Sep 24, 2017

Basic of artificial intelligence : Four stages of machine learning (人工知能の基本、機械学習の四段階)

Basic of artificial intelligence : Four stages of machine learning (人工知能の基本、機械学習の四段階)

Basic machine learning consists of two stages. In the first stage, let the machine learn judgment rules experientially. In the second stage, let the machine go out and experience new experiences, making judgments based solely on the learnings of the first stage.


The first stage learning is, in short, an operation of imagining a smooth curve graph (y = f (x)) from several input points (X, Y coordinates) of graph paper. If you have more points, the better imagining accuracy will be.


Furthermore, by stepping down and classifying and organizing the learning stages, learning efficiency can be raised by separating it from basic learning and applied learning. As an example, it is possible to apply learning by rotating, enlarging, reducing, inverting, dirty and blurred as basic learning of typeface recognition with fixed typeface and fixed position size. Furthermore, typeface change can be added to applied learning. If we can judge that the feature of the learning data is gentle later, we can also learn only pairs of (X coordinates) and (Y coordinates).


In the second stage, since learning has not been done, the machine head does not feel better and better for users of machines. It will be an artificial incompetent rather than an artificial intelligence because the system is fixed that everything will be decided in the first stage.


Futuristic machine learning further has a third stage and a fourth stage.


In the third stage, the dialog function of the user and the machine is incorporated in the artificial intelligence system. The machine constantly learns through dialogue with the user, and the user can feel that the head of the machine improves quickly. For the user, everyone can feel like everyone is talking to a real living human being.


Technical implementation method is that the user can instruct the machine to judge whether it is true or false with respect to the result of the second stage, the machine will accept that judgments. It is a problem how many percent of judgments of the user is acceptable. It is necessary to put in the basic principle (three principle of robot and human morality, constitution and regulation system) which can not absolutely yield to the machine. Embedding the evil ideas (communism, militarism, racial discrimination, evils, terrorism) here may be a serious thing that could lead to the demise of the liberal world or ruin of mankind .


There are various ways to raise the learning efficiency of the third stage. On the Internet, to read websites around the world, to learn photos and comments, to watch videos, to listen to the real voice. In games, shogi or go, to read the human score logs, to fight against the human, to fight against the machine.


In the future, it will be necessary to have hands and feet on the machine so that they can move independently and experience. This real experience is necessary for the development of automatic driving of cars. In the near future, robots with hands and feet for a smartphone as a head will be released as well. However, the machine field which is still inferior is a tactile sensor on a soft surface, a small actuator. Especially small and light artificial muscle. In order to put a robot that operates autonomously in human daily life, it is necessary to have a robot that is soft and does not hurt human beings.


The fourth stage is to gather each learning result from many machines spreading around the world and to integrate it into higher intelligence. And to distribute that intelligence to machines all over the world. At this stage, actions that organize miscellaneous knowledge and write textbooks will also be included.


In the current information science and technology, the first stage to the second stage are put to practical use. They are already in use in systems of world leading supercompanies. However, at the first stage, the size of the data is still necessary. The current situation of using big or deep terms such as big data or deep learning shows that the fourth stage has not reached yet. To realize the first stage with big data or deep learning method, we need a large amount of computers and mass data, and the world's biggest enterprises with both can only develop artificial intelligence systems.


Recently, a semiconductor circuit which performs the second stage judgment processing has been announced by several worldwide semiconductor manufacturers, and it is mounted on a smartphone (iPhone 8 / X etc.).


The third stage is being vigorously studied. The fourth stage will be.


Even after the basic learning like human childlen, elementary school entrance degree, for self-sustaining mobile robot, the second stage learning will be done only by both reading some textbooks and actual experience like human beings.


After the third stage and the fourth stage are implemented, a self - contained mobile robot with artificial intelligence will be equal to or better than human in terms of intellectual ability.