Awareness of human nature through the understanding of AI. Introduction Part 1


The introduction is devoted to how the course of events and conclusions led to the emergence of the idea and the beginning of its implementation.

For a long time, I have a problem with AI. In the beginning, these were attempts to make an advanced chat bot. Then came the realization that this will be just a program that will respond to a request in accordance with its database. Due to the fact that all knowledge of the system will be filled relatively randomly (in the process of speaking, learning), we get pseudo-unpredictable answers, but the system itself does not realize that it is responding. Next was an attempt to develop a system with more advanced sources of knowledge, in the hope that this would give some impetus to the emergence of AI. Processing images from the camera, sounds from the microphone, an attempt to create binocular vision, to identify the images of volumetric objects ... All this was just another data entry in the database. Undoubtedly, all this is useful and necessary in robotics, but it will not lead to the intellect itself.

Now we can see the wonders of artificial neural networks (INS): searching for faces, coloring pictures, writing texts, changing seasons in photographs, etc. All this is incredibly cool, and every year it is only gaining momentum, but unfortunately, this is again not an AI in the full sense of the word. These are libraries with databases that are trained to solve a specific task, and they solve it very well. We just learned to make fundamentally different libraries, which now do not just give, for example, translation of a word from a dictionary, but translate it in accordance with the vast knowledge base of the use of this word (And yes, coloring, changing faces on video is also everything, in principle , the library's response to the request, albeit unusual for our perception).

Most people believe that AI must transcend the human mind, or at least be similar, but should it be so? Dogs, cats - do they possess intelligence? Of course yes. He may not be like ours, someone even says that he is at the level of small children, but he is. And if you go further down the evolutionary ladder? In the end, it can be concluded that even the simplest or viruses have some sense of life and a behavioral model.

In 1970, John Conway came up with a cellular automaton called Life . Having laid down simple and reasonable principles, he got a terrific effect. Let's call the machine a game. It has become a new world in which its laws and rules appeared, completely unpredictable at first glance.

But only the following principles were laid:

  1. The game takes place in a certain cellular plane.
  2. Each cell on this surface can be in two states: to be “alive” or to be “dead” (empty). The cell has eight neighbors (surrounding cells).
  3. The distribution of living cells at the beginning of the game is called the first generation. Each next generation is calculated based on the previous one according to the following rules:
    • in an empty (dead) cell, next to which there are exactly three living cells, life begins;
    • if a living cell has two or three living neighbors, then that cell continues to live; otherwise (if there are less than two neighbors or more than three), the cell dies (“from loneliness” or “from overpopulation”).

Conway's Game of Life

Cellular machine "Life". One of the development options

For those who are not familiar with this entertaining discovery, I advise you to meet. There are many working examples and articles on the net.

Thus, I lead to the idea that intelligence itself is a very unambiguous concept. Understanding the difference between lack of intelligence, weak and strong for living beings is a non-trivial task. How reasonable is the bacteria? Or is the cat aware of itself?

Now a lot of people are actively studying the work of the brain, trying to understand and shift the biological mechanisms on silicon chips. Being a software engineer by profession, I did not dare to go into the wilds of neurobiology, but came to a very simple conclusion. Everything in our world can be described one way or another with the help of formulas or programs. It is not necessary to know thoroughly how this or that mechanism works, the main thing is to catch the algorithm, the principle. You can abstract complex elements, leading them to a formula or algorithm.

INS is an excellent example of our memory, and maybe even more. And most importantly - the algorithm works. The question remains for the search algorithm for the remaining elements.

So the idea was born that the key to understanding and creating AI is not in an attempt to create something that could live in our world and perceive information like you and me. The basis of the theory is that our mind works on simple algorithms, but due to the presence of multiple levels of images, as well as a huge variety of random data of our world, we get an absolutely unpredictable and complex system that seems impossible to know. A vivid example: almost no developer of a complex INS can say for sure why he received such an answer from the network, and not another. It seems almost unpredictable. Or Conway’s “Life” is a completely new, unpredictable world that seems complicated and thoughtful, with its inhabitants and laws, and in fact it contains a simple algorithm in a few lines.

It is necessary to simplify the environment and the object of study. In our world there are a lot of variables that are difficult to consider, interference. It is necessary to create a closed environment with limited laws, a certain model of the world in which an entity with a basic set of algorithms will be added. Basic concepts and goals are defined, guided by the following principles: they should be as close to human ones as possible; they will be added and introduced into the system gradually by the level of their importance for the life of the entity. But how to find and define these algorithms? Analyze yourself and other people around. Ponder the true causes of their behavior, actions, feelings, sensations, desires. A simplified example: an entity must want to eat. If you don’t eat, then you experience pain, which is a stimulant to search for food. When used, it will receive pleasure, and if it is excessive, pain.

As the behavior of the essence and laws of the world becomes more complex, we will be able to observe the development of the system and draw our own conclusions. If you manage to achieve some kind of breakthrough, you can simply replace the basic concepts of the entity, and transfer the entity from a closed system to an open one. For example, from a closed virtual sandbox - to our world on a robot platform, replacing hunger with a battery discharge.

Another important point. Since the basis of the behavioral model will be taken human needs and characteristics, we will be able to realize and understand many aspects of the very essence of man. Psychology of behavior, social interaction, the meaning of life and death - this is so, the tip of the iceberg, which comes first to mind.

By the way, does this remind you of anything? Development in a closed environment, where there is practically nothing, where the essence from one cell develops to a complex being, and then goes out into the world? How old do you remember yourself? Since 3 years old? And before? So it turns out that for 9 months you developed in a completely closed environment, and after another 3-4 years you learned how to perceive our world as it is. And only after that in your memory began to form images compatible with the current consciousness, which you remember to this day. Many believe that some complex actions or a sequence of actions in us are “laid down” ( instincts ). But how true is it? Is it possible to lay in the living creation at the gene level some actions with high images that do not appear immediately after birth? We intuitively fear many things (complex images). I had two dogs, one from childhood was afraid of water before the panic, and the second adored her, dived and did not climb out. Instikty? Can our mothers somehow convey some of their particularly “vivid” memories to us? After all, according to the logic, we are all someone's children, so how can complex instincts somehow be passed on to us at a higher level than the gene level? Indeed, in the umbilical cord, besides vessels, there are also nerve fibers. The baby feels when mom is nervous. Perhaps he defines all this in a different way, but still ... I really hope that the community will be able to give answers to at least some part of these questions.

In conclusion, I would like to say, perhaps as a result of this work, it will not be possible to create a strong AI, but to understand it better than myself is definitely. No one really knows how deep this hole is.

Part 1. Start

In this section, I will try to determine the primary requirements for the environment, as well as the basic algorithm of the entity.

First we define the environment in which we place the entity.

Let it be some kind of two-dimensional space divided into cells (50 by 50 cells), which is impossible to get beyond. Accordingly, in the first version of the world, a cell can either be empty, or contain an entity, or an obstacle, or food, or bad food, which all conventionally occupy one cell. It is possible to move in the environment only along the neighboring cells vertically and horizontally. Time in the world is quantized (divided into approximately equal periods of time). The essence is placed randomly, as well as obstacles with food around.


Let's think about it. What does an entity have to have in order to want to take the first step and define some task for its existence? If we put an entity without desires and needs in space, it will stand, and will not do anything. We ourselves can sometimes be lazy and not want to do anything - just lie down. Why is this condition manifested? Perhaps at that moment all our basic needs are met, and everything else is not so important that this can be neglected. What needs make you drop everything and do anything you want, if only they are silent? According to Maslow's pyramid, physiological needs underlie everything. Our mind is able to suppress them, but sooner or later, physiology will make us satisfy them, otherwise we may die. This may include the need to eat, breathe, go to the toilet, multiply, etc.

Let's unite them all and we will reduce to the abstract need to eat. Our essence must be hungry, otherwise it will be bad, painful. This is perhaps one of the most powerful motivators.

From here you can immediately determine that the entity should consist of different blocks, let one be responsible for the desire and satisfaction in food, and the other will receive all these signals and decide what to do.

Note that if our stomach did not tell us about hunger, we would not have thought about food (for now we’ll omit the moment with pleasure from food). Accordingly, we would not do everything related to the satisfaction of this need. And if you think about it, then we do a lot just to eat. And to multiply?

We added desire, but will the entity move now? Not. She will stand and, realizing this desire in one form or another, will simply starve, and in the end will die of hunger. Why? Because she does not know what to do to meet the need. This is where the most difficult and interesting begins. There must be some kind of trigger that will make her jerk off and stumble upon food. For example: initiate a cry for an infant to be fed.

I think everyone has come across or at least heard that in case of danger we can do things somewhat superior to our ordinary capabilities. Jump high, climb, etc. For example, when you run away from a dog. Why is this possible? Why in a conscious state, we can not do it, but here it turns out? You and I have one mechanism that is considered negative, but perhaps it is he who is called upon to save us in such situations. In a state of danger, we can lose control of ourselves and we can go into a certain state of affect.

