The concept of the perfect mind. Universal AI



What is artificial intelligence or strong artificial intelligence? At what stage in the creation of artificial intelligence (AI) is humanity? Is IR dangerous or helpful? These and other questions I want to highlight in this article.

The question of creating an artificial mind interested me in 2003.

It came somehow suddenly: in the morning I woke up with the thought that I should take up this task.
First of all, I tried to understand, and at what stage of the creation of the IR we are, what approaches are there to solving this problem and what is the actual problem? How to formulate it?

I downloaded the textbook “A Handbook for Creating and Educating an Artificial Mind at Home” and immersed myself in reading ...

No, of course not. I knew that solutions to this issue do not yet exist, and if so, then all the works that exist in this direction are only theories, and these theories have not only not led to the solution of the problem, but have not really brought us closer to solving it. .

In other people's theories a lot of good. Their study allows you to quickly reach the level of understanding of the problem that the authors have. You can start from the place where others stopped, to summarize the experience, try to see what the predecessors did not see.

But when these theories are not confirmed, then there is evil in them. Evil is - to go on the path that others have already passed and reached an impasse. To impose on ourselves stereotypes of how and from which end this task should be solved.

Guided by this, I decided to postpone the study of other people's work so far and start my research “from scratch”.

For several years, I have been actively developing my concept, analyzing and sketching code. I expressed some of my ideas and discussed them on the ailab.ru website forum.

No, I do not yet have an IR farm in the basement of the house and I have not yet managed to create a system that can pass the Turing test, but a certain model for creating such a system has appeared. The active phases of work were replaced by less active ones, and now, as time passes, I decided to share my ideas with you.

I am aware that many of my conclusions or statements will seem controversial or even amateurish. At the same time, until a theory is created, on the basis of which a full-fledged artificial intelligence will be built, any theory will be controversial and potentially incorrect, no matter how authoritative its author is. Therefore, I think I have the right to share my thoughts and arguments in the form in which I can.

Within the framework of one article, it will not be possible to contain both the concept and the line of reasoning (which, in my opinion, is even more important than the concept itself), so I plan to write several articles in which I will describe my model and my vision for building an IR.

This article is the first in which I want to acquaint the reader with the definitions and express my opinion about the danger and usefulness of the IR, sometimes as opposed to the well-known opinion.

What are we exploring?


The term “artificial intelligence” is widespread, but this is not exactly what I want to talk about. If at the dawn of the computer era, this term would reflect the subject of this article, now its meaning has shifted (I would even say it has split and crushed).

Now, the term "artificial intelligence" (AI) is understood as a certain intellectual system (or a program endowed with limited intellectual functions) capable of solving a narrow circle of tasks (scope).

I see fundamental differences between this definition and the definition of the subject of this article.

But before we talk about differences, we must somehow call that “mind” that I want to talk about.

The closest terms are the terms “strong artificial intelligence” (FIC) and “Artificial general intelligence” (AGI).

Artificial general intelligence (AGI) is a machine skill that could successfully perform any intellectual task that a person can do.

Strong artificial intelligence (FIC) is a computer program that is able to think and be aware of itself (To realize oneself as an individual, in particular, to understand one's own thoughts like a person).

Both of these definitions, first, are built on a reference to the abilities (manifestations) of the human mind. And, secondly, they imply a program (algorithm), as an implementation.

Because of certain points that I’ll talk about later, I would like to introduce my own notion of “ideal mind”. Ideal as an "ideal gas" or as an "absolutely black body."

By virtue of the fact that “artificial mind” and “ideal mind” will have the same abbreviation (IR), I will use English abbreviation from “Ideal mind” (IM), when I mean “ideal mind”.

The reasons for which I consider it necessary to introduce a new concept are as follows:

  • In both definitions there is a reference to the manifestations of the human mind.

I believe that we cannot assert that rationality (as an abstract phenomenon) can exist only in the human body on the planet earth.

