Intelligent cognitive system with multi-system knowledge integration: feasibility and approaches to formation

Gribkov Andrei Armovich ORCID: 0000-0002-9734-105X Doctor of Technical Science Leading researcher; Scientific and production complex 'Technological Center' 124498, Russia, Moscow, Shokin square, 1, building 7 andarmo@yandex.ru Other publications by this author

Zelenskii Aleksandr Aleksandrovich ORCID: 0000-0002-3464-538X PhD in Technical Science Leading researcher; Scientific and production complex 'Technological Center' 124498, Russia, Moscow, Shokin square, 1, building 7 zelenskyaa@gmail.com Other publications by this author

Introduction

The development of information technology in the last decade, manifested by the appearance of the first implementations of narrow artificial intelligence in the form of machine learning systems, has a noticeable impact on the organization of the knowledge system: its structure, methods and means of expansion. The observed transformation of society associated with the transition to a civilization of cognitive technologies ^[1], characterized by the further development of technologies in the direction of intelligent machine control systems, requires a methodology for the formation of knowledge beyond the capabilities of reason, i.e. the use of knowledge and methods accumulated in each subject area. It is necessary to build a knowledge system that will become the basis for the functioning of the mind in the form of creative artificial intelligence capable of solving creative tasks.

The problem of building this knowledge system consists of several components.

First, it is necessary to substantiate the possibility of a system on the basis of which artificial intelligence can be formed. It depends on whether creativity (creative intellectual activity) is an irrational and transcendental process ^[2] that cannot be formalized within a deterministic methodology, or whether creativity does not differ qualitatively from rational intellectual activity in the form of solving computational or logical problems ^[3]. The knowledge system required for the formation of artificial intelligence is possible only if the mechanisms of creativity are rational.

Secondly, it is necessary to determine the mechanisms of systematization and replenishment of knowledge in this system. Previous research conducted by the authors ^[4] allows us to identify a multisystem integration mechanism as the main one. The principle of operation of this mechanism, which natural cognitive systems (in particular, human intelligence) are endowed with from birth, is the ability to collect knowledge in all systems into which the subject of cognition is integrated (i.e., of which he is a part), identify isomorphism in them in the form of patterns of forms and laws, and use formalized knowledge from one system. in other systems. As a result, an associative database is formed in the mind ^{[5, p. 276]}.

Thirdly, it is necessary to define methodological tools that include collections of systematized patterns of forms, laws, and other knowledge generalized through multi-system integration, as well as means of universal (domain-specific) representation of patterns. The authors' research has shown that system primitives ^{[6, pp. 249-257]} (as a means for describing patterns) and neural circuits ^[7] (as logical elements for describing patterns, typical combinations of primitives or other elements of structure or connections) can play a significant role in this.

Creative artificial intelligence (AI) systems will inevitably implement the same methods and approaches that are used in natural (in particular, human) intelligence. However, the formation of objective theoretical foundations for building creative artificial intelligence systems in the form of correlation and expansion of ideas about natural intelligence is a dead–end development path. It is necessary to form these theoretical foundations on the basis of universal concepts that generalize natural and artificial intelligence, as well as include a wide range of cognitive systems.

According to the authors, the object of research should be cognitive systems – multilevel systems that perform the functions of recognizing and memorizing information, making decisions, storing, explaining, understanding and producing new knowledge ^[8].

According to another definition, a cognitive system is a constantly operating complex adaptive system that autonomously explores and reacts to the environment, with the ability to "survive" ^{[5, p. 229]}. "A cognitive system is not necessarily intellectual, but it can be so in principle" [ibid., p. 230]. Natural and artificial intelligence, complex adaptive control systems (technological processes, production facilities, vehicles, etc.) are special cases of cognitive systems.

According to the authors' definition, cognitive systems are autonomous cognitive systems with self–awareness. At the same time, self–awareness is just the ability to separate oneself from others, which is a necessary condition for cognition, which presupposes (within the framework of the epistemological interpretation of subject-object relations) the existence of a subject and an object of cognition.

