CC BY
3
УДК 004.94
РАЗВИТИЕ И ПРИМЕНЕНИЕ ИСКУССТВЕННОГО ИНТЕЛЛЕКТА
В СОВРЕМЕННОМ МИРЕ
Ю. А. Худоногова Научный руководитель - Л. В. Липинский Руководитель по иностранному языку - А. И. Куклина
Сибирский государственный университет науки и технологий имени академика М. Ф. Решетнева Российская Федерация, 660037, г. Красноярск, просп. им. газеты «Красноярский рабочий», 31
E-mail: j.khudonogova@gmail.com
Рассматривается развитие науки об искусственном интеллекте. Представлена сжатая историческая ретроспектива, проведен обзор трудов, сформировавших данную область науки. Помимо этого, представлен краткий обзор современных методов работы над ИИ, в частности методов машинного обучения, а также области их практического применения. Сформулирован вывод о том, на каком этапе развития находится ИИ в настоящее время.
Keywords: artificial intelligence, machine learning.
DEVELOPMENT AND APPLICATION OF ARTIFICIAL INTELLIGENCE
IN A MODERN WORLD
J. A. Khudonogova Scientific supervisor - L. V. Lipinsky Foreign language supervisor - A. I. Kuklina
Reshetnev Siberian State University of Science and Technology 31, Krasnoyarskii rabochii prospekt, Krasnoyarsk, 660037, Russian Federation E-mail: j.khudonogova@gmail.com
In this article the development of the artificial intelligence science is considered. A brief The paper considers the development of the artificial intelligence science. It also presents a brief historical retrospective and reviews the works that formed the given field of science. The paper also gives a short review of the modern AI methods (machine learning methods in particular) and the information about their application areas. Finally, it draws the conclusion of the state of modern AI development.
Keywords: artificial intelligence, machine learning.
The science of artificial intelligence is one of the most rapidly developing and advanced areas of knowledge in our time. It is also one of the most many-sided scientific areas, as it has been developing rapidly for more than half a century. As a result, the science of artificial intelligence includes many sub-fields and branches, so to answer the question of what level of development this area is in our time, it is necessary first to delve into the history of the formation of AI as a field of research, as well as highlight the main theses and definitions needed to draw conclusions.
The dawn of the theory of artificial intelligence is the second half of the thirties of the last century. The man who initiated this theory is the British mathematician Alan Turing, who in 1936 published the work entitled "On computable numbers, with an application to the Entscheidungsproblem". In this work, as well as in further ones, for example, in the "Computing
Актуальные проблемы авиации и космонавтики - 2020. Том 2
machinery and intelligence" of 1950, he first discusses the machine as a model of human consciousness. Such a view on the programming of computing devices turned out to be a breakthrough and raised a large number of questions the main one of which is naturally "is the machine capable of thinking?" This issue has become fundamental to the emerging science. In the future, attempts to answer this question put forward many scientists and researchers, thereby developing the science of AI.
However, in certain historical periods, interest to AI faded away and ideas and developments almost disappeared. Such periods were called "winters." In the history of the development of AI there were two winters: 1975-1980 and 1984-1993. The main prerequisites for the onset of winters were inflated expectations from AI, which were not justified due to the complexity of the tasks and limited resources, as well as the outflow of specialists to other fields of science.
Currently, AI is on the rise. Special interest is focused on the development of machine learning methods, since they allow solving a huge number of practical problems. Technical tools are currently capable of ensuring the implementation of these methods in practice. In particular:
• Computational capacities have increased many times and are accessible;
• A large number of databases with a huge variety of data is in the public domain;
• Industrial and information enterprises have developed enough to apply machine learning techniques to their operations.
In order to understand how the principles of AI are applied in practice, it is necessary to review the methods of machine learning. In order to delve into the review of machine learning methods, one needs to know some basic definitions.
