Bert G. J. Frederiks
The Time MachinePrototype of a Conscious Machine
6 The content of attentional neural machines
In this chapter I want to delve deeper into some facts, problems, and thoughts that one may think of with regard the content of my neural machinery in relation to neural hierarchy, activation hierarchy, immanence, and attention. I want to explain that, even without the inborn, biological neural hierarchy, the attention mechanism will create more or less hierarchical structures qua content.
The biological hierarchy is very small, but the fact that at the top only a few neurons can be active makes the relation between neural levels important with regard to its content. It also complicates things, since what one may attend to, may be represented at lower neural levels too, and vice versa. So qua content neural levels overlap each other.
Temporality and recurrence of images
To show you why some seemingly boring topics are important to me, I want to give you an introduction into my ideas on consciousness in the form of, what I call, temporal images, or, if you like, and maybe more precisely, partially recurrent, temporal images.[^]
[^]The concept of a programming language, from a mathematical point of view, is that it is something which is capable of calculating partially recursive functions. So, if I say that an image is like a function, then my machinery is a computer program, though not every computer program will behave like my machinery, of course. For a long time this seemed to me far out of odds. But if I consider that an ordinary computer uses, from moment to moment at least, many, relatively simple functions for a certain task, while our brains do the same with one, or a few, big, and complex functions, and further more, of course, that our brains have inborn, self-programming mechanisms, then I see no conflict in principle anymore. This said, I still do not really see how the idea of a computer program could help to gain an understanding of the working of our brains.
Let us try to forget the planning and programming mechanism, and also language again.
To put it simply, the attention mechanism is easily exhausted. It will, on itself, only ‘hold on to something’ for a few seconds – typically five or six seconds, I think, in humans – and then its attention will shift to something else. This works a little more complicated and also the five attention ‘spots’ (limiting myself to visual images here) do not necessarily shift simultaneously, but it is clear that, because of this, there is a possibility, that a temporal image, that is, an image in time, arises. I say “there is a possibility” since for this to be so, something else is required, namely circularity, or, especially with regard to speech and music, punctuation.
A string-in-time of images is not yet an image itself. One thing which, in my opinion, characterizes an image, is the freedom of movement that you have within it (or that it has within itself, if you like). By this I mean especially the freedom to move back in time, to something that you have imagined earlier. I do not necessarily mean this in an episodic sense, although this is a possibility in humans, of course.
This traveling back in time, to something which you have seen or imagined earlier, and attentionally recognizing this as such, is a moment of punctuation. At such moments connections in time are made in the ‘here-and-now.’ Time itself, when necessary, may be part of the meaning of this connection, but this is a rather complex phenomenon, part of which might be specific to human beings.
Traveling back in time is not possible in every interpretation since there also is an irreversible process called ‘learning’ at work. You cannot step into the same river twice.
Such an existence in time, or temporal existence, of an image is only possible because of a special organization of the neural network and its content, which for now I will see as being mainly the consequence of the attention mechanism.
As such I am claiming that the temporality of an image does have a material, non-temporal ground, next to other grounds.
It is as if the neural network is the ground in which the attention mechanism dances and plays, and in doing so, frequently it crosses places where it has been before.
Despite the circularity, the temporal image as such exists only in time. Nevertheless, especially human beings can self-consciously remember ‘the big lines,’ or ‘the general picture,’ or poetry, or music. So there is, as I said, some kind of a non-temporal ground. The point for me now is that, even though we have rituals, stories which we keep telling, and so on, one cannot say that there is a really ‘hard,’ absolute, non-temporal ‘memory,’ such as the celluloid of a movie is.
Non-temporal sources of temporal images
Disregarding inborn abilities there are at least two sources for a more or less non-temporal memory of a nevertheless temporal image.
With animals we share the ability of neural networks to abstract “the big picture” out of our environment. We could view this as a kind foundation on which other things can be build, except that this foundation, and the building on it, arise together. Another metaphor for this source is that it is like a landscape in which one has placed landmarks, and in which can wander around.
How a temporal image arises from such a static ground, and vice versa, requires more explanation, but for now I only want to point out that the attention mechanism, while wandering a landscape, will often cross over old terrain, which ensures the partial recurrence of the image.
