Ascent to the intellect. Experience of evolutionary hierarchical classification
The article discusses a variant of the evolutionary-hierarchical classification of autonomous automatic devices, which, first of all, include living organisms - from the simplest to humans. However, in order to abstract from the set of functions of a biological organism and focus only on the general principles of its interaction with the environment, and not to limit the classification to only living systems, the article uses the term “device” instead of “organism”.
The reasoning is based on a largely simplified pattern of the behavior of a living organism: it is represented as a device, the interaction of which with the environment is given by the set of algorithms embedded in it. An algorithm is an existing (saved, fixed) pattern of device behavior that can be unambiguously and repeatedly implemented by them in the form of specific actions. The reasons for the “launch” (application) of a particular algorithm — whether external influences or internal goal-setting — are not considered. But even with this simplified approach, it was possible to construct a sufficiently heuristically productive classification.
For the starting point of ascent along the hierarchical ladder, an elementary device is adopted, which unambiguously, not alternatively interacting with the environment, is an elementary device.
The algorithm of functioning (behavior) of elementary devices is laid down, fixed at their design (birth) and does not change during the operation. Adaptation of elementary devices to the environment is carried out only by a one-time modification of the full set of algorithms during their creation (birth) and selection of the most “adequate” devices with consolidation (inheritance) of successful solutions. Consequently, to adapt to changes in the environment of elementary devices, their constant reproduction with a variation of the set of algorithms is necessary.
It is reasonable to assume that the once effective (that is, previously passed selection), but replaced with new ones over time, the algorithms are not “erased”, but are preserved. Such an accumulation of “ready-made” algorithms, further allows the population to be shorter and more efficiently adapt to changing conditions - not “waiting” for random successful modifications, but using already workable already existing algorithms and their combinations.
An important consequence of the accumulation of algorithms and the first step in overcoming the uniqueness of the functioning of elementary devices is their consistent application of several different algorithms in the same type of conditions. Even a random device alternating two or more algorithms to solve a problem significantly increases the probability of achieving a positive result. However, it is clear that the elementary devices do not have the mechanism for selecting the most efficient algorithm in the course of operation (life) - the fixing of successful combinations of algorithms is realized only by selecting their full complex.
If for elementary devices, the efficiency of the application of algorithms is estimated only as a result of the survival of some and the extinction of others and adaptation is achieved only through many generations, then for more complex devices that have accumulated a redundant set of algorithms, there is a real possibility (and necessity) to evaluate the results of the algorithms during the operation itself . The mechanism of this assessment was realized in the form of a complex of additional algorithms, which can be called emotional. Emotion algorithms, unlike action algorithms, are not directly related to providing specific device functions. Their main task is to initiate some changes in the device, which would allow to evaluate the success of the action algorithms directly during their execution (or immediately after),
The accumulation of an excessive number of algorithms of actions, as well as the emergence of emotion algorithms evaluating their performance, created the basis for the formation of devices of a new type, which can be called adaptable. The main difference of the adaptable device from the elementary is its ability to fix the application of the most efficient algorithms when the environment changes. Consequently, the adaptation of devices is realized not through generations, but directly in the process of their functioning (life).
Adaptation of an adaptive device also occurs randomly, by trial and error, but the choice is made not from the variants of the complete sets of algorithms, but between several algorithms.
If the transition from an elementary device to an adaptive was caused by the internal differentiation of the device associated with the accumulation of an excess number of algorithms, then the next hierarchical leap is caused by the external non-identity of the adapting device to itself. The change of the adapting device during its functioning (life) makes it necessary, when evaluating the results of applying the algorithms, to include oneself as one of the elements of the environment, which gives impetus to the formation of a device of the following type - a reflective device.
The principal difference between a reflective device and an adaptive is its ability to prioritize (before performing actions) the choice of one of the existing behavior algorithms. This was made possible through regular repetitions by adapting devices to the process of choosing the best possible course of action. As a result, the mechanism of this choice was fixed in the form of an additional algorithm — the reflection algorithm, as if superimposed on the behavior algorithms. Unlike emotion algorithms, which evaluate the result of actions, the reflection mechanism is activated before executing external action algorithms.
Reflective assessment of the most appropriate behavior algorithm before performing the action itself allows, in the current situation, rather than by trial and error, to respond to changes in the environment. But the presence of the reflection algorithm does not eliminate the probabilistic moment in the functioning of the device - just a brute force of algorithms from the external sphere passed into the "internal".
As already noted, the formation of reflection, is a direct consequence of the differences in the adapting device in time, its non-identity to itself at different points in its functioning. It can be said that the reflection algorithm restores the time integrity of the device — with frequent changes in behavior algorithms, it is reflection that fixes its identity.