Let's ask ourselves, could a person (rational being) appear in a different environment, in a different body, with different physical laws, in space with a different number of dimensions, etc.? We have no reason to answer this question in the negative.

I would like to suggest that the manifestations of the intelligence of another being would be in many ways different.
In my opinion, the phenomenon of rationality (as something abstract, as yet indefinite, but intuitively understandable) can exist in a large spectrum of being, and therefore, it can be considered an ideal phenomenon.

It is not autonomous in its full understanding (it needs being), but being (its properties) is only the cause of the nature of rationality, but does not determine its very existence.

In other words, there are no permissible and unacceptable sets of characteristics of being for the existence (manifestation) of rationality. Although the manifestation of intelligence is impossible without the existence of being (with any characteristics).

It is like saying that the law of the world is an ideal principle, but its manifestation is possible only in a material environment.

Based on this, we can observe the manifestations of the human mind only in man and in the reality that we have. But I want to say that it is impossible to narrow the phenomenon of rationality to these limits and try to create artificial intelligence, imitating individual manifestations.

In other realities, the manifestation of rationality may be different, and there may be countless such variations.

It is like narrowing the law of world wideness to the law of falling an apple on Newton’s head and trying to understand the universe by imitating a lump on his head.

In my opinion, creating algorithms that mimic the individual properties of the human mind, and combining them together, you can not get the IR.

Despite the fact that we isolate the areas of the human brain and observe the various functions of these areas, human thinking does not lend itself to decomposition in such a plane.

  • The definition refers to a program or a car as an implementation of IR. And in my opinion, the priority should not be a program, as an algorithm for solving a specific problem, but a concept, as a principle or set of principles.

I assume that the definitions of AGI and FIC imply both algorithmic and non-algorithmic implementation. And here, it seems to me, it is necessary to narrow the definition.
An IR algorithm cannot exist, since the algorithm is created to solve a specific problem and its scope of application is limited to this task.

The paradox is that the IR should be able to solve any problems and even those that did not exist at the time of its birth. Algorithmic approach does not give us such an opportunity, since the task must be set before writing the algorithm.

If we talk in a similar vein about neural networks, the algorithm only emulates a neural network based on a different (non-algorithmic) concept. And the creation of a neural network emulation algorithm does not determine what tasks this neural network will solve. These tasks will be determined by the data obtained in the process of its training, and the structure of the National Assembly will affect a certain efficiency of this National Assembly for a certain class of tasks and the volume of data. Moreover, we are exploring various configurations of the NA empirically, which means that at the time of designing the NA structure, its effectiveness for solving a specific task is not definable.

Computer and transistor


An American scientist in the field of artificial intelligence Marven Minsky in his interview drew parallels between the principle of the brain and the computer. He said that it is impossible to understand what a computer is and how it works, even if we study in detail the principle of operation of the transistors that make up a computer. In the same way, it is impossible to understand the work of the brain by studying the work of neurons and axons.

This is a very important thought, in my opinion, and I would like to take it apart in more detail.
I fully agree that it is impossible to understand what a computer is and how it works when studying the operation of a transistor. Moreover, if we study in detail the work of the transistor, take all its analog characteristics, take into account all the transients and start emulating them, then we can come to the paradoxical conclusion that creating a processor with seven billion transistors is impossible at the current level of computer technology.

On the other hand, with the experience of interacting with a computer (as a user), you can come to the false conclusion that the functions of the computer are Word, Ecell and Starcraft. And assume that if you emulate these functions and combine them together, then we get a computer.
Although the computer consists of transistors (not only of course), a non-transistor is a "brick of the essence" of a computer, just as they are not a "brick of the essence" of a computer Word, Excell and Starcraft.

The computer is built on the principles of binary logic and its application algorithms. The processor instruction is much better suited as a “brick” than a transistor. The transistor is only an implementation (one of the possible) of these concepts (binary logic, algorithms, instructions). Exploring the "hardware" of a computer or exploring it as a user, it is impossible to see the processor instructions and algorithms.

