Figure 3.1. The metasystem transition This is chapter 3 of the "The
Phenomenon of Science" by Valentin F.
Turchin
Contents:
SUBSEQUENT STAGES in the development of the nervous system will be described
as stated above, on a more phenomenological level. For this we must summarize
the results of our investigation of the mechanism of evolution in the early
stages, using the terminology of general cybernetic concepts. Having begun
to think in this direction, we shall easily detect one general characteristic
of transitions from lower to higher stages: In each stage the biological
system has a subsystem which may be called the highest controlling device;
this is the subsystem which originated most recently and has the highest
level of organisation. The transition to the next stage occurs by multiplication
of such systems (multiple replication) and integration of them--by joining
them into a single whole with the formation (by the trial and error method)
of a control system headed by a new subsystem, which now becomes the highest
controlling device in the new stage of evolution. We shall call the system
made up of control subsystem X and the many homogeneous subsystems A1,
A2, A3 . . . controlled by it a metasystem
in relation to systems A1, A2, A3
. . . Therefore we shall call the transition from one stage to the next
the metasystem transition.
Figure 3.1. The metasystem transition
This concept will play a crucial part in our subsequent presentation. The metasystem transition creates a higher level of organization, the metalevel in relation to the level of organization of the subsystems being integrated.
From the functional point of view the metasystem transition is the case where the activity a, which is characteristic of the top control system at a lower stage, becomes controlled at the higher stage and there appears a qualitatively new, higher, type of activity b which controls the activity a. Replication and selection bring about the creation of the necessary structures.
The first metasystem transition we discern in the history of animals is the appearance of movement. The integrated subsystems are the parts of the cell that ensure metabolism and reproduction. The position of these parts in space is random and uncontrolled until, at a certain time, there appear organs that connect separate parts of the cell and put them into motion: cell membranes, cilia, flagella. A metasystem transition occurs which may be defined by the formula: control of position = movement.
In this stage movement is uncontrolled and not correlated in any way with the state of the environment. Nature's next task is to control it. To control motion means to make it a definite function of the state of the environment. This leads to irritability. Irritability occurs when--under the influence of external factors--there is a change in the state of some segments of the cell, and when this change spreads to other sectors--specifically those which ensure movement. Thus, the formula for the metasystem transition from the second stage to the third is: control of movement = irritability.
| Chemical Era | 1. Chemical foundations of life |
| 2 Movement | |
| 3 Irritability (simple reflex) | |
| Cybernetic Era | |
| 4 Nerve net (complex reflex) | |
| 5 Associating (conditioned reflex) |
Figure 3.2. Stages in the evolution of life before the era of reason.
The integration of cells with formation of the multicellular organism is also a transition from a system to a metasystem. But this transition concerns the structural aspect exclusively and is not describable in functional terms. From a functional point of view it is ultimately unimportant whether reproduction and integration of a certain part of the organism occur or whether organisms are integrated as whole units. This is a technical question, so to speak. Irritability is already manifested in unicellular organisms, but it reveals its capabilities fully after cell integration.
An important characteristic of the metasystem transition must be pointed out here. When the subsystems being integrated are joined into a metasystem, specialization occurs; the subsystems become adapted to a particular activity and lose their capability for other types of activity. Specialization is seen particularly clearly where whole organisms are integrated. Each subsystem being integrated in this case contains a great deal which is ''superfluous''--functions necessary for independent life but useless in the community, where other subsystems perform these functions. Thus, specialized muscle and nerve cells appear in the multicellular organism.
In general we must note that the integration of subsystems is by no means the end of their evolutionary development. We must not imagine that systems A1, A2, A3, . . . are reproduced in large numbers after which the control device X suddenly arises ''above them." On the contrary, the rudiments of the control system form when the number of subsystems Ai is still quite small. As we saw above, this is the only way the trial and error method can operate. But after control subsystem X has formed, there is a massive replication of subsystems Ai and during this process both Ai and X are refined. The appearance of the structure for control of subsystems Ai does not conclude rapid growth in the number of subsystems Ai; rather, it precedes and causes this growth because it makes multiplication of Ai useful to the organism. The carrier of a definite level of organization branches out only after the new, higher level begins to form. This characteristic can be called the law of branching growth of the penultimate level. In the phenomenological functional description, therefore, the metasystem transition does not appear immediately after the establishment of a new level; it appears somewhat later, after the penultimate level has branched out. The metasystem transition always involves two levels of organization.
