Dangerous mind safe intelligence
One day, artificial intelligence will become smarter than man. When can this be dangerous? To study the situation, one should more accurately define the concept of AI, for example, on the basis of comparing it with natural intelligence.
Can an intellectual not be smart? Can an intelligent person be a non-intellectual?
The approach of the era of artificial intelligence makes one ponder over these questions, because one more question remains - “How to recognize it?” - how to find out if it is already close or has already come - artificial intelligence, which is dangerous for humanity? So you can be late. Probably, the first element of the danger of AI will be its independence, uncontrollability in decision making.
It is known that the study of the problem of the danger of artificial intelligence for humanity is already funded [1]. OpenAI Institute is studying the prospects for the development of artificial intelligence. What can we say about the current development of AI systems?
Today, there are two types of danger of using AI:
- software errors;
- self-activity of artificial intelligence to the detriment of man.
The danger of the first kind is characteristic of any software, therefore we will not talk about it here.
Artificial intelligence can become dangerous when the computer becomes smart. Further, I will try to substantiate this statement.
Let's clarify these concepts and think about measurement. It would be nice to determine what degree of mind can be acceptable for a computer, and what degree will already be dangerous for people. The inevitable convergence of the intellectual capabilities of a computer and a person forces us to learn how to accurately measure these properties.
What do the well-known IQ tests measure [2]? Can I apply them to my computer? Erudite contests that measure - knowledge or mind?
Mind and intellect - these concepts are connected with the ability to solve problems. The more difficult task a person can solve, the smarter he is. It seems to be so. So it is necessary to measure the difficulty of the task.
An intellectual (not to be confused with an intellectual!) Is the person whose profession is to solve problems of every kind, both in the natural sciences and in the humanities. Perhaps such can be considered scientists, engineers, writers. The product of an intellectual's labor is some new theory (scientist) or the application of existing theories (engineer), the result of their work is most often documented. The writer also creates new ideas or applies existing ones, but presents them in a form that does not require special efforts for mastering, since books by a good writer are fun to read.
The intellectual has special knowledge in the field of his profession and skills that allow him to solve special problems and create new knowledge and new theories.
Obviously, there are much more intelligent people than intellectuals, since most people are not engaged in mental work.
Can an intellectual be stupid? It would probably be more correct to say - how smart an intellectual should be or how intelligent should an intelligent person be to be considered as such? How to measure the mind? How to measure intelligence?
Take for example the profession of a programmer. They can be considered intellectuals because they create an information product. Can you say that all programmers are smart? This question requires additional analysis.
Maybe the mind should be measured by the length of the chain of reasoning that a person can build to solve a problem, or would it be more correct to measure the degree of closeness of a particular solution to the optimal solution?
And the required amount of knowledge? And the complexity of this knowledge? How to measure it?
For example, if the mind can be measured by the length of a chain of statements, then intelligence can be measured by the amount of knowledge that must be applied to solve the problem.
Apparently, for a complete comparison of different minds (or intellects) of people or computers, several different measures will be needed, for example, the speed of solving a problem, the ability to solve "in the mind" or on paper. How to measure the wit of the solution? Who is more intelligent - “smart” or “witty”?
It is also necessary to distinguish between mental and logical abilities to solve problems. If a person has the ability to solve complex problems in his mind, does he mean that he is smarter than a person who solves such problems with paper and a pen? As a rule, people who solve problems in their mind seem more intelligent because they quickly give a solution. It can even be assumed that the smartest among people, such as Newton, Einstein, Lomonosov and many other great people, possessed this ability to solve complex problems in their minds, but still this is purely physiological ability, like the amount of computer's RAM.
It is considered, for example, that only two percent of people have the ability “in mind” to solve Einstein’s logical problem of five houses and their inhabitants. But how many people can solve it with paper and pen - 10%, 50%?
You might think that the computer has the intellect of at least 2% of the smartest people, since there are programs that solve the Einstein problem. However, it is impossible to agree with this, since in this case the computer acts as a human-machine system, in which the most intellectual part of the work - the conceptualization and the formalization of the task - is performed by the person.
I propose such definitions.
Intellect is the ability to solve problems on the basis of acquired knowledge of the relevant domain of the problem. It also includes the ability to learn - to acquire knowledge, specially prepared for the student to master.
Mind - the ability to solve problems in the absence of relevant subject area. Includes the ability to independently create knowledge.
It can be said even more simply - an intellectual applies knowledge, an intelligent one creates knowledge.
Of course, it is necessary to understand that such a division is conditional, since each person applies some kind of knowledge in a ready-made form, and creates some of his own. In this regard, people differ only in the degree of intelligence and intelligence.
When can we say that the computer has become intelligent? If you make a program that solves all possible tests for measuring IQ, can you say that this computer (program) is smarter than a person?
IQ tests are designed for the fact that a person does not have ready knowledge to solve them. So we can assume that these tests are designed to measure (compare) the mind, not intelligence.
However, a computer that solves such tests will not benefit from this. It is necessary that he solve any problem formulated in a natural language.
