Many studies in the field of artificial intelligence are faced with the problem of the lack of any powerful theory of consciousness and brain activity at the moment. In fact, we have sufficiently poor knowledge of how the brain learns and achieves an adaptive result. However, at the moment there is a noticeable increase in the mutual influence of the field of artificial intelligence and neurobiology. According to the results of mathematical modeling of brain activity, new goals are set for experiments in the field of neurobiology and psychophysiology, and the experimental data of biologists, in turn, in many respects affect the vector of development of AI.
Based on the foregoing, it becomes clear that for the future successful development of bionic AI, close cooperation between mathematicians and neuroscientists is necessary, which in the end will be fruitful for both areas. For this, in particular, it is necessary to study the modern successes of theoretical neurobiology.
At the moment there are three of the most developed and partly experimentally verified theories of the structure of consciousness in the field of theoretical neurobiology: the theory of functional systems PK Anokhin, the theory of selection of neuronal groups (neurodarvinism) by Gerald Edelman and the theory of global information spaces by Jean-Pierre Changet (originally formulated by Bernard Baars). The remaining theories are either modifications of these or not confirmed by any experimental data. In this article we will discuss the first of these theories - the Theory of Functional Systems P.K. Anokhin
Paradigms of reactivity and activity
First of all, it is necessary to say that with all the variety of theories and approaches used in psychology, psychophysiology and neuroscience, they can be divided into two groups. In the first group, the main methodological principle, which determines the approach to the study of patterns of cerebral organization of behavior and activity, is considered reactivity, in the second - activity  (Fig. 1).Fig. 1. Two paradigms of neurophysiology - reactivity and activity
In accordance with the reactivity paradigm, the stimulus is followed by a reaction — behavioral in an individual, pulsed in a neuron. In the latter case, the impulse of a presynaptic neuron is considered as a stimulus.
In accordance with the activity paradigm, the action ends with the achievement of the result and its evaluation. The scheme includes a model of a future result: for a person, for example, contact with a target object .
According to the reactive approach, the agent should not be active in the absence of incentives. On the contrary, when using the activity paradigm, we can admit the case when the agent did not receive any stimulus from the external environment, however, according to the expectations of the agent, he should have acted. In this case, the agent will act and be trained to eliminate the mismatch, which could not be the case in the case of the simplest unconditional response of the agent to the stimulus from the external environment.
Theory of Functional Systems
In the theory of functional systems, not the past is considered as a determinant of behavior in relation to the behavior of an event — a stimulus, but the future is the result . A functional system
is a dynamically folding wide distributed system of heterogeneous physiological formations, all parts of which contribute to a certain useful result . It is the anticipated value of the result and the model of the future created by the brain that allows us to speak not about the response to stimuli from the external environment, but about the full-fledged goal setting.Fig. 2. The overall architecture of the functional system 
(OA - situational afferentation, PA - trigger afferentation)
The architecture of the functional system is shown in Fig. 2. The diagram shows the sequence of actions in the implementation of a single functional system. First, there is an afferent synthesis that accumulates signals from the external environment, memory and motivation of the subject. On the basis of afferent synthesis, a decision is made, on the basis of which an action program is formed and an action result acceptor - a prediction of the effectiveness of the action taken After that, the action is performed directly and the physical parameters of the result are removed. One of the most important parts of this architecture is reverse afferentation - feedback, which allows you to judge the success of one or many actions. This directly allows the subject to learn, since by comparing the physical parameters of the result obtained and the predicted result, it is possible to evaluate the effectiveness of purposeful behavior. And it should be noted that the choice of one or another action is influenced by a lot of factors, the totality of which is processed in the process of afferent synthesis.
Such functional systems are developed in the process of evolution
and learning throughout life
. To summarize, the whole goal of evolution is the development of functional systems that will give the best adaptive effect. Functional systems developed by evolution develop before birth, when there is no direct contact with the environment, and provide the primary repertoire. This fact indicates the evolutionary nature of these phenomena. Such processes have received the general name - primary systemogenesis
The system-evolutionary theory developed by Shvyrkovy VB based on the theory of functional systems, even rejected the notion of “starting stimulus” and considered a behavioral act not isolated, but as a component of the behavioral continuum: a sequence of behavioral acts performed by an individual during his life (Fig. 3) . The next act in the continuum is realized after the achievement and evaluation of the result of the previous act. Such an assessment is a necessary part of the processes of organizing the next act, which, thus, can be considered as transformational or processes of transition from one act to another .Fig. 3. Behavioral-time continuum
From the foregoing, it follows that the individual, and even a single neuron, must have the ability to develop an image of the result of an action and the ability to evaluate the effectiveness of his behavior. Under these conditions, the behavior can be safely called purposeful.