One has to distinguish this source from another source, being that human beings can remember something more or less strongly at will, which we do with the aid of the programming mechanism. With regard to our mental abilities compared to other mammals this mechanism is probably the only apparatus that really distinguishes us from them. Realizing this makes that eating animal flesh feels like cannibalism to me.
Animals do not seem to have this kind of will-like thing, a will that can act on its own thoughts. They can have some discipline over themselves, but it seems to be more limited. This does not mean that there is not a possibility of a fixed sequence of actions ‘automatically’ repeating itself in animals, such as seen in a circus. One can compare this in humans with memory based on association and repetition. In animals certain temporal things can be burned into their brains with the help of food and punishment, but such ‘neurotic’ behavior can only be taught to them by humans. In such a case one cannot always speak of a consciously experienced ‘memory,’ I think, since this process of remembering consists of a chain of ‘memories,’ of which each one activates the next, but the next of the next can only arise after the next, with none of these associations having any real meaning. As such there is no circularity, let alone reflection. The same often happens to me when I play the violin, when my fingers play what I have trained, but what I cannot otherwise remember consciously. The use of consciousness and attention in learning to play the violin is, by the way, very interesting. There is a lot of practical knowledge on this among violin-players.
Aneural network as a multidimensional system
Real world neural networks, like our brains, may physically have only three, what mathematicians call, “independent dimensions,” but qua neural activation they usually have at least as many independent dimensions as there are neurons because no activation of any neuron is in any way completely dependent on any one other neuron, nor on any single group of neurons.[^]
[^]Strictly speaking this statement may not hold for every conceivable network, but in that case I claim that, in theory, there are redundant neurons – this might be beneficial in case a neuron dies, of course. Actually such a statement requires mathematical proof, of course. I take it for a fact.
The same argument goes for the synaptic sensitivity of neurons. Here the number of dimensions will be a manifold of the number of dimensions of activation, since each neuron contains many synapses. Our brains roughly contain 1012 neurons, with each on average 104 synaptic connections, so the number of dimensions with regard to neural sensitivity is about 1016.
Biological versus logical neural hierarchy
When I speak about “neural hierarchy” I usually mean the biologically fixed, inborn neural hierarchy. A little more abstractly I could refer to a hierarchy in inter-neural sensitivity. It is not correct to name a neural activation hierarchy a neural hierarchy.
The biologically prestructured hierarchy is logically equal to a non-prestructured neural network in which a lot of neural connections are made non-sensitive in advance.
Remark that the prestructured part of the hierarchy in a sense consists of relatively few logical dimensions compared to all the logically possible hierarchies in a neural network. But still, since we have to reckon logarithmically it easily cuts out 99% or much more of possible hierarchies, for instance reducing it from 1012 to 1010 possibilities.
My – implicit – claim herein is that the attention mechanism will ‘automagically’ create a neural hierarchy, even if it was not there in advance. Such a statement should be proved mathematically, but I will just ‘show’ you what it means and why it is so in this chapter.
“Gathering structures” in stead of “hierarchical structures”
The term “neural hierarchy” should in fact not be taken to strictly. Cognitive psychologists usually speak about “heterarchy.” For me the mathematical ideas of set-theory are more enlightening. In Dutch “set-theory” is called “verzamelingen-leer,” and “collecting” or “gathering” things together is called “verzamelen.” So in Dutch I would speak of “verzamelde structuren.” Unfortunately speaking about “set-structures” sounds a bit odd in English, so I choose to speak about “gathering structures.”[^]
[^]Another idea would be to speak of family-structure, as Wittgenstein did. Wittgenstein, Ludwig. 1945. Philosophische Untersuchungen, Teil I. Berlin: Suhrkamp Verlag. (In English: 1953. Philosophial Investigations. Oxford: Blackwell)
Within in a gathering structure sets are not ordered in a strictly hierarchical way, for instance in the sense that they do not overlap each other. They certainly will overlap. Further more, all kinds of sets can be distinguished throughout each larger set, based on the properties of all the elements in it. This is the multidimensionality mentioned earlier. Some thing may be higher in hierarchy with regard to one aspect, but lower with regard to another.