When discussing the problem of adaptation, it is customary to introduce the concept of “environment model”. In the general case, the model is considered as a certain substitute (internal substitute) for the surrounding world, allowing the device to respond adequately to external influences. However, with this approach, it can be stated that the environment model of devices of pre-reflective levels is absolutely identical to the set of their algorithms. That is, the external world for elementary and adaptable devices consists only of what they can react to and what they can affect, that is, of the elements of the algorithms of their operation (for example, the model of the flush tank world consists only of the water level and the event of pressing the knob plum). Only after the formation of the reflection algorithm - the discernment of the device from itself, highlighting oneself as an element of the environment — the environment model is separated from the complex of action algorithms. But, in essence, this new model, which includes the device itself as an element of the environment, is nothing more than a simple set of algorithms that make up the reflection mechanism.
It is this, not external, but internal and not separated in time, but the one-time distinction of the device itself from itself, the distinction in itself of two types of algorithms - two environmental models - creates new possibilities for adapting devices to changes in the external environment.
On the one hand, (1) the presence of a reflection mechanism not only allows an a priori assessment of the effectiveness of algorithms from the existing set, but also creates a real possibility of generating fundamentally new algorithms that are not laid down during its creation (birth). New algorithms are built as combinations of existing ones and are fixed when they are found useful.
On the other hand, (2) separating the model of the external world, formed by reflection, from the active-reactive model (model of actions and reactions) leads to the possibility and necessity of evaluating the reflexive device itself not only as (a) element of the external environment, but also as (b) an element of the model thereof. This internal separation of the device itself into (a) real and (b) ideal also leads to distinguishing itself from “relatives” (devices of the same type), which certainly opens another channel to replenish the set of algorithms directly during operation (life) devices - imitation, adoption of algorithms.
So, improving reflexive devices with regularity leads to the formation of new mechanisms for generating and transmitting algorithms, which with strict necessity requires the emergence of a new mechanism for their preservation, and, therefore, the formation of a fundamentally new level of organization of devices and their complexes.
For fixing a continuously growing stream of new algorithms generated during the operation of reflexive devices, the former, hereditary method of fixing algorithms (both action algorithms and reflection algorithms, that is, an a priori assessment of action algorithms) has become fundamentally inapplicable. The next logical step in the development of methods of operating with algorithms was the formation of mechanisms (1) for preserving algorithms outside of performing devices and (2) external “embedding” of these algorithms in new devices. Devices with such abilities will be called reasonable, and the form of accumulation (preservation) and distribution (transmission) of algorithms - culture, the main element of which initially becomes the language.
There are several fundamental differences between reasonable devices and reflexive ones:
(1) The set of algorithms that a reasonable device can operate with is not specified when it is created (born). For the formation of a full-fledged intelligent device, a process of “loading” algorithms, called learning, is necessary.
(2) A reasonable device can not only a priori choose the most efficient algorithm available, but also complete the set of algorithms from the set represented in its outer culture.
(3) A reasonable device is capable of fixing the algorithms created by it outside itself - in the elements of culture.
Considering the listed features, first of all it should be concluded that a reasonable device is a fundamentally social (collective) device. For its formation and functioning, it is necessary to have some kind of environment of rational devices-carriers of culture - society. Consequently, unlike the characteristics of lower-level devices (such as adaptability, reflection), rationality is a purely social, systemic concept. If earlier the device type was set “from birth” and the multiplicity of devices was necessary primarily to ensure the variability of the algorithms and the selection of the most effective of them, then for a reasonable device inclusion in the society is a necessary condition that determines it as such. The rationality of the device is not its immanent initial characteristic; it becomes reasonable only in society.
In fact, rational behavior itself does not differ from the adaptive behavior of lower-level devices, since at a particular moment of action it does not matter how the algorithm is formed (genetic), genetic, reflexive, or social. The essential difference of a reasonable device is that it can change and supplement the set of algorithms during operation and also fix new algorithms outside of it.
The speed of adaptation to the environment of intelligent devices is significantly higher than that of previous levels. This speed is provided mainly due to the formation of horizontal (timeless) connections between devices. That is, if a change in the principles of interaction with the environment of prudent devices requires many generations, since only a hereditary (vertical, spaced in time) method of transmitting and storing algorithms is available to them, then at a reasonable level, new efficient algorithms become available for each device almost instantly, at least for a single generation.
The initial unprogrammed reasonable device, on the one hand, and the avalanche-like growth of new algorithms fixed in the culture, on the other, naturally led to the specialization of devices. That is, during the training period, intelligent devices can receive different sets of algorithms and, therefore, significantly differ from each other in operation. Differentiation of devices according to the set of functioning algorithms is also observed at the previous hierarchical levels (sexual and other types of intraspecific separation in animals), but it was strictly fixed at the moment of creation and could no longer change during the operation. The need for external differentiation of intelligent devices is a consequence of how (1) the limited capacity of single devices to perceive the full range of algorithms accumulated in culture,
While it was a question of pre-reasonable levels, it was possible to dispense with the use of the concept “program”. A complete set of algorithms for a single device was the only constant program of its functioning. The formation of intra-societal differentiation of devices, their specialization requires the distinction between separate sets of algorithms that can be fixed as various programs. In addition to the fact that various reasonable devices, as already noted, can have various programs, they can also actively operate with several practically independent programs (professional, etc.) and replenish their composition in the course of operation.
