the individual sequences to complete. the result is likely a regulatory impulse that prioritizes the choices for the ACC. It is thought however that sometimes this regulatory impulse does not trigger an obvious favoring of one sequence over another, and in those cases, further processing is required. It is thought that in some cases resolution can be automated, and in those cases the context of having two sequences deadlocked, triggers the automation sequence that determines which sequence will run, we call this concept intention, because it does not involve conscious selection even though selection is done. If there is no automated process control sequence that will resolve the problem, it is thought, consciousness is engaged in order to create one, and the result is an effect we call volition. the end result of this effect, is the eventual release of one of the sequences, or a change in the context that causes a new range of sequences to come to bear. A side effect of the way the cerebellar cortex works is that the brain learns macros or sequences of commands in its base language, and since it can associate symbols with sequences of commands, such a construct is called a Macro in computer langauges. The brain can essentially trigger them at will creating something called a Macro-Language. The Macro-Language can be integrated into sequences of elemental commands causing more sophisticated sequences to be implemented simply by expanding sequences of Macros out into the elemental language through a process called an interpreter. Thus the brain is capable of higher order processing as if it had at least a medium level language.

Declarative Memory and the Meta-index
There are three things needed for Declarative memory, (Memory we know, we know)


 * There must be a demand memory system so that we can demand a memory.
 * There must be an index that we can use to find the memory
 * There must be a signal that the memory has been found in the index.

The Demand Memory System
There is no doubt that there is a need for a demand memory system, and although there is still some room for doubt as to how that system is implemented, I hope I have shown that the dual mode cortex, the bottleneck, and the early attention systems discussed so far, are enough to implement such a demand memory.

The Meta Index
When it comes to the existence of a Meta-index, or index of indexes, the candidate area of the brain is the hippocampus area, in which 4 separate pyramidal memory systems CA3, CA2, the Entorhinal Cortex, and the parahippocampus, seem to be integrated by an organ called the subiculum. All these areas, the subiculum, and the thalamus all project into the striate cortex, suggesting that the striate cortex is a memory area that remembers bottom-up addressing templates. However it is the organs in the hippocampal areas that are associated with specific types of data, such as the so called place cells in CA3 that seem to map location onto the environment. These can be contrasted with place cells in the parietal lobe that seem to map location onto the stage in the path the animal takes. It is important to note that the environmental mapping has to be derived from the parietal sequencing and some orienting sense as to what direction the organism was facing at the time it experienced the stimuli associated with a particular environmental area. The hippocampus is definitely implicated in the orienting sense since animals with reduced hippocampal function also seem to have a reduced sense of location within the environment and thus spend more time searching for landmarks than do animals with an intact hippocampus.

Metacognition and the Feeling of Knowing
Something that might be as simple as a success indicator for a search of the meta-index, the Feeling of knowing or FOK as it is affectionately called, is an indication that the mind contains memories about a particular subject. False positives, suggest that in fact it is only an indication, and that effects like Tip of the Tongue (TOT) might be required for cases where FOK is positive, but no valid memory can be found where the Meta-index suggests it should be. This might be possible for two reasons,


 * 1) The memory may resemble another memory but may never have been laid down. In this case instead of a feeling of knowing we would expect to get a feeling of familiarity. Which might simply be an indication that the area the memory is in, has been primed recently. When we get both a feeling of knowing and a feeling of familiarity we will be convinced we know the subject of the search.
 * 2) The memory may have been indexed, but then moved within the cerebral cortex so that when it is retrieved a different memory is found. This implies that there is some search parameter that must be met in order for the memory to be acknowledged, and if it has moved an error mechanism must set a meta-cognitive signal saying it has not been acknowledged. If the entry is positive this might indicate a loss of synchronization between the index entry and the memory, which a search of surrounding memory areas might alleviate. The tip of the tongue feeling might be a report on the success of the peripheral search and any changes in strategy used to narrow the search and find the data.    The important thing to know about declarative memory is that you can't know you know something until you have told yourself that you know it.

Slow Wave Patterns and Slow Wave Sleep Stages
Research into the activations within the hippocampus area, have noted that certain slow wave patterns noted during the daytime, are repeated during slow wave sleep stages. While it is too early for scientists to be sure, one possible interpretation is that these slow waves are associated with the passing of signals from the parietal lobes through the hippocampus area, as the meta-index attempts to map signals from the experiential parietal lobe place cells, onto the episodal memory. One interpretation of this, is that the repeating of the mapping signals might indicate that the same signal needs to pass through the meta-index a number of times in order to be fully indexed. Obviously we need some method to select which sequence we are passing through the meta-index, so I call this the Slow-Wave Attention System.

The Meta-index Image in the Cerebral Cortex
Since the days of H.M. we have known that it takes nearly 2.5 years for the complete set of knowledge of an experience to be written to a Cerebral Cortex index image, and out of the hippocampus area. If the hippocampus is removed, or silenced in that time, some aspects of the long-term memory are lost. This is called a retrograde amnesia. The fact that it is limited to 2.5 years, suggests that there is some process that scientists have learned to call consolidation that moves the meta-index information out of the hippocampus area. It has been proposed that this movement happens during REM sleep. That it takes so long, is an indication that it is a housekeeping task that is low priority, and that it probably includes a serial dependency similar to the bottleneck that slows it's processing down. Evidence seems to indicate that memories that are referenced more often tend to be transferred first, and other content filled in as necessary. This suggests another form of attention that keeps track of what has been written to the cortex, so that the data eventually gets there in one piece.

Conclusion
Although, this discussion has not been as Academically Rigorous as some would like, I believe that I have shown at least a rationale based on my current understanding of memory for a 9 phase attention system. There is reason to believe that in fact the attention system is more complex yet, since I haven't attempted to deal with Awareness, Intentionality, or Volition in any depth. This multi-phase attention model is not a complete model of attention but includes known attentional effects and parallels known neural correlates according to my admittedly naive viewpoint. Dr. LaBerge himself has expressed some disquiet with the portions of this that I have sent to him, so I know it will not receive complete acceptance, but I hope it will at least open some questions that have heretofore been left closed.