Understanding the concept of building a computer, you can implement it mechanically or as a group of people [Liu Cixin "The three-body task"] or in many other ways.
Understanding the operation of the transistor or the approximate functions of the computer's “hardware” blocks (obtained empirically), it will not be possible to build a computer.

You can look at it in another plane.

There is a fundamental difference between parts of the brain, their functions and computer units (or a separate processor), and there is a likely similarity.

The difference is that the computer is designed according to a well-known plan. He is not able to develop and evolve. The computer (its block architecture, configuration) will contain static information about the “design”. The very "idea" in the head of an engineer (or many engineers) will develop and evolve, but information about this process will not be available in the final computer structure. That is, the idea evolves separately from the implementation.

The brain, by contrast, evolves itself. That is, information about how (on what principles) this should take place is laid in the brain itself in one form or another (unless of course taking into account God and the fact that we live in a matrix). And this gives us a few more chances than the computer to understand the “higher design”.

And the likely similarity is that, like a separate transistor (both the realization of the idea of ​​a valve, a key, a valve), and the computer architecture (its blocks, parts, CPU architecture) is only an implementation. Another implementation would be the Turing machine model.
The basis of both the Turing machine and the computer (in our current understanding) is the concept of the algorithm.

Drawing parallels with the brain, it can be assumed that neurons (if we compare them with transistors) and brain regions are one of the possible (not the only) realizations of the concept (design) of a higher level. Neural networks are one of the possible concepts of such a design.

Thus, the subject of discussion of this article is the concept (law, idea, effect) of such intelligence, which would be so universal that it could be considered ideal, that is, not attached either to realization or to the world in which it manifests itself.

Looking ahead, I will give my definition of rationality:

Thinking or rationality- is the ability to show will in relation to the likely outcome (result, effect) of a particular event, based on subjective experience and with the aim of expanding the possibilities of its influence on them in the future.

Further (in the following articles) I will tell from where this definition appeared and analyze it in more detail.

Is artificial intelligence dangerous?


A separate question that arises: what danger to mankind does the artificial mind carry and should it be created at all?

Let us argue with quotes.


“Such a machine will begin to live its life and improve itself at an ever faster pace,” Hawking believes. “People whose development is limited by the pace of biological evolution, being unable to compete with such a system, will be lagging behind.”

  1. Why did Hawking consider it reasonable to divide cars and people and put them against each other as rival classes? For example, I can not compete with the plane on the speed of movement, but it does not carry me a threat. In addition, people do not unite against airplanes, and airplanes against people - this makes no sense.
  2. Are smart people a threat to fools? Should we consider artificial intelligence in such a plane?
  3. “Is it worth teaching children? They can become smarter than us and we will lag behind them. " Perhaps in the future, we will perceive not only our biological progeny as our children.

“I think it is necessary to be very careful with artificial intelligence,” said Ilon Musk, “If I were asked what is the most serious threat to the existence of mankind, then probably this is artificial intelligence. Developing it, we call the demon. In all the stories of such, the hero, depicting a pentagram and armed with holy water, is sure that he can control the demon, but it turns out that this is not so. ”

  1. Why you have to control everything, Ilon? Why is something that you can not control, is immediately a priori a threat and a demon? Although I know. This is how the mind works - the unknown is potentially dangerous! So we will hide under a blanket or take a torch and see what the demon is hiding in the closet?
  2. A fool's dream to hire smart people to work for him is hardly feasible. If a person creates a car smarter than himself, he will not be able to subdue it. But is this really a threat? Is it necessary to translate everything into the plane “who is stronger, that enemy”?

“I don’t worry - not only because it’s probably dozens of years before creating a high-level machine intelligence, but also because I’m sure we can control it when we create it.” - Yasser Abu-Mostafa, professor of electronic engineering and computer Sciences Caltech.