Objective non-intelligibility of subjective experiences
A certain neural activation structure may qua content – that is, subjectively – be hierarchically structured, but this does not necessarily mean that it is objectively structured as such too. It might objectively not even be intelligible.
For instance, something which I perceive of as consisting of parts and wholes, will usually not have exactly this part-whole structure qua neural activation, like with the parts each being active neurons at certain places, and with a neuron for the whole circumventing these, or whatever. This said, an immanent network may contain something like such a structure, but immanence alone is not enough to explain this.
Immanence means, to put it simply, that if a certain thing is perceived by a neural network, then a certain neuron gets activated mostly. This does not on itself implicate anything with regard to the activation structure leading to this. That might be completely transcendent.
However, if we introduce the principle of fatigue and attention, then the neuron being most activated will, from moment to moment, each time be another neuron, while the environment will usually stay the same. As such the network will learn to immanently recognize things surrounding something (its exostructure), and the elements this thing consists of (its endostructure).[^]
As these elements arise together they will, of course, help each other to recognize each other, and there will come into existence all kinds of heterarchical, gathering structures qua neural sensitivity, and – from moment to moment – qua neural activity.
[^]The terms “endostructure” and “exostructure” I took from Chris Sinha. 1987. Language and Representation. A socio-naturalistic approach to human development.
In a neural network with a build-in hierarchy, in which the final attention mechanism is attached only to the top level, this will work out even more nicely, since there are many smaller attention mechanisms at lower neural levels. At very low levels their scope may be limited, for instance neurons in the visual cortex will not react to sound, but above the middle levels most things are connected with most other things.
But suppose you are a brain surgeon, and you have given someone some chemical, so that you can see the immanent neurons in your patient lightning up when he sees certain things. Would the going on and off of the lights then come to be intelligible to you? Any object which is immanently recognized, is in our brains just a light, with many lights leading to this light, whether it is an elephant or a mosquito, whether it is my fingernail or my body or my house.
Specific, intelligible (gathering) structures may subjectively be there, but you won’t find them easily by – objectively – looking at neural activation, except maybe at the outer ends of a neural network. The build-in neural hierarchy in our brains, and the structure of lower-level activation mechanisms may help a little though.
Opposed to this, in stead of looking at the immanent neurons being activated, one can also objectively cause them to get activated by electrically stimulating them, and then see what happens subjectively. In certain severely epileptic people, in whom their epilepsy stems from the temporal lobe, which is rather high in the biologically fixed neural hierarchy, it happens that, when neurons in the temporal lobe of their brains are externally activated during brain surgery, they remember things such as complete, past events. But this may be extraordinary. This phenomenon may not arise in ordinary people, probably because of stabilization from silencing master neurons at higher neural levels. Maybe one could achieve the same effect in non-epileptic people if one would activate neurons hidden deeper inside our brains – in, or maybe near, the hippocampus, is my guess. Maybe this somehow is what happens in some forms of schizophrenia – I know too little about it to be able to tell, I just think it might work this way because of the destructiveness of this decease with regard to mental structures.
Parts and whole without a homunculus
All the above said, let us, for educational purposes, conceive of intelligible, neural activation heterarchies nevertheless, and think of a very nicely immanized neural network.
Take the example of things having parts and wholes. A part-whole structure is a kind of hierarchical structure. The parts are low in the hierarchy, the whole is on top. Think of your body, consisting of arms, legs, and so on, which again consist of fingers, and so on. This need not be a strict hierarchy. I think of it as what I earlier called a heterarchy, or a gathering structure, or what Wittgenstein calls a “family” structure. I may have a nose which look like my mother’s, and eyes which look like my father’s, and so on, and in the end such characteristics can somehow be attributed to my ancestors.[^]
[^]Wittgenstein, Ludwig. 1945. Philosophische Untersuchungen, Teil I. Berlin: Suhrkamp Verlag. (In English: 1953. Philosophial Investigations. Oxford: Blackwell)
Notice that it is very special to look at the world as consisting of parts and wholes. It requires attention – I mean, it requires the attention mechanism. In this case we would ‘look at’ abstract ‘things,’ being ‘formal’ parts and wholes. This is a very complex and self-conscious action, and that is not what I am speaking about here! This said, the hierarchy or gathering structure I am speaking about now is not something of which it (=the gathering structure) is conscious itself, even though its content may be part of its consciousness.