Let us continue our survey of the stages of evolution. We shall apply the principle of the metasystem transition to the level of irritability. At this level, stimulation of certain sectors of a unicellular organism or a specialized nerve cell in a multicellular organism occurs directly from the external environment, and this stimulation causes direct (one-to-one) stimulation of muscular activity. What can control of irritability signify? Apparently, creation of a nerve net whose elements, specifically the effectors, are not stimulated by the environment directly but rather through the mediation of a complex control system. This is the stage of evolution we related to the concept of the complex reflex. The control of irritability in this stage is seen especially clearly in the fact that where there is a goal, stimulation of the effectors depends not only on the state of the environment but also upon this goal--that is, on the state of certain internal neurons of the net. Thus, the formula for this metasystem transition (from the third stage to the fourth) is: control of irritability = complex reflex.
What next'?
NO MATTER how highly refined the nerve net built on the principle of the complex reflex may be, it has one fundamental shortcoming: the invariability of its functioning over time. The animal with such a nervous system cannot extract anything from its experience; its reactions will always be the same and its actions will always be executed according to the same plan. If the animal is to be able to learn, its nervous system must contain some variable components which ensure change in the relations among situations and actions. These components will therefore carry out control of reflexes. It is commonly known that animals have the ability to learn and develop new reflexes. In the terminology introduced by I. P. Pavlov, the inborn reflex included in the nervous system by nature is called an unconditioned reflex while a reflex developed under the influence of the environment is called a conditioned reflex. When we speak of a complex reflex we have in mind, of course, an unconditioned complex reflex. The presence of components that control complex reflexes manifests itself, in experiments with teaching animals, as the ability to form conditioned reflexes.
We cannot, however, equate the concept of the conditioned reflex with the concept of control of a reflex. The latter concept is broader. After all, our concept of the complex reflex, taken in the context of the description of general principles of the evolution of the nervous system, essentially signifies any fixed connection between the states of classifiers, representation fixers, and effectors. Therefore, control of reflexes must be understood as the creation, growing out of individual experience, of any variable connections among these objects. Such connections are called associations of representations or simply associations. The term ''representation" here is understood in the broad sense as the state of any subsystems of the brain, in particular the classifiers and effectors. We shall call the formation of associations associating (this terminology is somewhat awkward, but it is precise). Thus, the fifth stage of evolution is the stage of associations. The formula for the metasystem transition to this stage is: control of reflexes = associating.
THE CONCEPTS of the reflex and the association are functional not structural concepts. The connection between stimulus S and response R in the reflex (see figure 3.3) does not represent the transmission of information from one subsystem to another, it is a transition from one generalized state to another. This distinction is essential to avoid confusing the reflex, as a definite functional diagram which describes behavior, with the embodiment of this diagram, that is, with the cybernetic device that reveals this diagram of behavior.
Figure 3.3. Functional diagram of the unconditioned reflex
Confusion can easily arise, because the simplest embodiment of reflex behavior has a structural diagram that coincides externally with the diagram shown in figure 3.3, except that S and R in it must be understood as physical subsystems that fix the stimulus and response. This coincidence is not entirely accidental. As we have already said in defining the functional diagram, breaking the set of all states of the system down into subsets which are ascribed to vertices of the graph is closely tied to breaking the system down into subsystems. Specifically, each subsystem that can be in two states (yes/no) can be related to the set of all states of the system as a whole for which this system is in a definite state, for example ''yes.'' More simply, when defining the generalised state we consider only the state of the given subsystem, paying no attention to what is happening with the other subsystems. Let us assume that the letters S and R signify precisely these subsystems, that is to say, subsystem S is the discriminator for stimulus (set of situations) S and subsystem R is the effector that evokes response R. Then the statement that ''yes'' in subsystem S is transmitted along a communications channel (arrow) to subsystem R, putting it also in the ''yes'' state, coincides with the statement that the ,generalised state S switches (arrow) to state R. Thus the structural and functional diagrams are very similar. It is true that the structural diagram in no way reflects the fact that ''yes'' evokes a ''yes" not a "no,'' whereas this is the very essence of the reflex. As we have already said, the reflex is a functional concept.
