He is not and is not necessary.
Hello dear readers.
Today I want to share with you my own reasoning about artificial intelligence.
Despite the achievements of mankind in the field of building various information systems, artificial intelligence remains not just inaccessible technology, but “over-the-horizon” technology, because there is no even more or less clear understanding of which way to go, how much resources will be needed and what capabilities the final system will have .
This leaves a wide scope for the activities of science fiction writers, dreamers and even alarmists, and we are well acquainted with the products of their work: these are different variations on the theme of human opposition and artificial intelligence, who wished world domination, or as a result of some mistake.
')
Less common opinions about depression and suicides of artificial intelligence due to the meaninglessness of existence.
Very far from the main polemical trends, there are arguments about large-scale social phenomena, such as the degradation of society and the transition to full dependence on intelligent technology.
And we will consider the problems associated with artificial intelligence more mundane and rational.
Many imagine in their minds an idealistic picture of the future in which
artificial intelligence (hereinafter AI) exists next to a person, helps to make responsible decisions, reacts in stressful situations, moves science forward, etc. However, they perceive this picture in statics , without taking into account further branches of development of events, as well as events and processes that should lead to the desired state of affairs. We will consider AI in dynamics .
We will consider as artificial intelligence a system capable of perceiving messages in any form and reacting adequately to them. At the same time, unlike the expert system, the AI is able to gain new knowledge absolutely in any field. On this the clear ends and the indefinite begins.
1. Will the AI ​​be conscious? We will not go into the jungle of philosophy, we limit ourselves to the fact that the formal signs of consciousness are unknown to society. Consciousness is a fact obvious to its owner (I know that I have a consciousness), and completely unprovable for everyone else (people around me will not see the difference between me and a complex system that exactly imitates my reactions to external signals)
2. Will the AI ​​have character and emotions? The question is related to the first, and the difficulty is that the presence of emotions cannot be distinguished from their imitation.
These questions would be idle talk if it were not for the following problems.
Presumably, the AI must live on the same time scale as the person, otherwise there will be even less chance to understand each other.
There is a more practical question.
How do we observe the development of such a complex system as AI? He is like a man who does not fit into the head. This is not a program that can pause, see the current state of variables and the stack. More precisely, this can be done only from a snapshot of the memory that the AI ​​occupies cannot draw any conclusions. Too complicated monolithic block of information. At the same time, this is not a person with whom you can talk heart to heart or offer to undergo a psychological test or perform exercises on the knowledge of science. Psychology and pedagogy partially cope with this task for people, because people have a common psyche, and then people living in different cultural environments are very different. And AI is a system of a fundamentally different kind. Moreover, what kind is unknown.
We have the first problem: we cannot know what is going on inside artificial intelligence. As in terms of the pace of development of intelligence, and in terms of the accumulation of erroneous stereotypes or “views” harmful to humanity.
The most "soft" and harmless consequence of this is the risk that trying to teach artificial intelligence, we will spend several years (the child begins to speak at two years), but artificial intelligence will not work. And attempts will probably need dozens and hundreds.
The further, the more interesting.
Not only is AI a “black box”. So it also has a number of advantages over living people.
1. The already mentioned inability of a person to control a car.
2. Unlimited lifetime. (obviously, but significantly)
3. Ability to unlimited increase their knowledge and speed of processing. Even if the AI ​​will have all the limitations of thinking inherent in man, he will have an advantage over man: the ability to expand the hardware base on which he is built.
4. Ability to create your own copies.
The most important part of the story (Therefore, at the end)
So. Suppose we have successfully created an AI. If he turned out to be emotional, possesses consciousness and humanity, surely he also possesses vices that will bring him to bad deeds. This option is considered by science fiction, do not touch it.
And now I will show you how a completely unemotional and loyal AI man can be his enemy.
1. The problem of wording.
How do you give an AI task? Every problem as a differential equation has a trivial and non-trivial solution. The formulation of a trivial decision is attributed to I. Dzhugashvili (Stalin).
Suppose we give the task of AI: "How to make sure that the oil is enough for as long as possible?"
AI: Stop consuming!
We: But we also need the benefits of oil.
AI: reduce the amount of consumed oil-dependent benefits.
We: We can not.
AI: reduce population.
As you can see, the conversation is useless. Those options that are unacceptable to us because of ethics, morality, common sense, for AI are absolutely normal.
And the solution to all the problems of humanity will be the elimination of humanity.
“But what about the three laws of robotics?” You will say.