Machine learning is an extensive subsection of artificial intelligence that studies methods for constructing algorithms capable of learning. In order to train the machine to do this, we need three main components:
Objects (use cases, situations) are a certain final set of instances some parameters of which are collected.
Objects and parameters can be of various kinds depending on the problem to be solved. For example, if we want to write an algorithm for creating contextual advertisements, we will need information about the interests of the user, the pages he frequently visits, etc. In this example, the set of users is the set of objects, and the various criteria for their activity (number of likes, activity, and attendance) are a set of parameters.
Thus, if objects are represented as rows of a table, and parameters as columns, after filling this table with data we will receive a training dataset.
For each task, the machine is expected to receive a certain response, the answer to the question that it must give after processing the input data. Since there is an explicit relationship between the input data and the response, to predict the result we also need an algorithm that can reconstruct the implicit relationship between inputs and outputs and, for any new input, predict a reasonably accurate result on the basis of the existing training dataset.
Thus, the objects, parameters and algorithm are the three main components of machine learning.
Machine learning algorithms can be classified in various ways. The most basic classification is "supervised learning", "unsupervised learning" and "reinforcement learning". Let us consider examples of each of the types [1].
The following methods can be classified as a method of machine learning, called "supervised learning", because with this method the system under test is forcibly trained using a training dataset.
• Classification - there are many objects, subdivided in some way into classes. For a certain number of objects in the dataset, class membership is known. The task of the algorithm is to classify an arbitrary object on the basis of data from the training dataset.
Usually the classification problem is solved using the following methods:
o K-nearest neighbors algorithm
o Naive Bayes Classifier
o Support vector machine (SVM) o Decision tree and others[2].
• Regression analysis is a statistical research method whose goal is to predict the value of the dependent variable with the help of some independent ones.
Solution methods: o Linear regression
o LASSO and Ridge Regression and others.
In contrast, there is "unsupervised learning", that is, a way in which the test system is spontaneously trained to perform the task without interference from the experimenter. As a rule, this is suitable only for tasks in which descriptions of the set of objects (the training set) are known, and it is required to detect internal interrelations, dependencies, regularities existing between objects. This method is typical for the following tasks:
• The task of clustering is the splitting of similar objects into disjoint sets, called clusters. A common task in the design of pattern recognition systems.
It is solved by: o K-means method o Deep trust network and others.
At first glance, the tasks described above do not seem practical, but in the modern information space they appear in almost every sphere of life. Here are some examples:
• Robotics (creating so-called intelligent robots that can learn to recognize faces, emotions, etc.)
• Recognition and classification (for example, faces and fingerprints to create passwords, or any other images to classify them)
• Sorting content for users of social networks and other platforms (for example, recommendations for videos and photos based on data about user views, as well as contextual advertising)
However, it is obvious that all of the above tasks are solved using AI in the narrow sense of the word. A machine capable of recognizing faces cannot be called an AI in the broad sense of the word. AI in the broad sense (or "strong AI") is the construction of a complete intellect, equivalent to the human one. Modeling individual things that are only accessible to the human intellect on a computer is the construction of AI in the narrow sense of the word. Currently, specialists in the field of AI are able to create only AI in the narrow sense of the word. Creating a strong AI at the moment is an impossible task and to solve it, it is necessary to invent new methods or synthesize several existing ones.
Creating a strong AI will be a breakthrough step in science and will open many exciting new opportunities to humanity. However, on the very way to solve this problem, humanity will open up many new opportunities and make many technological breakthroughs. Therefore creating a strong AI is one of the most important tasks that humanity faces nowadays.
References
1. Müller, A. Introduction to Machine Learning with Python: A Guide for Data Scientists, Williams. 2017. - 480 p.
2. Vanderplas, J. Python Data Science Handbook: Essential Tools for Working with Data, Sebastopol, O'Reilly Media. 2016. - 576 p.
© XyflOHoroBa ro. A., 2020