THESE PRELIMINARY considerations were required in order for us to be able to better grasp the concept of association and the connection between a functional description using associations and a structural description by means of classifiers. Because each classifier can be connected to one or several generalized states, there is a hierarchy of generalized states corresponding to the hierarchy of classifiers. When introducing the concept of the classifier we pointed out that for each state of the classifier (we can now say for each generalized state of the system as a whole) there is a corresponding, definite concept at the input of the system--that is, the input situation is affiliated with a definite set. The concepts of the Aristotelian ''concept'' and the ''generalized state'' are close to one another; both are sets of states. But the "generalised state'' is a more general concept and may take account of the state not just of receptors but also of any other subsystems, in particular classifiers. This is essential to follow the dynamics of the state of the system during the process of information processing.
Let us see how the generalized states of the K level of the hierarchy and the next level, K + 1, are interconnected. As we know, the chief task of the classifiers is to store ''significant'' information and discard ''insignificant'' information. This means that there is some set of states on level K which in the functional diagram has an arrow going from each of the states to the same state at level K + 1. In figure 3.4 below, the representations (generalized states) T1 andT2 evoke representation U equally.
Figure 3.4. Association of representations
If T1and T2 always accompany one another this diagram will unquestionably be advantageous to the animal. He does not have to know that T1 and T2 are occurring; it is enough if he knows that U is occurring. In this way superfluous information is discarded and useful information is compressed. The compression of information is possible because T1 andT2 are always encountered together. This is a fact which is external to the nervous system and refers only to the stream of situations being fed to it. It testifies to the existence of a definite organization in the stream of situations, which is a consequence of the organized nature of the environment surrounding the animal. The organization of the nervous system and its activity (the system of reflexes) reflect characteristics of the environment. This happens because, by testing different ways to discard information, nature finally finds the variation where the information discarded is indeed superfluous and unnecessary owing to the partially organized nature of the environment.
In the stage of the unconditioned reflex the structure of such connections, as shown in figure 3.4, does not change during the life of the animal and is the same for all animals of the given species. As we have already said, however, such a situation is not satisfactory. The metasystem transition occurs, and the connections between generalized states become controlled. Now if T1 andT2 in the individual experience of the animal always (or at least quite often) accompany one another, new connections form in the animal brain which are not determined uniquely by heredity. This is associating--the formation of a new association of representations. It is clear that associations form among representations of the highest level of the hierarchy. Thus, the most general correlations in the environment, those which are the same for all times and all places of habitation, are reflected in the permanent organization of the lowerlevels of classifiers. The more particular correlations are reflected by variable connections at the highest level.
THE DIAGRAM shown in figure 3.4 may cause misunderstanding. When speaking of an association of representations we usually mean something like a two-way connection between T1 andT2, where T1 evokes T2 and T2evokes T1. But in our diagram both representations evoke something different, specifically U, and there are no feedback arrows from U to T1 and T2.
In fact, the diagram shown in figure 3.4 more closely corresponds to the concept of the association of representations than a diagram with feedback does. Specifically, it contains an evocation, in a certain sense, of representation T2 by representation T1 (and vice versa), but this is evocation by complement. The representation U contains both T2 and T1; after all, it was conceived by our nervous system as equivalent to the simultaneous presence of T2 and T1. Therefore, when T1 evokes U in the absence of T2, then T2 is contained concealed in U itself. By evoking U we, so to speak, complement T1 with the nonexistent T2.
This process of mental complementing is in no way related to the fact that the association is developed by learning. Only the method by which the brain processes information plays a part here. When inborn lower-level mechanisms operate the effect of the complementing shows itself even more clearly; no kind of learning or training will weaken or strengthen it.
Look at figure 3.5. In it you see not just points but also a line, an arc. In fact there is no line at all. But you mentally supplement (complement) the drawing with points so that a solid line is formed. In terms of figure 3.4. T1 here is the actually existing points, U is the line, and T2 is the complementary points. The fact that you discern a nonexistent line testifies to the presence in the brain (or in the retina) of classifiers which create the representation of U.
Why did these classifiers arise? Because the situations arriving at the input of our visual apparatus possess the characteristic of continuity. The illuminations of neighboring receptors of the retina are strongly correlated. The image on the retina is not a mosaic set of points, it is a set of light spots. Therefore, translating the image into the language of spots, the brain (we say ''brain'' arbitrarily, not going into the question of where the translation is in fact made) rejects useless information and stores useful information. Because ''consisting of spots'' is a universal characteristic of images on the retina, the language of spots must be located at one of the lowest levels and it must be inborn. The line which we "see" in figure 3.5 is a long, narrow spot.
NOTICE, we have reduced the concept of the line to the concept of the spot. We had to do this because we were establishing the theoretical basis for the existence of the corresponding classifiers. In reality, it is possible to conclude from the two-dimensional continuity of the image on the retina that the basic concept for the brain should be the concept of the spot. not the line. The line can be included as either an exotically shaped spot or as the boundary between spots. This theoretical consideration is confirmed by numerous observations.
Figure 3.6. Concealed circle formed by the vertices of angles
A circle formed by the vertices of angles is clearly seen in figure 3.6 a. In figure 3.6 b the vertices of the angles are located at exactly the same points, but their sides are directed every which way, some outside and some inside. As a result the circle disappears. It is possible to follow the vertices along, switching attention from one to another, and ascertain that they are set in a circle, but you cannot see this as you can in the first drawing, even though the points which make up the circle are all vertices of angles and all lie on the circumference of the circle. The simplest machine program for recognizing circles would "see'' the circle in figure 3.6 b (as well as figure 3.6 a). But our eye does not see it. In figure 3.6 a, where all the rays are directed out, however, our eye glosses them into something like a rim and clearly sees the internal circle, a two-dimensional formation, a spot. The circumference, the boundary of this spot, also becomes visible.
There are many visual illusions resulting from the fact that we ''see in spots.'' They offer instructive examples of inborn associations. One of the best ones is shown in figure 3.7.
Figure 3.7a is a square, and its diagonals intersect at right angles. Figure 3.7 b is constructed of arcs, but its vertices form precisely the same square as in figure 3.7 a. and therefore the diagonals also intersect at right angles. This is almost impossible to believe, so great is the illusion that the diagonals of figure 3.7 b are approximated to the vertical. This illusion may be explained by the fact that alongside the microcharacteristics of the figure--that is, the details of its shape--we always perceive its macrocharacteristics, its overall appearance. The overall appearance of figure 3.7 b is that of a spot which is elongated on the vertical. The degree of elongation may be judged by figure 3.7 c. This figure is a rectangle whose area is equal to the area of figures 3.7 a and b, while the ratio of its width to its height is equal to the ratio of the average width of figure 3.7 b to its average height. The hypothetical classifier which records the overall elongation of the figure will arrive at the same state in contemplating figure 3.7 b as in contemplating figure 3.7 c. In other words, whether we desire it or not, figure 3.7 b is associated in our mind with the rectangle in figure 3.7 c. Following the diagonals in figure 3.7 b in our mind, we equate them with the diagonals of figure 3.7 c, which form acute vertical angles. The classifier that records elongation of the spot is unquestionably a useful thing it was especially useful for our distant ancestors who did not perceive the world in more subtle concepts. But because we cannot switch it on or off at will, it sometimes does us a disservice, causing visual deception.
BUT LET US RETURN from inborn associations to developed ones, that is, to the actual associating of representations. The very essence of the metasystem transition from the fourth stage of evolution to the fifth lies in the difference between the suffixes of two words from the same root. The association is simply one of the aspects of the complex reflex, while associating is control of associations: the formation of new associations and disappearance of old ones.
The capability for associating representations appears most fully as the capability for forming (and therefore also recognizing) new concepts. The dog that recognizes its master from a distance may serve as an example.
The Pavlovian conditioned reflex is a more particular manifestation of the capability for associating. The diagram of this reflex is shown in figure 3.8.
Figure 3.8. Diagram of the conditioned reflex.
The unconditioned stimulus S1 (food) is always accompanied by the conditioned stimulus S2 (a whistle), and as a result they become associated in one representation U, which, because of the presence of S1 in it, causes the response R (salivation). Then stimulus S2 causes U, and therefore R, even where S1 is not present. The whistle causes salivation.