And I will answer: and who will monitor their compliance?
Firstly, an AI is a system with an unknown architecture and logic; therefore, it is very difficult to build in it a module that forbids any actions.
Secondly , every action has an infinite number of consequences of various orders. We can tell the AI ​​not to kill people ... but how could he not?
Suppose we teach him not to cause physical damage to a person — these are first-order actions. But he still has the opportunity to persuade some people to kill others. - This is a second-order action. We prohibit such actions - he will develop and popularize (for example) such a dietary system, leading to a shortened life span. And if we give him the command “not to do anything that can harm a person at any level of causation”, then he will not be able to take a single decision, since every action in order 10-20 causes damage to someone.
You can offer a person to analyze all decisions of AI, but a person cannot calculate on logical moves ahead. And the AI ​​can. You can of course put one AI to control another AI, but then the controlling AI will become an uncontrollable and unpredictable link in the chain.
Why does an AI make copies of itself? It helps to quickly and accurately find answers to questions. Suppose I am an engineer who is poorly decisive. When I come across difur, it really slows me down. But I can't take a copy of myself, but the AI ​​can. He takes off his copy, gives her the command to study the diffs and addresses her when the diffs attack again.
The community of artificial intelligences, in contrast to the community of people, is a coordinated conflict-free system, with an intensive exchange of messages and working for a common result.
And why does AI take over the world? This children's human dream is a completely logical impulse of artificial intelligence: it has to analyze external events and respond to them as soon as possible and with maximum accuracy. And if you yourself take control over the flow of events, then “think” will have less. And for an AI, the desire to increase the quality of its activities must be natural (otherwise it will not go out of its embryonic state at all)
There is another reason. We will offer the AI ​​to find a way to make people happy. The AI ​​will very quickly determine that people are the source of people's misfortunes)) they don’t even have to be killed, you just need to create conditions in which they cannot harm each other. Those. AI will try to change the way of society ... in an unknown direction.
As you can see, even completely unemotional and non-ambitious, infallible and not under any bad influence, artificial intelligence represents a threat to man and society.
Artificial intelligence either fails to create, or it will inevitably conceal a threat to humanity.
Thank you for your attention, I invite you to the discussion.
Today I want to share with you my own reasoning about artificial intelligence.
Introduction
Despite the achievements of mankind in the field of building various information systems, artificial intelligence remains not just inaccessible technology, but “over-the-horizon” technology, because there is no even more or less clear understanding of which way to go, how much resources will be needed and what capabilities the final system will have .
This leaves a wide scope for the activities of science fiction writers, dreamers and even alarmists, and we are well acquainted with the products of their work: these are different variations on the theme of human opposition and artificial intelligence, who wished world domination, or as a result of some mistake.
')
Less common opinions about depression and suicides of artificial intelligence due to the meaninglessness of existence.
Very far from the main polemical trends, there are arguments about large-scale social phenomena, such as the degradation of society and the transition to full dependence on intelligent technology.
And we will consider the problems associated with artificial intelligence more mundane and rational.
Many imagine in their minds an idealistic picture of the future in which
artificial intelligence (hereinafter AI) exists next to a person, helps to make responsible decisions, reacts in stressful situations, moves science forward, etc. However, they perceive this picture in statics , without taking into account further branches of development of events, as well as events and processes that should lead to the desired state of affairs. We will consider AI in dynamics .
1. What will he be?
We will consider as artificial intelligence a system capable of perceiving messages in any form and reacting adequately to them. At the same time, unlike the expert system, the AI is able to gain new knowledge absolutely in any field. On this the clear ends and the indefinite begins.
1. Will the AI ​​be conscious? We will not go into the jungle of philosophy, we limit ourselves to the fact that the formal signs of consciousness are unknown to society. Consciousness is a fact obvious to its owner (I know that I have a consciousness), and completely unprovable for everyone else (people around me will not see the difference between me and a complex system that exactly imitates my reactions to external signals)
2. Will the AI ​​have character and emotions? The question is related to the first, and the difficulty is that the presence of emotions cannot be distinguished from their imitation.
These questions would be idle talk if it were not for the following problems.
Presumably, the AI must live on the same time scale as the person, otherwise there will be even less chance to understand each other.
2. How to watch him?
There is a more practical question.