It is not in every aspect evident that a neural network can or will represent something as having parts and wholes. After all, isn’t it so that all that a neural network does, is to associate two or more things in the real world with each other by somehow interconnecting the replacing, or ‘standing-for,’ immanent neurons with each other?
Having adopted the principle of the attention mechanism, it is clearly possible that a neural network can and will attend to both parts and whole at different moments in time, simply because attention exists and because attention will shift. So the network will learn to recognize all elements.
Next one notices that, for instance, fingers are not, or less, associated with legs, but are associated with hands, and thereby with arms, and thereby with our bodies. You see that as such our body is always associated with its parts but the parts are not necessarily associated with each other in an equally direct way. In other words, just as the parts are always part of the whole, so the whole is always part of its parts, but parts of the whole are not always parts of each other, and a part has only a few wholes, while a whole has many parts. This is a hierarchy, constructed without assuming neither a top nor a bottom, not even the whole is assumed to be standing on top, so again I do not need a homunculus.
Temporal neural activation hierarchies within the biologically fixed neural hierarchy
When there is also a fixed hierarchy, then the above is not enough to explain the hierarchical neural activation or sensitivity structure with regard to something, or at least we should notice some other things.
The attention mechanism only works at the highest level. So, we may ask, do we objectively find the parts and wholes as recognized by the attention mechanism only, to put it simply, in the top layer of the network? If so, then here there will only be five of them activated at each moment. It is clear that something which is consciously perceived as some kind of hierarchical structure, in our example parts and wholes of our body, will objectively not arise only in the top of the neural network. It must extend deeper into the neural hierarchy.
To recapitulate the working of the attention mechanism: Of all the neurons firing in the deconstructing neural network, the attention mechanism finds the most active ones (five, in case of vision), and next it activates (five) neurons at the same place, but in the constructing neural network (or it activates them more than they are already activated, or it deactivates all the others). As such we temporally get a five-headed hierarchy in the distributed activation of the neural network. Since the neural network remembers some of this in its synapses, the neural network will in time, qua hierarchy in sensitivity, become a mixture of all the temporally established hierarchies.
My neural network is, as far as it is immanent, a system of ‘attended to’ neurons. Neurons can be directly ‘attended to’ by the attention mechanism, or they can, lower in the neural hierarchy, be indirectly ‘attended to,’ by neurons which are more directly ‘attended to’ by the attention mechanism.
What happens then is, that in the neural level below the attention mechanism, for each thing seen, in our case parts and wholes of our body, large groups of neurons will be activated. Lower in the hierarchy even more neurons will be activated.
Taken that the activation mechanism will only attend to five or less neurons at a time, five things will be associated together from moment to moment. This association will for the biggest part take place in the neural levels below this attention mechanism for the simple reason that there are much more active neurons there.
As our attention moves in five seconds or less from one thing to the next, in time a structure such as a part-whole structure will be build up, consisting of all these ‘associations.’
The important conclusion which we may draw from this is, that most aspects which we are conscious of are not only immanently represented in the top-level neural layer, but also in the neural level below this, and many aspects are also represented in an again lower neural level, then a few again lower, and so on. As such there is, next to the double work taking place within neural levels, also some redundancy between neural levels.
Having concluded this, one might wonder if nature has also invented something to exploit this even more. My guess is that this double representation at different neural levels is so natural to an in-born, hierarchical neural structure with an attention mechanism and mirroring, and also with top-down attentional inhibition, that nothing special is needed. The combination of specific top-down activation through the constructing neural network, combined with more overall, top-down neural depression by master neurons, does exactly what we need.