The genesis of creativity

The assumption that creativity is not something more than an ordinary mental activity that generates tangible or intangible objects obviously does not correspond to the ideas that have long been unconditionally accepted in culture. Intuition and epiphany, images and ideas that appear as if from nowhere, etc. – all this creates the illusion of irrationality of creativity, its transcendence, the impossibility of its logical comprehension.

Creativity is a diverse phenomenon, realized in a multitude of forms that do not fit well into formalized definitions. Thus, scientific creativity and other creativity, which consists in solving intellectual problems for which there is no known method of solution ^{[9, pp. 39-40]}, differs significantly from artistic creativity, which results in the birth of original and artistically valuable tangible and intangible objects. Artistic creativity, as research shows ^[10], is a phenomenon for which even the definition of basic concepts (for example, the concept of artistic value) presents significant difficulties.

It can be assumed with a high degree of confidence that the nature of creativity in all its forms and manifestations is fundamentally of a common nature. If our assumption is wrong, then further reflection based on this assumption should reveal contradictions that will expand as the field of study grows. The need for this growth of contradictions follows from the rule of consistency: "the absence of contradictions between limited knowledge covering a sufficiently large area of being and the data of its sensory perception is an indicator of the reliability of knowledge" ^{[6, pp. 195-205]}.

Assuming a common nature of creativity, we will focus our attention on a less complex form of its manifestation – solving intellectual problems. How is the solution of an intellectual problem found in the process of scientific creativity? Where does it come from? To answer these questions, it is necessary to state two facts that are reliably empirically confirmed and logically difficult to dispute.

First, as Aristotle wrote, "the objects of thought are in sensually comprehended forms" ^{[11, p. 405]} or, as J. R.R. Tolkien wrote. Locke, "all ideas come from sensation or reflection" ^{[12, p. 154]}. This means that the creator draws all his ideas from the world around him.

Secondly, knowledge of the world reveals the ubiquity of isomorphism – the phenomenon of similarity of forms and laws at different levels of the organization of the universe, in various subject areas. Particular manifestations of isomorphism of forms and laws in the universe can be formalized in the form of patterns – patterns of forms of relations of elements within the system, widely distributed in various subject areas.

These two facts give grounds for affirming the connection of creativity with the real world in its unity, manifested in the form of isomorphism of forms and laws: creativity is the implementation of the integrity of the world ^[10], i.e. simultaneously the realization of this integrity (manifested through the use of patterns in creativity) and one of the mechanisms of its formation (based on the creation of new material and intangible objects compatible with and complementary to the universe).

If creative thinking does not differ qualitatively from rational thinking (for example, related to calculations or logical constructions), then it can be determined, i.e. to determine the mechanisms of its implementation and the necessary methodological tools for this. Since, as we have stated, creativity is based on the integrity of the universe, when defining creativity, one should rely on the general theory of systems, a field of knowledge, the subject of which is the integrity of the universe, manifested through the isomorphism of its forms and laws ^{[13,14,15,16]}.

The successful development of the general theory of systems, which makes it possible to comprehend the phenomenon of creativity, requires the expansion and systematization of existing knowledge about the manifestation of isomorphism in the universe. This systematization involves the creation of representative collections: patterns – widespread patterns of forms and relationships of elements within systems, primitives – typical elements from which patterns are "assembled", as well as secondary laws that are reliably confirmed empirically, but for which the internal mechanisms are not fully deterministic. In general, the indicated expansion of the knowledge system means their ontologization ^[17], i.e. a more reliable correlation with reality.

Multisystem knowledge integration

To date, science does not possess reliable knowledge and a detailed understanding of the mechanism of implementing multi-system integration of knowledge in human consciousness. We can judge this instrument of cognition, which makes creativity possible, mainly as external observers. Our individual experience of observing the process of our own thinking can also serve as a source of knowledge about the multisystem integration of knowledge. Unfortunately, the significant contribution of subjectivity makes the reliability of knowledge obtained from this source low. Despite these difficulties of cognition, it is possible to state some significant observable facts.