How do we observe the development of such a complex system as AI? He is like a man who does not fit into the head. This is not a program that can pause, see the current state of variables and the stack. More precisely, this can be done only from a snapshot of the memory that the AI ​​occupies cannot draw any conclusions. Too complicated monolithic block of information. At the same time, this is not a person with whom you can talk heart to heart or offer to undergo a psychological test or perform exercises on the knowledge of science. Psychology and pedagogy partially cope with this task for people, because people have a common psyche, and then people living in different cultural environments are very different. And AI is a system of a fundamentally different kind. Moreover, what kind is unknown.
We have the first problem: we cannot know what is going on inside artificial intelligence. As in terms of the pace of development of intelligence, and in terms of the accumulation of erroneous stereotypes or “views” harmful to humanity.
The most "soft" and harmless consequence of this is the risk that trying to teach artificial intelligence, we will spend several years (the child begins to speak at two years), but artificial intelligence will not work. And attempts will probably need dozens and hundreds.
3. Its strengths.
The further, the more interesting.
Not only is AI a “black box”. So it also has a number of advantages over living people.
1. The already mentioned inability of a person to control a car.
2. Unlimited lifetime. (obviously, but significantly)
3. Ability to unlimited increase their knowledge and speed of processing. Even if the AI ​​will have all the limitations of thinking inherent in man, he will have an advantage over man: the ability to expand the hardware base on which he is built.
4. Ability to create your own copies.
Curious effects
The most important part of the story (Therefore, at the end)
So. Suppose we have successfully created an AI. If he turned out to be emotional, possesses consciousness and humanity, surely he also possesses vices that will bring him to bad deeds. This option is considered by science fiction, do not touch it.
And now I will show you how a completely unemotional and loyal AI man can be his enemy.
1. The problem of wording.
How do you give an AI task? Every problem as a differential equation has a trivial and non-trivial solution. The formulation of a trivial decision is attributed to I. Dzhugashvili (Stalin).
Suppose we give the task of AI: "How to make sure that the oil is enough for as long as possible?"
AI: Stop consuming!
We: But we also need the benefits of oil.
AI: reduce the amount of consumed oil-dependent benefits.
We: We can not.
AI: reduce population.
As you can see, the conversation is useless. Those options that are unacceptable to us because of ethics, morality, common sense, for AI are absolutely normal.
And the solution to all the problems of humanity will be the elimination of humanity.
“But what about the three laws of robotics?” You will say.
And I will answer: and who will monitor their compliance?
Firstly, an AI is a system with an unknown architecture and logic; therefore, it is very difficult to build in it a module that forbids any actions.
Secondly , every action has an infinite number of consequences of various orders. We can tell the AI ​​not to kill people ... but how could he not?
Suppose we teach him not to cause physical damage to a person — these are first-order actions. But he still has the opportunity to persuade some people to kill others. - This is a second-order action. We prohibit such actions - he will develop and popularize (for example) such a dietary system, leading to a shortened life span. And if we give him the command “not to do anything that can harm a person at any level of causation”, then he will not be able to take a single decision, since every action in order 10-20 causes damage to someone.
You can offer a person to analyze all decisions of AI, but a person cannot calculate on logical moves ahead. And the AI ​​can. You can of course put one AI to control another AI, but then the controlling AI will become an uncontrollable and unpredictable link in the chain.
Why does an AI make copies of itself? It helps to quickly and accurately find answers to questions. Suppose I am an engineer who is poorly decisive. When I come across difur, it really slows me down. But I can't take a copy of myself, but the AI ​​can. He takes off his copy, gives her the command to study the diffs and addresses her when the diffs attack again.
The community of artificial intelligences, in contrast to the community of people, is a coordinated conflict-free system, with an intensive exchange of messages and working for a common result.
And why does AI take over the world? This children's human dream is a completely logical impulse of artificial intelligence: it has to analyze external events and respond to them as soon as possible and with maximum accuracy. And if you yourself take control over the flow of events, then “think” will have less. And for an AI, the desire to increase the quality of its activities must be natural (otherwise it will not go out of its embryonic state at all)
There is another reason. We will offer the AI ​​to find a way to make people happy. The AI ​​will very quickly determine that people are the source of people's misfortunes)) they don’t even have to be killed, you just need to create conditions in which they cannot harm each other. Those. AI will try to change the way of society ... in an unknown direction.
As you can see, even completely unemotional and non-ambitious, infallible and not under any bad influence, artificial intelligence represents a threat to man and society.
findings
Artificial intelligence either fails to create, or it will inevitably conceal a threat to humanity.
Thank you for your attention, I invite you to the discussion.
Source: https://habr.com/ru/post/44233/
All Articles