It should be noted that the introduction of the concept of “program” is advisable only when a statement is made that programs are independent of the devices themselves, of their belonging to a culture, and not to an individual. Although all programs as concrete sets of algorithms are functionally realized only by specific single devices, but (1) in their origin, (2) in distinguishing from other algorithms, (3) in the method of fixation (storage) they are of a purely social, general cultural character. Therefore, it would be more correct to call them social programs.
Of all the social programs, the most important, requiring priority “loading” into the devices when they are taught, is the universal social program - the language. Unlike application programs, that is, programs that implement the direct functioning of reasonable devices, the language determines, sets the functioning of society as a whole. It directly provides both the preservation of new algorithms and their “loading” into devices. Practically, the language in the society of intelligent devices performs the function of reflection (algorithm control algorithms), fixes the self-identity of society.
Just as the presence of the over-algorithm and the internal distinction of conscious devices led to the transition to a new level of organization, the development of rational society towards differentiating culture into many social programs and forming a single over-program (language) with regularity leads to the possibility and necessity of a new evolutionary leap - the emergence of intelligent devices.
The essence and the result of a reasonable developmental stage were: (1) the separation of algorithms from the devices themselves, (2) the integration of algorithms into complexes - programs, (3) the differentiation of programs with the simultaneous formation of a unified over-program - language. All this, in fact, is a statement of the fact of the emergence of a new reality, a new environment, a second space of existence of single intelligent devices - culture. The logical consequence of the formation of a new reality was the emergence of specialized programs that were no longer focused on adapting devices to the environment, but on operating with elements of cultural space. Such programs and reasonable devices capable of functioning according to these programs can be called intellectual.
The main result of the functioning of an intelligent device is not its adaptation to the environment (as for devices of previous levels), but new social programs. If it is possible to talk about the adaptive role of intellectual programs, then only referring it to the whole society, and not to a specific intellectual device.
Although a reasonable device can fix, save new algorithms generated by it, making them available to other devices, this ability to produce new algorithms is not its necessary characteristic. The production of new social programs is the only and defining function of smart devices.
Attention should be paid to the fact that intelligent devices are only operational, that is, at the time of the "work" of the intellectual program. In other periods of its operation, the intelligence of devices does not manifest itself in an explicit form - they interact with the environment, based on the existing set of rational and other adaptation programs. Consequently, intellectuality is more a characteristic of a specialized social program, and not of the device itself that performs it. Therefore, it would be more correct to speak generally not about intelligent devices, but about the intellectual activity of intelligent devices.
Moreover, since the functioning of the intellectual program is not directly related to the adaptation (effective adaptation to the environment) of a single device, the intellectual program has a social (system-wide) status not only in its origin (like other social programs), but also in its essence. Intellectual programs are not individual either (1) in content (they are not aimed at adapting a particular device), or (2) in terms of volume - single devices in the process of intellectual activity are able to “load” and work out only some part of any of the intelligent programs. That is, it can be said that intelligent programs are detached from specific devices not only in their origin, methods of fixation and transmission (as reasonable programs), but in their essence, gradually forming into new self-sufficient objects of reality, developing according to their own law (for example, such are religious, scientific systems). Individual intelligent devices, becoming for a time intelligent, only realize, fuel this development, being necessary, but already secondary, auxiliary elements. (The issue of classification of intellectual programs deserves a separate discussion.)
In conclusion, brief characteristics of devices of the selected evolutionary hierarchical levels are presented.
- The elementary device operates according to the algorithm (set of algorithms) laid down at the time of its creation (birth).
- An adaptive device is capable of selecting the most efficient algorithms from the redundant pre-set using the method of statistical evaluation of the results of their actual operation.
- The reflective device performs an a priori (without real use) assessment of the success of the application of existing algorithms.
- A reasonable device functions according to external (loaded in the learning process) programs — complex algorithms; it is able to keep new algorithms outside itself in the form of cultural elements; it selects and changes various programs.
- The intelligent device generates new programs.
The proposed evolutionary-hierarchical classification of devices (organisms) gives quite unambiguous criteria for their recognition. Although, of course, there is no clear boundary between the selected levels. For example, a reasonable device without “loading social programs into it” is only reflexive, and an intelligent device outside of intellectual activity (that is, outside the process of creating new programs) is an ordinary intelligent device and in its “rationality” and even “adaptability” can be much inferior to others. (non-intelligent) devices (for example, the notorious scattered professor).
This hierarchical classification of conditional devices can be perceived as hypothetical, only indirectly reflecting the real process of evolution of the nervous activity of living organisms. But, probably, it is this abstraction from real systems that makes it possible to more freely, without delving in particular, to realize, to understand the pattern of evolutionary formation of higher nervous activity.
St. Petersburg, 2004, May