One reason to study as many aspects of conscious machines as I can find, is that I want to “program” my own, and your, lower level neural levels with as much truths as I can find, so that we get a reasonably trustworthy intuition, that will at least warn us when I do or say something stupid, through giving us some association which we need. With regard to things we do, we speak not of intuition but of talent or genius. This too can largely be trained, although a ‘good’ brain helps a lot, of course. I think that ideally, to gain the best of all possible structures with regard to some part of reality – or fantasy – it requires a certain, active, meditative, reflective, skeptical, and practical live style, if possible from childhood on. I also believe that it helps if one does not allow to much rubbish in one’s head, something which sometimes makes schools and universities, where other people try to decide what I should believe – people who even try to examine me on this with quite severe public and social penalties if I do no comply – horrible places to be, despite the fact that I like knowledge.
The attention mechanism as an association mechanism
Subjectively the attention mechanism is also an association mechanism. Subjectively the faculty of association is even more to the foreground than attention, since attention moves exactly as fast as our consciousness and leaves no memory other then associations.
An association may entail the movement within the image itself, something which gives animals a certain self-consciousness too, just like us. For this one needs time-delay neurons and feedback coupling. I do not want to mention this any further yet, as it would shift the accent of my story to this mechanism instead of the importance of temporality sec and temporality with regard to consciousness. But do notice that it is possible for a neural network to notice and act on movement itself, including movement of the attention mechanism. I think you will feel that this makes my story even more plausible.
Zooming, spatiality, and parts and wholes
With regard to parts and wholes which are visible as such there are other mechanisms than mentioned thus far at work too. Some of these consist of “mechanical,” physiological focusing mechanisms, such as those which move our eyes and control its focus. Another mechanism, which is the neural equivalent of this, allows us to focus at things on a neural base, using different neural channels than the attention mechanism. Neuro-physiologically the center of the former mechanism is, in our brains, localized in the amygdaloid nuclei, and the latter is, among others, localized in the thalamus, which is directly connected to our visual cortex.[^]
[^]See Kosslyn, Stephen M., 1994. Image and Brain: the resolution of the imagery debate. Cambridge, Massachusetts, London: The MIT Press.
In essence these mechanisms allow us to zoom in and out on things. As such there is a physiological aid for actively structuring the world into parts and wholes. Again one could conceive of this as a kind of prestructuring of our neural network, but in a rather different way.
For this to work I would need another thing to add to my machine, namely the ability to act. The acting, in this case, is to zoom in and out, and to turn our eyes. I will do so in another chapter.
Let us, for educational purposes, simplify reality. Imagine a newborn child first learning to recognize fingernails, thumbs, bodies, houses, and so on. Ones it recognizes these things, it – or its neural network if you like – can start integrating them. In doing so it will integrate them together with its zooming activities, and also together with other movements of its body. All this will become part of, among others, the part-whole structure. So perception is in again more ways not at all a passive activity. The part-whole structure will, of course, itself also become part of other structures, such as the “hierarchy” or “dimension” close-by – far-away, thereby maybe, with regard to some other aspect, becoming a hierarchy of a hierarchy.
Remark that such a relation between hierarchies or dimensions, for example between [part – whole] and [close-by – far-away], is not in every sense intelligible. It is probable that there are many such relations which, although they help us in perception and cognition, are never made conscious, even if we could make them conscious.
Remark that zooming is in itself also some ’thing’ which is being associated with something else, in this case certain parts and wholes. Above I said that part-whole structures are abstract things which require self-consciousness for us to become conscious of them. In another sense it is, however, also very concrete and close to us, and partly conscious, in the form of zooming in and out of bodily parts. For human adults this is often covered by culturally established veils – Sigmund Freud and Jacques Lacan have had a lot to say about this. This, in the end, according to Lacan, constitutes our ‘drives.’[^]
The amygdaloid nuclei, are not only the main source of “automatic,” saccadic eye movements, but also of sexuality and aggression, that is Eros and Thanatos.
[^]Jacques M. Lacan. 1977 [1966]. Écrits, A Selection. New York/London: W.W. Norton & Company.
Further more, I think one can sense that our own actions, either imagined or real, are an important part of the meaning which we attribute to something.
Also remark that these build-in faculties pre-adapt us to a three dimensional world. Since a neural network tends to go for the most efficient resolution of what is fed into it, the world, as replaced in our neural network, will tend to get a structure which is in conformity with the spatial, three dimensional world around us.
In Western cultures, and especially in English language communities, the structure which, by way of metaphoric predication, is used most often to represent abstract things, such as time, is the spatial structure. We think of minutes as being parts of hours, and so on, even though this is wrong. Not every language contains this defect in such a strong way. In Hopi, for instance, it is said to be virtually absent.[^]
You may also have noticed that I nowhere seriously speak of a “mind.” In my country, the Netherlands, we luckily do not have a word for ‘it’ – whatever ‘it’ is. The Dutch language does have much of the errors of the English language, but less so.
[^]Whorf, B.L. 1956. The relation of habitual thought and behavior to language. in Whorf, B.L. Language, thought and reality. Page 58-67. Cambridge, MA.: MIT Press.
So, this is another way of looking at (the working-out of) the attention mechanism: The attention mechanism can create a neural activation hierarchy in a flexible way. With the help of an inborn focusing mechanism, and our ability to act, this hierarchy will tend to get into some relation with our three-dimensional world. But few things really are three-dimensional, even regarding spatial matters, if only because such things have many non-spatial properties, like color, weight, smoothness, and so on.
Attentional activation hierarchies versus distinctions
With regard to patterns of neural activation one always has to speak about patterns of something. Patterns ‘in general’ make no sense. But what the ‘something’ is, is not necessarily clear, so it is not always right to speak about patterns of something.
In speech any adjective can be seen as a dimension of something. Take blueness. Everyone has an idea of it, but some people have more complex, and more meaningful structures about it than others. In its simplest form blueness to someone can mean that something is more blue or less blue. Someone, a painter maybe, or you, may, by reading this paragraph, develop, in his brains, a neural structure for blueness which entails more than other people can think of. When one thinks of blueness, one gets all kinds of vague, almost conscious thoughts with regard to blueness – it is problematic to speak of things which are ‘almost’ conscious, but I hope you can skip this problem for now. This is the image that I have of a neural hierarchy for blueness as caused by the attention mechanism. Blueness is the most immanent. What I think of as vaguely conscious is less immanent. Further more there is a lot of neural activity which I don’t even know of, but which does influence my thoughts, and this I name more transcendent.
This said there seems to be a very interesting relation between opposites, like blue or not blue, light and dark, and so on. It almost seems as if opposites not only come into existence automatically, they also seem to be associated fairly automatically. This might not be as strange as this seems, since things always oppose each other with regard to something more encompassing. Blue, for instance, is the opposite of not-blue with regard to color, or more specific, blueness. Dark can be contrasted to light with regard to lightness or darkness. So again this is not only just a matter of immanization, with the neural network distinguishing something that opposites have in common, and grouping them accordingly, but also of a neural activation hierarchy as caused by the attention mechanism, whereby blueness at the top neural level leads to activation of blue and not blue, or blue and not blue things at lower levels. As such we have the same mechanism as with regard to parts and wholes again.
Above I recapitulated much of what I said earlier. Why is this so important to me? It is important because of the fact that in a neural machine with an attention mechanism, a mirroring mechanism, and a small build-in hierarchy, there will, in any neural activation structure, or sensitivity structure, always be at least one hierarchy, namely the build-in hierarchy. Therefore I tend to think of neural activation as always being hierarchical, and therefore I tend to see parts and wholes in everything, at least objectively. In this way I can subjectively understand an otherwise objective process.
If there is objectively a neural hierarchy with regard to, for instance, blueness, will it then not also be subjectively there? No, certainly not necessarily, and probably not even usually. Realizing it would anyway be a self-conscious action. So that will be beyond most earthly beings.
Mixing attentional hierarchies of things
Since the attention mechanism can attend to five things at a time, we may conceive of attentional hierarchies as consisting of five intermingled hierarchies. In an immanized neural network this has very interesting consequences, for instance with regard to the perceived meaning of things. It also allows for new, more ore less sensible discoveries to happen within our brains – how sensible depends on lower level immanization, and therefore, among others, on training. Further more it is a prerequisite for language.
To put it simply, the intermingling of attentional activation hierarchies is one of the mechanisms which allow for the possibility of a mixture of words – read: a (part of a) sentence – to gain a new meaning which goes beyond each of the words. The mixing takes place below the top neural levels, especially in the middle neural levels, and above. In so far as successful, this generates one ore more new points of attention – this I earlier named “chunking.” As such our attention keeps shifting. Remark that the new point of attention will probably not be expressible in language using a single word, even if we can attend to it singularly, except by referring to it as ‘that’ or ‘this’ or ‘whom’ or ‘which,’ and so on.
One example is that attending something blue, and attending something not blue, may lead to the more abstract notion of blueness – that is its exostructure.
Another example: you look at a bicycle, and next you notice the wheels, the spokes, the chain, and the rider – that is its endostructure.
The mixing of attentional neural hierarchies will, if sensible, and tested through actions and/or thought, and so on, incline to a better, more sensible immanization of the lower neural levels, which will in turn help to understand the world, and it may also lead to a new point of attention. Important is that this understanding is actively employed, if only through trial and error. The mixing-in of language and culture through the programming mechanism, further more, leads to a more culturally established view on the world.
This said, I do not believe that the structures in the middle level of the biologically fixed, neural hierarchy are, even in as far as they are immanent, in general, and certainly not in most people, very ‘rational’ – whatever that is. Its content is largely the domain of what Lacan called “l’imaginaire,” and for the rest we are generally no wiser than our parents where, or in other words, no wiser than our culture lets us be. As we are in our mothers’ womb, and the first years after birth, while our brains are forming themselves, then also the first patterns of activation form themselves. They do so in a largely ‘empty,’ in the sense of ‘non-determined,’ system – John Lock’s “tabula rasa,” but better to be conceived of as filled with noise (and some structure too) than as really being empty. Lacan speaks of a specific, too early adultness of humans, which causes us to form l’imaginaire. With this not always so very lucky start, we are stuck for the rest of our lives. Not that it cannot change, but this is the basis, and changes will require effort. If you accidentally make an unhappy start, then, for a happy end, conditions for you must, later in live, be even better than for someone with more luck, while the opposite, bad conditions because of a bad start, is more likely to happen. It is, anyway, quite certain that most men come to nothing without the aid of other men.
Attention and temporality
A neural network with an attention mechanism does more than simply associate or connect things with each other. Imagine to be able to look into your head, and see the attention patterns dancing in the ‘top’ part of your brain. Then you can also imagine how these attention patterns make you gain an existence in time. In the lower, sensory part there is even more change in pattern, but this doesn’t look like a dance.
The middle and upper middle parts of the biologically fixed, neural hierarchy make sure that the attention patterns are, from moment to moment connected together in a meaningful way. In these (upper) middle layers, the attention patterns, and also much of language and speech, condense, to use a Freudian word, into meaning patterns. Here we find much of the metaphoric aspects of speech, especially as far as these can’t be imagined in a sensory sense, like with ‘strength.’ The other way, the meaning patterns will lead to new attention patterns. As such the meaning patterns have, among others, a stabilizing function with regard to the drifting of our attention, and thereby of our thoughts and consciousness.
Together all the sensory, meaning, and attention patterns constitute our consciousness of events, whereby the meaning patterns form the more or less static bindings, for the more temporal sensory, and attention patterns.
An event is some‘thing’ which to us exists in time, and sometimes in space too. It is because of the essential importance of temporality in the working of my machine that I like to call it a time-machine.
The temporality at the top of the network is quite different from the temporality at the bottom. At the top temporality shows itself in the dancing of ever-changing attention patterns, each consisting of a small number of neurons, or maybe collections of neurons, at a time.
At the bottom of the network there are clear peaks of activation too, but here there are millions of them. The dominant characteristic here is not the movement or shifting of neural activation, but the seemingly chaotic change of activation.
Compared to the activation in the middle, the activation at both bottom and top is more fluctuating, by which I mean that the activation of each neuron differs a lot from moment to moment. The middle levels fluctuate too, maybe modulated by general stabilization and synchronization mechanisms, as mentioned in chapter 3, but they have a bigger chance to be reactivated by the attention mechanism, the senses, and some activation of their own – see the section on background knowledge on page 104. The relative continuity in the middle gives us a sense of who and where we are. Be ware if you lose your connection with this.