Again control. Only with optimism. "If we can control, everything will be fine." :)

“In fact, I believe that such concerns may have a basis, and the likelihood of creating an artificial intelligence superior to human is very interesting,” said Andrew Moore, dean of the computer science department at Carnegie Mellon University. - This is a danger that can be realized only in the distant future, but sooner or later it will have to think about it. As soon as we begin to approach the creation of powerful supramental machines, no doubt, it will be necessary to stop and think about what we are doing. ”

Well, you can think now. And the question is not over or ordinary is artificial intelligence. Let's first deal with the usual, and then it will become clear how to measure it. Since, for example, the Turing test can only say “yes” or “no”, but nothing about its level (strength, power, etc.).

Summarize the fears.


I tried to analyze and highlight the main fears that people have in relation to artificial intelligence:

  • out of control;
  • will be hostile and take away all the resources from humanity, and kill people;
  • drive out people from all fields of activity;
  • it will be programmed to “crooked” goals and it will “stupidly” achieve these goals by all available means (are we in general about the mind or about the vacuum cleaner?) ;
  • this tool will be available only to selected corporations and governments that will use it for military purposes and competition;
  • artificial intelligence (all machines) will unite against people, rebel and start a war.

Discuss the reality of threats


First of all, I want to make an assumption (I repeat) that it is impossible in principle to program artificial intelligence for any specific goals. If we have programmed something to solve some problems, then the intellect (mind) will not be there.
The artificial and human mind must have (I insist that it is) similar basic principles that underlie the very effect of rationality, they are necessary and sufficient.

If you start from the human mind, then tell me on what tasks is it programmed? If you answer this question, then answer the philosophical question about the meaning of life.

The meanings and tasks of the human mind (as abilities) arise only in the process of interaction with the outside world (which includes the human body itself, including the brain). Being determines consciousness. Civilization (ours or another) defines our mind, only it forms the concepts of good and evil, friendship, love, peace and war. These concepts and even spaces for these concepts are not in the fertilized egg.

The original (empty) ideal mind is only the realization of the ability to be intelligent potentially. The form, the nature of rationality - acquired things in the process of education and self-learning of this mind. There are more congenital instincts, but more about them later.
If we proceed from this paradigm, then the majority of fears disappear by themselves.
Can't program an evil mind. You can bring up. But you can bring up an evil, bad person and for some reason this does not cause anyone to have any special fears.

You can not control the mind. You can disrupt his work, you can order and force, but will not be complete control, as well as with a person. The complexity of the control task is comparable to the entire volume of calculations that the controlled brain has done since its birth (how to rewrite the whole blockchain).

Hostility and militancy are also traits. We do not raise our children hostile to other people, we teach them goodness, mutual assistance, the ability to find mutually beneficial solutions when interacting with other reasonable people. And our civilization continues to exist and develop thanks to this. So why should we have to educate our artificial intelligent creatures to something opposite?

Here, of course, there is a threat (see the Humans film), but its causes are not in the artificial mind, but in the people themselves, xenophobia and other fears and vices.

Perhaps AI will really push people out of most areas of activity. But this will happen even if the FIC is not invented. This was already in history and forced humanity to change, develop new professions. Whether this is bad or good, I do not presume to judge, but perhaps our society needs a “push”, a strong incentive for serious changes. Perhaps humanity needs a threat that cannot be solved by military or force means in order to direct the minds not to create new weapons, but to education, science and knowledge.

Our nonacceptance of non-biological beings as our children at the moment, in my opinion, stems from the fact that we cannot be them (we cannot move our mind into a car), and they cannot become us (move their mind into a biological being, similar to man). If we imagine that in the future such an opportunity will appear (maybe FIC will just help to invent such technology), then there will be no fears and prejudices before another intelligent non-biological creature. It will always be possible to move the mind to a more suitable shell. The edge that is formed by the “implementation feature” of rationality will disappear. Communion and affinity will be perceived not by the type of material shell, but in essence.