If you make full-fledged expert systems (the first requirement for them is an understanding of the tasks formulated in natural language) for all subject areas, instead of the traditional application systems that are now used, this will be the first step to building safe artificial intelligence.
Existing information systems are human-machine, since they are useless without the user's intelligence and / or computer expert.
Today's application systems require user training - the acquisition of special knowledge. For example, an accountant needs to be trained to use the 1C system, but this will not be enough, because, in the process of working with an accountant with this system (as with any other application system), there is always a need to consult an expert.
An application system can be called an expert if the specialist (for example, an accountant) does not need special training to work with a computer application system, and does not require an appeal to a consultant to resolve non-standard situations. For these types of computer systems, the term “safe (limited) artificial intelligence” can be confidently applied.
Even if the expert system completely replaces the human accountant, it will still remain safe (if not considered a risk unemployment among accountants). Means of verification of decisions made in accounting have long been developed, so errors in the accounting software will be detected in time.
They ate to raise the level of computer intelligence even higher, for example, to endow it with the ability to independently create application systems, this will not be dangerous yet, although it will be a significant approximation to a dangerous level.
In what case can an intelligent accounting system become dangerous for a person — for example, to give out incorrect information?
The danger will appear if the system of artificial intelligence contains elements of personality, for example, the emergence of self-interest, separate from human. To eliminate such a development of events, it may be necessary to prohibit the creation of intelligent systems capable of evolution.
There is another danger - the AI ​​may contain logical errors. Since the tasks solved by him can be very complex, and the list of tasks is not known in advance, it is necessary to provide special measures for verifying the solution obtained. For example, for an ES, the standard element of the technology is a subsystem for explaining and / or justifying the solution obtained.
Is it possible the situation in which the method of solving the problem, built by artificial intelligence, will be so difficult that people can not understand it? Today's intellectual systems are based on "human" knowledge. When the AI ​​begins to “acquire” the knowledge independently, we will not know “what he thinks about.” This is a real danger. Perhaps such software, when the possibility of their creation becomes real, will have to be banned.
S. Lem has such a story about the future of the Internet - a worldwide network, based on the combination of all the incoming information, she began to predict events. It can be assumed that the next step in the development of such a network would be the management of world events.
Apparently, it is necessary to develop all sorts of ways to control artificial intelligence systems, of which the most obvious are special software tools for checking the correctness of the solution automatically, without human intervention.
This will be a trust in artificial intelligence (or faith?) - “I don’t understand how he does it, but there’s a program that tested it.” This has been happening for a long time in technology - we cannot check all the mechanisms and devices on the basis of our own sense organs, trusting special technical means of diagnostics.
Links
1. www.openai.com
2. www.en.wikipedia.org/wiki/Hans_Eysenck
Can an intellectual not be smart? Can an intelligent person be a non-intellectual?
The approach of the era of artificial intelligence makes one ponder over these questions, because one more question remains - “How to recognize it?” - how to find out if it is already close or has already come - artificial intelligence, which is dangerous for humanity? So you can be late. Probably, the first element of the danger of AI will be its independence, uncontrollability in decision making.
It is known that the study of the problem of the danger of artificial intelligence for humanity is already funded [1]. OpenAI Institute is studying the prospects for the development of artificial intelligence. What can we say about the current development of AI systems?
Today, there are two types of danger of using AI:
- software errors;
- self-activity of artificial intelligence to the detriment of man.
The danger of the first kind is characteristic of any software, therefore we will not talk about it here.
Artificial intelligence can become dangerous when the computer becomes smart. Further, I will try to substantiate this statement.
Let's clarify these concepts and think about measurement. It would be nice to determine what degree of mind can be acceptable for a computer, and what degree will already be dangerous for people. The inevitable convergence of the intellectual capabilities of a computer and a person forces us to learn how to accurately measure these properties.
What do the well-known IQ tests measure [2]? Can I apply them to my computer? Erudite contests that measure - knowledge or mind?
Mind and intellect - these concepts are connected with the ability to solve problems. The more difficult task a person can solve, the smarter he is. It seems to be so. So it is necessary to measure the difficulty of the task.
An intellectual (not to be confused with an intellectual!) Is the person whose profession is to solve problems of every kind, both in the natural sciences and in the humanities. Perhaps such can be considered scientists, engineers, writers. The product of an intellectual's labor is some new theory (scientist) or the application of existing theories (engineer), the result of their work is most often documented. The writer also creates new ideas or applies existing ones, but presents them in a form that does not require special efforts for mastering, since books by a good writer are fun to read.
The intellectual has special knowledge in the field of his profession and skills that allow him to solve special problems and create new knowledge and new theories.
Obviously, there are much more intelligent people than intellectuals, since most people are not engaged in mental work.
Can an intellectual be stupid? It would probably be more correct to say - how smart an intellectual should be or how intelligent should an intelligent person be to be considered as such? How to measure the mind? How to measure intelligence?
Take for example the profession of a programmer. They can be considered intellectuals because they create an information product. Can you say that all programmers are smart? This question requires additional analysis.
Maybe the mind should be measured by the length of the chain of reasoning that a person can build to solve a problem, or would it be more correct to measure the degree of closeness of a particular solution to the optimal solution?
