Simple architecture of strong artificial intelligence

Everything incomprehensible seems complicated, confusing and difficult to explain. Such an attitude is completely understandable with respect to a strong AI. Our brain can do so many things that it involuntarily seems that in order to explain the principles of its work, it is necessary to understand the most complicated intricacies of many different principles and various systems. Actually, this generates appropriate approaches to the study of AI, which are described from time to time on this server. Anyone who is professionally engaged in AI, over time, is determined with a certain philosophical view that takes on a religious connotation. Trying to refute or even question this view is a thankless task. In the end, only a working model of AI is a strong argument in favor of one of the theories. So please take my story solely as my personal point of view. So, I am convinced that we have already managed to formulate the basic principles necessary for building a strong AI and are mostly understandable, and that there is a way of thinking and how to model it in full. Details under the cut.

I will very briefly describe the basic principles giving a combination of clues to thinking. Not everything may seem obvious, so at the end I will give a link to a more detailed and demonstrative presentation.

Behavior
')
From everyday experience, it seems: that there are reflex actions, there is an instinctive behavior, it happens that actions are a result of emotions experienced, that often an act is the result of a planned, deliberate decision. Unwittingly it is assumed that the brain contains modules that implement each of these strategies. In fact, all actions are either reflexive or dictated by memory, which has experience of behavior in such situations. It can be shown that the whole diversity of variants of human behavior fits into this scheme. The main principle: everything that happens to us receives a rating of “good - bad”, memory fixes what is happening, taking into account how our state changes, having encountered a familiar memory situation causes us to repeat actions that promise a change of assessment towards “good” and blocks actions leading to deterioration. states. This great principle is the basis of behaviorism and is used in a number of psychological theories and in cybernetics in reinforced learning systems. The main question that arises from this principle is what is the nature and mechanism of “good to bad” assessments?

Emotions

Different theories interpret differently what emotions are. According to Darwin, emotions are strongly changed as a result of evolution of the reaction to various vital introductory. Called in simple situations, these reactions manifest themselves in subsequent, more complex life circumstances that are connected with the original ones. Darwin was mainly concerned with the explanation of mimic reactions. Reasoning about the relationship of emotions and evaluation, demanded assumptions about the mechanisms responsible for the behavior. And such theories appeared a little later.

According to Anokhin, any organism knows its optimal state, and strive to stick to it. Then negative emotions are signals of a deviation from the norm, and positive emotions are incentives for a return to optimum.

In Leon Festinger's cognitive dissonance theory, emotion is a signal that describes the consistency of interacting systems. A positive emotional experience appears when the implemented plan of action does not encounter obstacles in its path. Negative emotions are associated with a mismatch between the current activity and the expected result.

Pavel Simonov, who linked emotions and the likelihood of satisfaction of a particular need, came closest to understanding emotions. According to Simonov, "... emotion is a reflection by the brain of man and animals of any actual need (its quality and size) and the probability (possibility) of its satisfaction, which the brain evaluates on the basis of genetic and previously acquired individual experience."

What unites, perhaps, all these theories is an explicit or implicit indication that all basic emotions are certain properties resulting from natural selection. And although it was assumed that the acquired individual experience could correct innate emotions, the hereditary genetic nature of emotions was never questioned.

This statement is the main delusion. Only assessments of sensations that have a sensory nature are genetically predetermined. What we perceive, for example, as “bad,” caused by cold, hunger, pain, or “good,” follows from taste or erogenous receptor signals. All other assessments of what is happening, what we actually call emotions, are in no way laid down genetically and are not inherited.

Describing the behavior, we said that the memory captures what is happening and the accompanying signs, taking into account the change in the state of “good - bad”. This allows you to subsequently form a behavior. So it turns out that exactly the same neural mechanism, applied not to the motor zones of the cortex, but returns the excitation back to the structures responsible for the state “good - bad” and gives rise to emotions.

That is, emotions are the ability of memory to form a “good - bad” state, based on accumulated experience. From the moment of birth, with respect to all observable signs and unobservable factors, experience is accumulated about which state “good - bad” corresponds to them. If a symptom occurs equally with both the “good” and the “bad” condition, then its final score becomes neutral. If it turns out that a sign or factor is statistically more associated with one of the estimates, then an emotion is formed. All positive emotions can be interpreted as anticipations of certain pleasures, all negative ones as fears of something bad. Initially, "good" and "bad" are the evaluation of sensations. Fears of “bad” and anticipation of “good” are formed on their basis. Since these fears and anticipations are themselves evaluations and create a state of “good - bad”, then fears of fears and anticipations of anticipations are formed on their basis. At the same time, emotions are not associated with any analysis of the situation, but arise as a reaction to a symptom, in relation to which the brain has determined a statistical connection with the state. As in Simonov’s theory, emotions are connected with the probability of receiving pleasure or displeasure, but the idea of ​​purpose and need is rejected. Emotions are purely statistical in nature and reflect the patterns of the influence of the surrounding world on our state of “good - bad”.