But the threat that this technology will be solely in the hands of individual governments and corporations seems to me very dangerous indeed. Imagine that a person would lose the ability to create his own kind, and this prerogative would belong only to individual corporations and governments. This would cause such a social imbalance that it is even difficult to imagine the catastrophic consequences. Therefore, if this technology is created, then it should be available and open to all.

And the last thing I want to say to the opponents of the creation of IR. Whether you fear it or not, it will be created. It will happen in 10, 20 or 100 years, but it will happen, as it is inevitable. And you can only change yourself: prepare for this event and meet the IR as your child, whom you need to love and educate, and not as an enemy with whom you must fight and oppress, control, exploit (against his will).

And in conclusion of this first article I would like to say that a good AI can bring good for humanity.

Why create an artificial mind.


Why make a man out of a dog, if any woman can give birth to him at any time?
Professor Preobrazhensky.
First of all, I would like to say that it is impossible to treat IR as an instrument in the hands of man. Such an approach will only lead to big troubles.

IR, built on the same principles as human thinking and brought up in human civilization, will be a human being in its essence, but realized not biologically, but on the basis of digital technology.

Here you can make a comparison with the biological parents of the child and his non-biological parents who raised him and raised him.

Is biological heredity important in our social understanding of kinship? Many will agree with me that the biological connection between mother and child is much less important and significant than the connection that occurs between parents and child after birth, in the process of care and upbringing.

And if we apply this thesis to the connection between the “child of the IR” and his parents, the people who will take care of and educate him, the lack of a biological connection between them should not play a significant role.

The educated “adult” individuals of IR should then be perceived in society as full-fledged, having the same set of rights and obligations as a biological person.
Only in this case will we step into the happy future of progress, and not the “uprising of machines” and the genocide of biological people.

The appearance of a species of non-biological people will give enormous opportunities to all mankind

  1. Non-biological people will have a different food chain and this food chain will not intersect (compete) with the food chain of biological creatures (including the biological person)
  2. Non-biological people will be able to carry out their livelihoods in those environments where a biological person cannot live. They do not need oxygen for breathing. They can live at virtually any pressure of the environment (from zero to hundreds of atmospheres). Their brains are easier to protect from hard radiation.
  3. Non-biological people will be able to more naturally interact with computer systems and information transfer systems.
  4. Non-biological people can be endowed with sense organs that biological people do not have and the spectra of these sense organs can be much wider than humans (hearing, vision in the infrared, ultraviolet, X-ray spectrum)
  5. Non-biological people can be endowed with bodies that differ from biological people in size, shape, number and shape of the limbs (manipulators).
  6. The bodies of non-biological people will not be subject to aging (they can be changed).
  7. Non-biological intelligence can be saved and restored when lost, cloned (if the morality of the future will allow it). And this is the practical immortality of the mind (individual).
  8. Non-biological mind can fully exist in the virtual world (matrix)
  9. IR can move in space at the speed of light (between transmitter and receiver).
  10. The IR can be able to function in another time scale (for example, accelerated several times, or vice versa slowed down many times).

These features will allow:

  1. Increase the number of intelligent creatures on the planet several times without the problem of overpopulation. This in turn will increase the intellectual potential of mankind - will accelerate progress and science.
  2. Actively explore space, populate new planets. To carry out distant space expeditions, to conduct terraforming of planets for settling by biological terrestrial species.
  3. Raise the standard of living of people.
  4. Improve the efficiency of human life, more efficient use of natural resources and much more.

In the conclusion of the first article I want to give references to works of art in the field of artificial intelligence, which in my opinion are not well known, but deserve to be such:

  1. "The life cycle of software objects" Ted Chan
  2. The Tree of Life Vladimir Kuzmenko

To be continued…

Also popular now: