Распознавание образов в эйдетическом искусственном интеллекте

"[The neural network needs] 300 million images of cats to learn to understand that there is a cat, a cow or a dog. But intelligence is not based on big data. (Big data). On the contrary, it is based on" small data. "If you can watch on a cat, extract the abstract principles of a “cat” in the way children do, and then always know what a cat is - this is the intelligence “

Pascal Kaufmann, Swiss scientist, neurologist, founder of the company“ Starmaind ”.

In fact, why do children learn to recognize cats faster and easier than neural networks do? In other words, what is the schematic diagram of the human pattern recognition mechanism?

Let's take a deeper look at this process!

But first, simplify the task to better understand the principle of operation.

Take an apple as a subject for recognition. This item is simpler than a cat, it is easier to recognize, which means that it is easier to understand the recognition mechanism.

I will describe the mechanism of the brain in epistemological terms, i.e. in the key of the philosophical theory of knowledge. In other words - how do we see it in our own mind, our “inner sight”, for ourselves? I do this intentionally, because only by creating a living, i.e. native description of the thinking process; in the future, you can use the basis for algorithmization. Including for the creation of mathematical formulas, research in the field of neurophysiology, the creation of an artificial organism (strong artificial intelligence).

To solve this problem, we will conduct a mental experiment. Imagine yourself in the place of a one-year-old child, who recently learned what an apple is and is not yet very good at distinguishing between them. Therefore, its recognition process is too slow, which is very useful for science: it can be traced.

In general, pedagogy, especially the pedagogy of the first years of life, is no end for the study of learning algorithms, but for some reason, computer scientists do not look into this area, but go straight to mathematical formulas.

Let me ask myself a question: when does a person master complex mathematics, algebra? That's right - already at school, i.e. from 10-12 years of age. Wait, how can we teach the car to think, if we don’t understand how children learn to think at the very beginning of their lives and their learning? We do not understand how learning is possible in principle? And feel free to ask our children about it!

And if you just remember how to study yourself?

- I remember, I imagine myself small. I am 1 year old. Mom hands me an apple.



Here I look at the apple. I do not know what to do with it. It is round [1st pattern, shape] *.

Round can be a toy ball.



It becomes interesting to me, and I take an apple in my hand. It is colored evenly, in red-yellow color [2nd pattern, color], and the ball, usually painted in different colors. So, most likely, this is not a ball.



I drop it on the floor. It does not jump. So, definitely not the ball.


I smell it - it smells sweetly [3rd pattern, smell]! So maybe this is food! Oh, I already know a lot of edible tasty round sweet items: this is an orange, and an apple, and a tangerine, and a peach ... I even know what they are called - these are fruits!



The circle of searches is narrowed. The extra eidos patterns are rejected, I focus on the recognition of fruits! I lightly press this round fruit.


It is not soft [4th pattern, sensation], which means that the mandarin and peach fall away.


The orange has a hard bumpy crust, I remember that, and orange is also orange [the second pattern, color again], which means that the orange also disappears.



What is left?

An Apple!

Just in case, I carefully pick this fruit and feel the incomparable taste [of the 5th pattern, taste] of apple juice.

Exactly, this is an apple! The recognition task is solved!



The mental experiment is over.

As you have noticed, when recognizing an object, the child used vision, which can operate with patterns, (or more precisely, eidos) of the form; Smell (pattern of smell), touch (eidos of sensations), etc. - all 6 sense organs and associated eidos: concepts-ideas about the world. This feature of our thinking was noticed by the ancients and is well described by Aristotle in his “Categories”, remember: categories of place, time, category of space ... From the category of state, the category of form (1st pattern), from the category - taste, smell, colors, etc. Those. Aristotle considered the highest categories, which are the top of the pyramid. And at the bottom of the pyramid, we can easily find household categories. This is a category hierarchy.

But wait, isn't that the very principle of categorical description used by scientists in research?

Take, for example, the requirements for the content of a thesis. They sound like this:

• introduction (category of reason, answers the question "why?");
• description of the subject of the research (category of the subject, answers the question "what?");
• justification of relevance (time category, answers the question "when?");
• references to sources (category of space, answers the question “from where?”)
• conclusions (quality category, considers the value of this research for science), etc. etc.

This begs the question: there is no fundamental difference between the description of an apple made by a child and the description of a scientific invention made by an experienced scientist?

In terms of eidetics, yes!

In this is possible and hidden the secret of human learning ability. We do not increase the volume of knowledge (Big Data, Big Data). We make the processing of quantitative knowledge into qualitative in a special way. This is the so-called law of dialectics, derived by Hegel.

And back, when you need to expand the description from a brief idea, we take its basis, the pattern and saturate it with examples.

Look - the main vocabulary is formed by a child up to 7 years. 90% of the child’s knowledge is absorbed by up to 7 years. So in the first 7 years, the brain is formed and develops. Further, only a minor improvement. Remember the well-known words of the Jesuits: “Let me raise a child up to 7 years old, and then do with it what you want!”

With age, the neural connections only slightly improve and become more complicated, but not the number of neurons. This fact created a trap for the creators of computer neural networks. They decided that the complexity and versatility of thinking is ensured by neural connections within which intelligence arises due to the self-organization of neurons.

Therefore, the ideologists of neural networks are trying to create more and more deep, more and more complex neural networks so that self-learning (Deep Learning) is “born” in them.

This theory believes that there is really no universal intelligence. Intellect is simply the immanent, inherent property of complexly organized matter. Just as when looking at the "behavior" of a typhoon, you might think that he has some kind of demonic mind.

In this sense, any system is inherent intelligence. With the internal complication of the system, its external manifestations become more complex, which we, as rational people, are more accustomed to interpret as rational behavior.

May be so. But let's think about what happened before, the chicken or the egg?

Categories appeared after Aristotle, or did Aristotle generalize already existing laws of language, logic, human intellect?

Perhaps, the growth of connections between neurons is only a consequence of the growth in the number of patterns, eidos, this semiotic tree of knowledge?

Or does the one without the other not exist, but interact organically?

Can a pattern of taste arise for a car if it does not have a body of taste?

Can a person understand how to train a machine if he has no processor in his brain?

Can a neural connection be formed if there is nothing to connect? If instead of a bizarre mixture of human feelings, animal instincts, social education and inner “will” - just a stream of photos from the social network?

The answers to these questions, I hope, will be given by the new science - computer eidetic, integral science of eidetic artificial intelligence.

When the exact sciences come to a standstill, an integral method of philosophy that can become a bridge between computer science, mathematics and neurophysiology comes to the rescue.

---

* Here and hereafter: the pattern, it is the image, it is the eidos - an element of the description of a single thought, model, representation, definition. In psychology, they are called prints, imprints of impressions. In everyday understanding, we can consider eidos ideas. The term eidos is borrowed from Plato, who considered ideas material. If we agree that thinking patterns or images can at least be drawn, or described, then in this sense, yes, they are really material. From Plato's dialectics, Aristotle's logic and his theory of categories grew. And without logic, as it is known, mathematics would not have developed, and later physics, and indeed all modern science and our modern world, as we know it. From the grain, from the eidos of philosophy, as a new form of knowledge, in contrast to the mythological way of the world and thinking, a new, science-driven world was born.