A question may arise here. The conditioned reflex arises on the basis of the unconditioned reflex whose diagram is S -> R. At the same time, if the conditioned stimulus is removed in figure 3.8, we shall obtain the diagram S1-> U -> R. How do we know that step U exists? Is this an arbitrary hypothesis?
In reality the diagram shown in figure 3.8 contains absolutely no hypotheses. We shall emphasize once more that this diagram is functional, not structural. We are making no assumptions about the organization of the nerve net; we are simply describing observed facts, which are these: first, state S1 leads, through the mediation of some intermediate states, to state R; second, state S2 in the end also leads to R. Therefore, at some moment these two processes are combined. We designate the state at this moment U and obtain the diagram we are discussing. In this way our diagram, and our approach in general, differ from the Pavlovian diagram of the reflex arc, which is precisely the structural diagram, a physiological model of higher nervous activity.
The process of learning, if it is not reduced to the development of certain conditioned reflexes (that is, touching only the discriminatory hierarchy) also includes the element of acquiring know-how, development of specific skills. The process of learning also fits within the diagram of associating representations in the general meaning we give to this concept. After all, learning involves the development and reinforcement of a detailed plan to achieve a goal. a new plan that did not exist before. The plan may be represented as an organized group of associations. Let us recall the regulation diagram (see figure 2.6). With a fixed goal the comparison block must juxtapose a definite action to each situation. The ''untaught'' comparison block will test all possible actions and stop at those which yield a reduction in the discrepancy between the situation and the goal (the trial and error method). As a result of learning a connection is established between the situation and the appropriate action (which is, after all, a representation also) so that the ''taught'' comparison block executes the necessary action quickly and without error.
Now for a few words about instinct and the relationship between instinctive behavior and behavior developed through learning. Obviously, instinct is something passed on by inheritance--but exactly what? In the book already referred to, Miller, Galantier, and Pribram define instinct as a ''hereditary, invariable, involuntary plan.'' Plans, as we know, are organized on the hierarchical principle. It is theoretically possible to assume the existence of an instinct that applies to all stages of the hierarchy, including both the general strategy and particular tactical procedures all the way to contracting individual muscles. ''But if such an instinct does exist,'' these authors write, ''we have never heard of it.'' The instinct always keeps a definite level in the hierarchy of behavior, permitting the animal to build the missing components at lower levels through learning. A wolf cub which is trying to capture a fleeing animal unquestionably acts under the influence of instinct. But it is one thing to try and another to succeed. ''It may be considered,'' Miller, Galantier, and Pribram write, "that copulation is an instinctive form of behavior in rats. In certain respects this is in fact true. But the crudeness of copulative behavior by a rat which does not have experience in the area of courting shows plainly that some practice in these instinctive responses is essential.''
As the organization of an animal becomes more complex and its ability to learn grows in the process of evolution, the instincts ''retreat upward,'' becoming increasingly abstract and leaving the animal more and more space for their realization. Thus the behavior of animals becomes increasingly flexible and changes operationally with changes in external conditions. The species' chances for survival grow .
IN OUR DISCUSSION of associations of representations thus far we have completely ignored their dynamic, temporal aspect; we have considered the representations being connected as static and without any coordinate in time. But the idea of time can enter actively into our representations. We can picture figures that are moving and changing at a certain speed and we can continue the observed process mentally. A wheel rolls down the road. We close our eyes for a second or two and picture the movement of the wheel. Upon opening our eyes we see it in exactly the place where we expected it. This is, of course, the result of an association of representations, but this means an association, or more correctly representations, which are organically bound up with the passage of time. The wheel's position x at moment t is associated with the position x1 at moment t + [Delta]t with position x2 at moment t + 2[Delta]t , and so on. Each of these representations includes a representation of the time to which it refers. We do not know the mechanism by which this inclusion is made and, in conformity with our approach, we shall not construct any hypotheses regarding this. We shall simply note that there is nothing particularly surprising in this. It is commonly known that an organism has its own time sensor, the ''internal clock.''