If you think about it, then there are so many similar mechanisms in us that are not manageable, i.e. our brain disconnects us from the control of the body, and does everything to survive. It is funny, the feeling that when creating a person, someone made a “protection against a fool”. For example, you will not breathe for long - the brain will interrupt control anyway and you take a breath. Bring your hand to the fire - the brain itself will pull it away. Something will fly in the eye - it will close it without your knowledge. Yes, you can refuse to eat or come up with an ingenious way to kill yourself, but all this is too difficult for this “protection against a fool”, perhaps that is why we can get around it in some cases. But there is a higher level of "protection" - affects, when our brain just dulls the mind, or enters us into a short state of increased activity. The brain continues to work and can carry out some calculations based on the available in memory, as well as incoming data. He does it more quickly, turning off some non-critical areas. One goal - to survive. In this state, we are able to make more accelerated decisions, either to panic altogether, which may save us, or kill us ... (Sometimes it seems that we are prisoners. We were given control levers and put in a cage that lives on its own, we can only manage a small part of it.)

If anyone is interested in the details read about the affect and reflexes in the relevant literature, perhaps you will discover many new things for yourself, and we will go further.

So, our essence should have a mechanism similar to the affect that will push it in search of food. If she is lucky, she will find it before death, and then the positive effect of eating will be fixed in the memory, and according to the logic of the process she will have to realize this and start looking for food further herself, without the work of protective mechanisms.

I will not publish early ideas and concepts, in view of their current irrelevance. Instead, I will try to abstract as much as possible what is already there for initial understanding. A more detailed analysis of the units and systems will be carried out in the following articles.

As a result, the following blocks were defined in the current version:

Diagram of interconnections between AI blocks

Further, all the blocks in the text for convenience will be written in bold to make it clear that we are talking about the block.


We define the following system entities:

World version 1.1

So, imagine that the entity appeared in the void. After a while attention registers the “hunger” signal in information systems , which is becoming more and more high. Having reached a certain limit, the “hunger” begins to be recorded with an already negative assessment, a too strong signal. If the signal level reaches a certain level, we will assume that the entity has died of pain. Notice that for the entity it is a completely new and unknown signal. It only implies that if any signal has a level greater than some constant (pain threshold), then the signal is considered negative. She is asked to remember what the signal is, how to evaluate it? And there is not a single record, and she can neither assess it nor determine what to do next. This is a new feeling. Accordingly, it does not threaten life by obvious signs, and it is ignored by the shadow and the ghost, simply by writing it in memory in both the short-term and the long-term .

In the first versions, the entity was waiting for the moment when the defense mechanism would turn on, when the level would exceed the pain threshold in order to start moving. Now the ghost itself makes a decision without this mechanism. He waits until the signal becomes negative and then begins a systematic enumeration of all possible actions with writing the result into memory. This condition can be described as "tedious." Something is bothering you, but you don't know what. And you diligently try to understand what it is, by doing this one, then the other, if only the anxiety is gone.

It is also worth noting that in the first versions there was a separate universal pain signal. Which imposed an additional negative color. He walked with a couple hunger and could go with any other. Later it was decided to get rid of it and evaluate the pain as exceeding a certain threshold by any signal. We are uncomfortable to hear loud sounds or to look at the bright light.

So, as the signal passes through critical levels, the shadow will take on the role of a ghost . It will be more and more aggressive to sort through the possibilities until the entity dies or stumbles upon food. With death, everything is clear: the end and start all over again. But what will happen if the entity stumbles upon food?

The entity will satisfy its needs, receive a whole pile of positive signals, and eventually the first sequence of actions will be recorded to get rid of hunger.

Here you can see that hunger will receive a negative color in the memory, but after it is satisfied, this color will be normalized by a positive effect. In the case of the bite of the beast, for example, there will be no normalizer, and the signal will remain negative.

Then the essence will begin again to slowly hunger. And here the algorithm for predicting events from memory will work. The signal level becomes more negative, and in the foreseeable future it will be very bad. Need to find a solution. And notice to look for a solution, the entity will begin before it really becomes bad like the first time.

A similar situation will happen with a bad meal. Having satisfied the hunger and experienced pain afterwards, she will no longer eat bad food, unless in a pinch, since the final pain from bad food is still lower than the death pain.

With a large number of iterations, the entity will understand how to deal with pain, how to bypass blocks, what to eat and what not.

It is worth noting that at this stage the entity does not need to have a focus of attention or a shadow. They are listed for greater authenticity, and a smooth transition to the next version of the entity. I did not pay attention to how exactly the signals fall into both memories and how they interact at all, there will be separate articles for this.

And in conclusion I will make a retreat. In charge, I raised the issue of the transfer of experience to children from parents without training. On one of the future versions, when the offspring appears, the question of laying some basic features from the parents will be acute. Character, fears, way of thinking. And logically, the need arises for a mechanism to transfer some of the critical experience from parents to children. , , ( ). , . , (, ). . , … , - . I will be grateful. .

, . , , ; - ; - , -, , .


All Articles