And the required amount of knowledge? And the complexity of this knowledge? How to measure it?
For example, if the mind can be measured by the length of a chain of statements, then intelligence can be measured by the amount of knowledge that must be applied to solve the problem.
Apparently, for a complete comparison of different minds (or intellects) of people or computers, several different measures will be needed, for example, the speed of solving a problem, the ability to solve "in the mind" or on paper. How to measure the wit of the solution? Who is more intelligent - “smart” or “witty”?
It is also necessary to distinguish between mental and logical abilities to solve problems. If a person has the ability to solve complex problems in his mind, does he mean that he is smarter than a person who solves such problems with paper and a pen? As a rule, people who solve problems in their mind seem more intelligent because they quickly give a solution. It can even be assumed that the smartest among people, such as Newton, Einstein, Lomonosov and many other great people, possessed this ability to solve complex problems in their minds, but still this is purely physiological ability, like the amount of computer's RAM.
It is considered, for example, that only two percent of people have the ability “in mind” to solve Einstein’s logical problem of five houses and their inhabitants. But how many people can solve it with paper and pen - 10%, 50%?
You might think that the computer has the intellect of at least 2% of the smartest people, since there are programs that solve the Einstein problem. However, it is impossible to agree with this, since in this case the computer acts as a human-machine system, in which the most intellectual part of the work - the conceptualization and the formalization of the task - is performed by the person.
I propose such definitions.
Intellect is the ability to solve problems on the basis of acquired knowledge of the relevant domain of the problem. It also includes the ability to learn - to acquire knowledge, specially prepared for the student to master.
Mind - the ability to solve problems in the absence of relevant subject area. Includes the ability to independently create knowledge.
It can be said even more simply - an intellectual applies knowledge, an intelligent one creates knowledge.
Of course, it is necessary to understand that such a division is conditional, since each person applies some kind of knowledge in a ready-made form, and creates some of his own. In this regard, people differ only in the degree of intelligence and intelligence.
When can we say that the computer has become intelligent? If you make a program that solves all possible tests for measuring IQ, can you say that this computer (program) is smarter than a person?
IQ tests are designed for the fact that a person does not have ready knowledge to solve them. So we can assume that these tests are designed to measure (compare) the mind, not intelligence.
However, a computer that solves such tests will not benefit from this. It is necessary that he solve any problem formulated in a natural language.
If you make full-fledged expert systems (the first requirement for them is an understanding of the tasks formulated in natural language) for all subject areas, instead of the traditional application systems that are now used, this will be the first step to building safe artificial intelligence.
Existing information systems are human-machine, since they are useless without the user's intelligence and / or computer expert.
Today's application systems require user training - the acquisition of special knowledge. For example, an accountant needs to be trained to use the 1C system, but this will not be enough, because, in the process of working with an accountant with this system (as with any other application system), there is always a need to consult an expert.
An application system can be called an expert if the specialist (for example, an accountant) does not need special training to work with a computer application system, and does not require an appeal to a consultant to resolve non-standard situations. For these types of computer systems, the term “safe (limited) artificial intelligence” can be confidently applied.
Even if the expert system completely replaces the human accountant, it will still remain safe (if not considered a risk unemployment among accountants). Means of verification of decisions made in accounting have long been developed, so errors in the accounting software will be detected in time.
They ate to raise the level of computer intelligence even higher, for example, to endow it with the ability to independently create application systems, this will not be dangerous yet, although it will be a significant approximation to a dangerous level.
In what case can an intelligent accounting system become dangerous for a person — for example, to give out incorrect information?
The danger will appear if the system of artificial intelligence contains elements of personality, for example, the emergence of self-interest, separate from human. To eliminate such a development of events, it may be necessary to prohibit the creation of intelligent systems capable of evolution.
There is another danger - the AI ​​may contain logical errors. Since the tasks solved by him can be very complex, and the list of tasks is not known in advance, it is necessary to provide special measures for verifying the solution obtained. For example, for an ES, the standard element of the technology is a subsystem for explaining and / or justifying the solution obtained.
Is it possible the situation in which the method of solving the problem, built by artificial intelligence, will be so difficult that people can not understand it? Today's intellectual systems are based on "human" knowledge. When the AI ​​begins to “acquire” the knowledge independently, we will not know “what he thinks about.” This is a real danger. Perhaps such software, when the possibility of their creation becomes real, will have to be banned.
S. Lem has such a story about the future of the Internet - a worldwide network, based on the combination of all the incoming information, she began to predict events. It can be assumed that the next step in the development of such a network would be the management of world events.
Apparently, it is necessary to develop all sorts of ways to control artificial intelligence systems, of which the most obvious are special software tools for checking the correctness of the solution automatically, without human intervention.
This will be a trust in artificial intelligence (or faith?) - “I don’t understand how he does it, but there’s a program that tested it.” This has been happening for a long time in technology - we cannot check all the mechanisms and devices on the basis of our own sense organs, trusting special technical means of diagnostics.
Links
1. www.openai.com
2. www.en.wikipedia.org/wiki/Hans_Eysenck
')
Source: https://habr.com/ru/post/372047/
All Articles