The association of representations that have a time coordinate enables us to foresee future situations in our imagination. We have just established the existence of such representations relying on internal, subjective experience. But the fact that animals also reveal the capability for foresight (look at the way a dog catches a piece of sugar) leads us to conclude that animal representations may also have a time coordinate.
Speaking in the language of cybernetics, the interconnection of representations which have a time coordinate and the resultant capability to foresee the future is simply modeling, constructing a model of the environment.
Figure 3.9. Diagram of modeling.
Let us give the general concept of the model. We shall consider two systems a and b. Let us assume that to each state Ai of system a we can somehow juxtapose one definite state Bi of system b. The inverse correspondence does not have to be unique (single-valued); that is, many states of a may correspond to one state of b. Because, according to our definition, the genenalized state is a set of states, this proposition may be described as a one-to-one correspondence of the states of system b to the generalized states of system a . This is necessary but not sufficient to consider system b a model of system a . Additionally there must be a transformation T(t) of system b which depends on time t and models the natural passage of time in system a . This means the following: Suppose that system a is initially in generalized state A1 which corresponds to state B1 of system b. Suppose that after the passage of time t the state of system a becomes A2 Then the conversion T(t) should switch system b to state B2, which corresponds to generalized state A2. If this condition is met we call system b a model of system a .
The conversion T(t) may involve nothing more, specifically, than permitting system b by itself to change its state with time. Such models are called real time models.
The besiegers dug an opening under the fortress wall and placed several barrels of powder in it. Next to them a candle was burning and from the base of the candle a trail of powder ran to the barrels. When the candle burned down the explosion would take place. An identical candle lighted at the same time was burning on a table in the tent of the leader of the besieging forces. This candle was his model of the first candle. Knowing how much time remained until the explosion he gave his last orders.... Wild faces leaned over the table, hairy hands clutched their weapons. The candle burned down and a fearsome explosion shook the air. The model had not let them down.
The image on a television screen when a soccer game is being broadcast may also be formally considered a model of the soccer field and stands. All conditions are in fact observed. But one senses a great difference between the case of the two candles and the case of the soccer broadcast--a difference in the information links between systems a and b. Any image b of object a is a model of it in the broad sense; but there is a continuous flow of information from a to b and it is only thanks to this flow that the correspondence between states a and b is kept. With information access to b, we in fact have access to a . System b operates as simply a phase in the transmission of information from a .
The situation is quite different in the case of the two candles. Candle b burns at the same speed as candle a , but independently of it. The leader of the besieging forces does not have access to candle a and cannot receive any real information regarding its state. By modeling he compensates for this lack and obtains equivalent information. System b here plays a fundamentally different and more significant role. A spatial barrier is overcome, so to speak, by this means and this is done without establishing any new information channels.
Even more important is the case where the model helps overcome a barrier of time rather than space. One cannot, alas, lay an information channel to the future. But a model permits us to operate as if there were such a channel. All that is required is that execution of the conversion T(t) on the model take less time than time t itself. Many other examples could be given of the use of such models in modern life, but that is hardly necessary. Let us return once again to associations of representations.
We have seen that associations of static representations reflect the existence of spatial correlations, interrelationships in the environment. In exactly the same way associations of dynamic representations (models created by the brain) reflect dynamic temporal correlations that characterize the environment. Situation x after time t evokes (or may evoke) situation y--that is the general formula for such correlations, and in the brain these correlations are imprinted in the form of the corresponding associations.
WHAT IS knowledge? From a cybernetic point of view, how can we describe the situation where a person or animal knows something or other? Suppose we know there are two people in an adjacent room. Since they really are there, if we go into the room we shall see two people there. Because we do know this, we can, without actually entering the room, imagine that we are opening the door and entering it; we shall picture the two people who are in the room. In our brains therefore, an association of representations takes place which enables us to foresee the results of certain actions: that is. there is a certain model of reality. For the same reason, when we see a rolling wheel, we know where it will be a second later, and for the same reason when a stick is shaken at a dog the animal knows that a blow will follow, and so on.
Knowledge is the presence in the brain of a certain model of reality. An increase in knowledge--the emergence of new models of reality in the brain--is the process of cognition. Learning about the world is not a human privilege, but one characteristic of all higher animals. The fifth stage of evolution may be called the stage of individual cognition of the world.