It turned out that in this approach all the emotions peculiar to man inevitably arise as a result of the interaction of those evaluations of sensations that are predetermined by the genetically and natural and social environment in which the person is formed. It was possible to describe the ways of formation of all basic emotions.

Beauty and harmony

The traditional view of beauty and harmony dates back to Plato. His ideas, one way or another, are paraphrased in various arguments about the nature of the beautiful. The basic idea: the idea of ​​beauty is inherent in us genetically, as some kind of higher knowledge about the world around us. The development of the ability to perceive beauty allows a person to penetrate deeper and deeper into the ideal world of harmony. Harmony itself is perceived as a kind of objective entity that exists independently of a person.

Based on the fact that the perception of beauty is nothing more than an emotional assessment, we were able to show that all our perception of the beautiful is a combination of individual experience of interacting with the outside world. In other words, that there is no certain objective beauty, but our adaptation to the existing environment, and if we were surrounded by a different world, there would be other ideas about beauty. But each of the existing factors of beauty nonetheless has a completely rigorous explanation of why it is formed in this way and why this formation is more or less the same for all people. Within the framework of the described theory of emotions, it was possible to give an explanation of all the basic forms of beauty.

Humor

All existing attempts to give an explanation of humor are explicitly or implicitly based on evolutionary theory. They proceed from the fact that humor or the elements from which it is composed, arose as a result of natural selection and carried with them a certain expediency. It was possible to show that in the framework of the proposed theory of emotions, humor arises as an inevitable consequence of the social interaction of people. Humor is described as a combination of two emotions: the emotions of the beauty of information transfer and the emotions of funny. For each of these emotions, the path of its formation and the laws of its manifestation are clearly visible.

It is noteworthy that by rejecting the idea of ​​the genetic nature of the ridiculous, it was possible to give an exhaustive description of everything connected with humor. Everything, even the smallest details, found their obvious explanation.

Thinking

Understanding the nature of emotions allows you to describe the nature of thinking. Thinking itself is the essence of modeling some virtual situations that receive emotional evaluation and are imprinted in the memory in the same way as real experience. Such virtual experience contributes to the formation of behavior along with real experience.

But the description of thinking requires an answer to the questions: about the nature of thought, about the form of its representation in the neural structure of the brain, about the nature of memory, about how memory fixes an existing thought, how focusing happens, how the next thought is born. And here it turns out that a thought can be described as the activity of a certain set of concepts previously singled out by the brain. But not an arbitrary set, but one that causes the strongest emotional evaluation. In other words, thought has an emotional basis. That is, if a person perceives and recognizes a certain set of objects, then a certain subset of them, activating a previously formed emotion, forms the focus of attention. Simply put, we can see and hear a variety of stimuli, but only those that are associated with the strongest emotion will form a thought. The principle of selection from a large set of perceived and associatively related images of a subset that forms a thought, through focusing with the use of emotions, allows us to describe the process of thought formation. The most interesting thing is that such a construction is elegantly described only under the assumption that emotions are statistically isolated assessments of a universal nature, localized in the areas of the cortex, and not tied to specific genetically determined brain structures.

Brain structure

Unlike the classic computer, the design of which does not tolerate the slightest mistake, the brain has an amazing ability to self-organize and plasticity in the use of its resources. The unity of the mechanism of formation of behavior and the mechanism of formation of emotions allows us to build an extremely simple and at the same time extremely plausible model of interaction of brain structures. The model is based on the idea of ​​the cortex zones as structures, onto which information is projected as a “picture” zone distributed over the surface. Memorizing and generalizing the experience gained through self-organization, the cortex zone forms an “output picture” as a set of activity of efferent neurons, corresponding to the activity of the cortical columns of the corresponding zone of the cortex. The zones of the cortex have a complex system of mutual projections. The structure of the projections is genetically predetermined and determines the probability of obtaining a zone of a particular specialization. Zones onto which sensory information is directly projected acquire specialization as primary sensory zones. The specialization of other zones is determined by the type of information projected on them. Beams of signals from pleasure centers and pain centers are projected onto some zones. This allows generalizations forming on such zones of the cortex to acquire an emotional coloring. If we assume that the cortex zones are able to reproduce the memorized patterns and return the projections back, restoring the original images in other zones, then a reverse projection on the motor zones will give us actions, on the sensory zones - fantasies, on the “good - bad” centers - emotions. That is, the diversity of brain functions can be explained by a single universal mechanism, besides self-adjusting and fault-tolerant. If you use patterns that cause emotions to slow down everything that is not related to the information that caused their activation, you can get a realization of the focus of attention and thought formation.