Firstly, to implement the mechanism of multisystem integration, it is usually necessary to have a substantial body of knowledge gained from their various systems. This body of knowledge is formed by the education and life experience of the bearer of consciousness and can be interpreted as wisdom.

Secondly, in some subject areas, the degree of formalization of knowledge generalized through multi-system integration is very high. For such fields (music, mathematics, and partly painting), it is possible to productively transfer knowledge between a limited set of systems: in all of them, the laws of harmonic, hierarchical, and other forms of object construction are clearly manifested. In this case, multi-system integration becomes possible even without a significant amount of knowledge or wisdom.

It follows from these two facts that the success of multisystem integration depends on the amount of knowledge about various systems and the scope of application of this knowledge. Different subject areas are unequal in the complexity of the patterns that define them, and as a result, different amounts of knowledge from different systems are required to identify and identify these patterns.

The mechanism of multisystem knowledge integration can be conventionally represented as consisting of two functional components: the mechanism of the birth and consolidation of neural circuits and the mechanism of their generalization and determination, including the use of patterns and primitives. Natural intelligence allows both of these components to be realized, therefore, the mechanism of multisystem integration is fully activated, providing the potential for solving intellectual problems and creativity in general, which expands with the growth of knowledge. Currently existing implementations of narrow artificial intelligence in the form of machine learning systems do not include the use of patterns and primitives by their creator (human). As for the formation and consolidation of neural circuits, this mechanism develops and activates spontaneously in the case of artificial neural networks. According to the authors' research, "an artificial cognitive system in the form of a deep machine learning system implemented on the basis of an artificial neural network will independently identify and generalize complex dependencies between input and output data in the form of appropriate communication coefficients between neurons during the learning process" ^[18]. The training of an artificial cognitive system at the initial stage of its formation can be autonomous, when learning and work are separated in time. For complex systems with extensive functionality, online learning is necessary, i.e. learning in the process.

In the case when an artificial cognitive system is trained on data from a large number of heterogeneous systems, the formed neural circuits can reflect not only the elementary relationships between the elements of one system, but also record (based on the results of comparison) the typical relationships of elements in different systems. We are equally unaware of the form in which neural circuits exist in natural and artificial cognitive systems. The process of their formation is not monitored, however, there is no doubt that something conceptually corresponding to neural circuits exists and can serve as a basis for further generalization of knowledge within the framework of the implementation of the multisystem integration mechanism.

An indirect evidence of the reality of this mechanism of formation and functioning of neural circuits is the possibility of constructing topological neural maps based on general self-regulating feedback. An example is a topological map connecting the retina to the primary optical cortex ^{[5, p. 260]}.

How can we make the transition from neural circuits (non-deterministic and non-formalized) to systemic holistic knowledge formulated in the form of patterns, primitives and secondary laws? Probably, considering a set of neural circuits (implemented, for example, in the form of a "trained" artificial neural network) as a "black box" ^{[19, pp. 127-169]}, i.e. an object whose properties are studied based on the reaction to external influences. Based on the generalization of the totality of the reactions obtained, it is possible to build a "black box" model that allows predicting the reaction of the "black box" to their changes within certain limits of the values of the input parameters. A model demonstrating high reliability should be isomorphic to the modeled system ^[20], which opens up the possibility of a qualitative (in the form of a choice of appropriate patterns of forms and relationships) description of the original system, in this case, a set of neural circuits.

The formation of systemic holistic knowledge does not necessarily have to be carried out solely through the study of models that generalize a set of neural circuits. For example, the systemic holistic knowledge available to mankind is largely formed deductively as a consequence of a priori accepted metaphysical concepts. Within the framework of the empirical-metaphysical general theory of systems ^[6], the acceptance and use of a priori metaphysical knowledge is formalized in the form of primary properties of being, basic and primary laws of being.