Non-hierarchical consciousness despite neural activation hierarchies
It is important to mention that the hierarchies in neural activation and neural sensitivity – whether or not caused by our in-born neural hierarchy – nor its content, imply that our consciousness is – subjectively – necessarily hierarchically structured too, although it is likely structured as such at least partly. Even if consciousness is structured in the same way indirectly, then we are usually not conscious of this – that is we are not conscious of our consciousness in this respect, in other words we are not self-conscious with regard to this.
We are hardly ever self-conscious of gathering structures too. Since we will consciously (but not self-consciously) often slip from one gathering structure to the next, and since there is always a complex mix of gathering structures at work at many different levels in the neural hierarchy, one cannot really say that our consciousness as a whole has a gathering structure. Even with regard to non-narrative knowledge one can attribute the post-modern idea of many sorts of discourse hooking on each other, with the discourses here being privately, culturally, and/or discursively constituted gathering structures.[^]
[^]Here I am referring to the post-modern philosophy of Jean-François Lyotard. See: Jean-François Lyotard. 1979. Het postmoderne weten: een verslag. Kampen: Kok Agora.
Background knowledge
The being immanent of the nevertheless ‘distributed’ activation may account for the possibility of background knowledge. Secondly it makes it possible for instincts, and even Lacanian, language-based ‘drives,’ to somewhere attach to the neural network ‘intelligibly’ with bidirectional interconnections.
Immanence and the attention mechanism make for meaningful background knowledge, and meaningful instincts and emotions. Master neurons and the attention mechanism create a meaningful difference between neurons. This process I called “immanization.” As such a little ‘background activation’ of these neurons may have a sensible influence on what we do and think.
To get a meaningful background activity from which I can, for example, know or take into account where I am and what I’m doing, it is sufficient for the neurons which ‘sample’ – or ‘stand for,’ or ‘replace’ – the images which go with these places and activities, to be a little more active or sensitive than others. This is something which can be achieved easily and automatically at neural level. This way mammals can at least take into account, if not remember, their environment, and other background knowledge, in their learned actions.
I do not say that this is the only background knowledge we have – this certainly is not true for human beings – neither do I say that there are not a number a refinements to be made. Now, in 2021, I think the essence of background knowledge is caused mostly by our immense memory. We can remember up to 40 things per second, and we have a massive memory, so we can live in unique moments continuously. But in storing all the moments this continuity may indeed be expressed in meaning patterns. The advantage of immanized networks is that knowledge can be reused.
Connecting instinct and intellect
In the same way that background sensitivity can be meaningful, instincts can extend themselves into the neocortex, and create a meaningful position for themselves by simply being associated with our knowledge. They may influence our thinking and behavior by creating a kind of background activation or background sensitivity from outside the cortical neural network, but from within the animal itself.
The other way, if we are conscious of certain things, then we may experience the accompanying emotions or instinctual reactions because of the neural projections to our instinctual and emotional systems. This is, I think, part of an explanation of what emotions, and qualia like pain and pleasure are with regard to consciousness. Firstly emotions cause a general arousal of both our body and our cortical system, from which all kinds of reactions will follow, which we experience as qualia. Secondly they often cause specific arousal in the sense of activating a specific piece of neural tissue as if being sensory information.
How much a certain instinct is at the background or at the foreground seems to differ from animal to animal. It seems likely that intelligence is hindered when instinctive signals are very much to the foreground in the higher neural networks. I say this because it always strikes me how guide-dogs seem to have so much less of the typical dog-like behavior – but I know nothing about dogs, let alone guide-dogs.
The girl who couldn’t act
I ones saw a movie based on a true story about a little girl with a cerebellum that was badly injured from birth. So badly could she control her movements that she could not make any contact at all with the world, at first. She was in a mental institution, treated like a plant with no mental abilities, but she was very smart, she understood what happened, and what was said around her, and she could talk in herself, that is, she could think self-consciously – and what a beautiful, proud and rightfully stubborn girl she was! Later she could communicate, but only with the help of a woman who had discovered it. Since it was an American movie they had to go to court to get it recognized. The little girl also pointed out another boy who was “smart too.” A beautiful movie – a real tearjerker – based on true events, whatever that means. Well, to get back to my topic, this girl could not act, at first, but she could see – though she could not really control her eye-movements – and hear!