However, the processes of system genesis occur in the brain not only in development (primary system genesis), but also during the life of the subject. System genesis
is the formation of new systems in the learning process. In the framework of the system-selection concept of learning - the formation of a new system - is considered as the formation of a new element of individual experience in the learning process. The basis of the formation of new functional systems in learning is the selection of neurons from the "reserve" (presumably low active or "silent" cells). These neurons can be designated as pre-specialized cells .
The selection of neurons depends on their individual properties, i.e. on the characteristics of their metabolic "needs." Selected cells become specialized with respect to the newly formed system - system-specialized. This specialization of neurons with respect to newly formed systems is constant. Thus, the new system turns out to be an “addition” to the previously formed ones, “layering” on them. This process is called secondary systemogenesis
The following provisions of the system-evolutionary theory:
• about the presence in the brain of animals of different types of a large number of "silent" cells;
• on increasing the number of active cells during training;
• that the newly formed neuron specializations remain constant
• that when learning occurs, the involvement of new neurons rather than retraining of old ones,
consistent with the data obtained in a number of laboratories .
Separately, I would like to note that according to modern concepts of psychophysiology and system-evolutionary theory, the number and composition of the individual's functional systems is determined both by the processes of evolutionary adaptation, which are reflected in the genome, and by individual in vivo training.
The theory of functional systems is successfully investigated by means of simulation modeling and on its basis various models of control of adaptive behavior are built [11,12].
Instead of conclusion
The theory of functional systems in its time was the first to introduce the notion of goal-directed behavior by comparing the prediction of the result with its actual parameters, as well as learning as a way to eliminate the body’s mismatch with the environment. Many of the provisions of this theory now need substantial revision and adaptation, taking into account new experimental data. However, at present this theory is among the most developed and biologically adequate.
I would like to note once again that from my point of view, further development of the field of AI is impossible without close cooperation with neurobiologists, without building new models based on powerful theories.
. Aleksandrov Yu.I. "Introduction to systemic psychophysiology". // Psychology of the XXI century. M .: Per Se, pp. 39-85 (2003).
. Aleksandrov Yu.I., Anokhin K.V. and others. Neuron. Signal processing Plastic. Modeling: Fundamental guide. Tyumen: Publishing House of Tyumen State University (2008).
. Anokhin P.K. Essays on the physiology of functional systems. M .: Medicine (1975).
. Anokhin P.K. "Ideas and facts in the development of the theory of functional systems." // Psychological journal. Vol.5, pp. 107-118 (1984).
. Anokhin P.K. "System genesis as a general pattern of the evolutionary process." // Bulletin of experimental biology and medicine. Number 8, t. 26 (1948).
. Shvyrkov V.B. Introduction to objective psychology. Neuronal basis of the psyche. M .: Institute of Psychology, RAS (1995).
. Aleksandrov Yu.I. Psychophysiology: Textbook for universities. 2nd ed. St. Petersburg: Peter (2003).
. Aleksandrov Yu.I. "Learning and memory: a system perspective." // Second Simon readings. M .: Izd. RAS, pp. 3-51 (2004).
. Theory of system genesis. Under. ed. K.V.Sudakova. M .: Horizon (1997).
. Jog MS, Kubota K, Connolly CI, Hillegaart V., Graybiel AM “Bulding neural representations of habits”. // Science. Vol. 286, pp. 1745-1749 (1999).
. Red'ko VG, Anokhin KV, Burtsev MS, Manolov AI, Mosalov OP, Nepomnyashchikh VA, Prokhorov DV “Project“ Animat Brain ”: Designing the Animatics Control System” // Anticipatory Behavior in Adaptive Learning Systems. LNAI 4520, pp. 94-107 (2007).
. Red'ko VG, Prokhorov DV, Burtsev MS "Theory of Functional Systems, Adaptive Critics and Neural Networks" // Proceedings of IJCNN 2004. Pp. 1787-1792 (2004).