The effectiveness of using metaphysical knowledge is determined by several factors.

Firstly, metaphysical (a priori) knowledge generated by pure reason, if it is correct, has high reliability. On the contrary, knowledge obtained a posteriori on the basis of generalization of experience and its integration into the knowledge system is inevitably distorted. This distortion is due to the variability of possible interpretations of empirical knowledge and, most importantly, the fundamental difference between the logic of cognition, based on hierarchical constructions, generalizations, probabilistic representations, etc., and the "logic" of constructing the universe, in which everything is concrete, not repeated and not generalized.

Secondly, systems formed by pure reason as a consequence of metaphysical knowledge are open to knowledge and can serve as the basis for building patterns of forms and laws that can later be applied to more complex, not fully deterministic systems. Taking into account the isomorphism of the universe, these patterns of forms and laws will correspond to the organization of cognizable systems at all levels of the organization of material existence.

Earlier, we stated that both natural and artificial cognitive systems are capable of forming aggregates of neural circuits, forming cognitive models of these aggregates. The reliability and completeness of the holistic knowledge system formed in this way inevitably remain low. According to the authors, it is not possible to rise in generalization to the level of patterns of forms and laws based on induction. Achieving this goal requires movement not only from below (from empirical experience), but also from above (from a priori metaphysical concepts).

The question of whether an artificial cognitive system (or a networked set of such systems) is capable of forming the necessary metaphysical representations, according to the authors, is not important. Approaches to the formation of such representations are defined, and representative collections of patterns and secondary laws are being formed ^[6]. Their formalization will open up the possibility of embedding not only in artificial intelligence systems based on neural networks, but also, probably, in rigidly deterministic computing machines operating with strictly defined algorithms, which will include algorithms for searching and matching patterns and secondary laws from existing representative collections.

Further development of cognitive systems requires the formation of a much broader theoretical framework than currently available. The key areas of necessary theoretical research are the general theory of systems, synergetics and the theory of cognitive systems. A promising version of the general theory of the system is the empirical-metaphysical general theory of systems mentioned above ^[6], which already outlines all the main development vectors necessary for cognitive systems. The synergetics of cognitive systems can be built on the basis of G. Haken's research: a synergetic interpretation of cognitive activity ^{[21, 243-314]}, a synergetic approach to the study of complex nonequilibrium systems ^{[22, pp. 36-37]}, a study of the hierarchy of instabilities in self-organizing systems and devices ^{[23, pp. 36-38]}, etc. The final generalized version of the theory of cognitive systems has not yet been formed, but work is underway in this area. Among them, one can single out the research of Prof. K. Gros from the University of Frankfurt ^[5].

Conclusions

Let's summarize the research conducted in the article:

1. Further development of information technologies in the field of intelligent control of machines requires the construction of knowledge systems for them that will ensure their functioning in the form of creative artificial intelligence capable of solving creative tasks.

2. The possibility of creating such knowledge systems depends on whether the creative process is formalized through a deterministic methodology similar to rational intellectual activity. The authors' research so far shows that creativity is an implementation of the integrity of the world and can be formalized by means of the general theory of systems.

3. The main practical mechanism for building a knowledge system for creative artificial intelligence is multi-system knowledge integration, which consists in the ability to collect knowledge in all systems into which the subject of knowledge is integrated (i.e., of which he is a part), identify isomorphism in them in the form of patterns of forms and laws, and use formalized knowledge from one system to another systems.

4. An important tool for low-level generalization of data and knowledge in general, which is one of the sources of the formation of systemic holistic knowledge, are neural circuits that reflect the elementary relationships between the elements of one system, as well as (based on the results of comparison) the typical relationships of elements in different systems.

5. Further development of cognitive systems, including at the level of creative artificial intelligence, will require in-depth theoretical research in the field of general systems theory, synergetics, and cognitive systems theory.