1. The main directions of research in the field of artificial intelligence. Classification of intelligent systems

Artificial intelligence is one of the directions of computer science, the purpose of which is to develop hardware and software tools that allow a non-programmer user to set and solve problems that are traditionally considered to be intellectual, communicating with computers on a limited subset of natural language.

The main task of AI is the development of paradigms and algorithms that provide computer solutions to cognitive tasks - tasks that require mental perception and processing of external information. An example of such an information processing task is vision.

Artificial Intelligence (AI) as a science has existed for about half a century. The first intellectual system is the program “Logic-Theorist”, intended for proving theorems and calculating statements. Her work was first demonstrated on August 9, 1956. Such famous scientists as A. Newell, A. Turing, C. Shannon, J. Low, G. Simon, and others participated in the creation of the program. Since then, in the field of AI developed a great variety of computer systems, which are called intellectual.

Consider briefly the main directions of development of intelligent systems.

Development of intelligent information systems or systems based on knowledge. The main purpose of building such systems is to identify, study and apply the knowledge of highly qualified experts to solve complex problems arising in practice. When building systems based on knowledge (POPs), knowledge gained by experts in the form of specific rules for solving certain problems is used. This direction aims to imitate the human art of analyzing unstructured and semi-structured problems. In this area of ​​research, the development of models for the representation, extraction and structuring of knowledge is being carried out, and the problems of creating knowledge bases (KBs) constituting the core of POPs are being studied. A special case of POPs are expert systems (ES).

Development of natural language interfaces and machine translation. Machine translation systems are built on the basis of the BR about a specific subject area of ​​the text and complex models that provide translation according to the scheme: the original source language is the language of meaning is the language of translation. It uses a structural-logical approach, which includes the sequential analysis and synthesis of natural language messages. In addition, they carry out an associative search for similar text fragments and their translations in special databases (DB). This direction also covers the research of methods and the development of systems that ensure the implementation of the process of human communication with a computer in a natural language (the so-called NL-communication systems).

Generation and speech recognition. Speech communication systems are created in order to increase the speed of information input into computers, voice control, etc. In such systems, text is understood as phoneme text, i.e. how to pronounce.

Visual processing. In this scientific direction, tasks of processing, analysis and synthesis of images are solved. The task of image processing is associated with the transformation of graphic images, the result of which are new images. In the analysis task, the original images are converted into data of another type, for example, text descriptions. In the synthesis of images, an image construction algorithm is input to the system, and the output data are graphic objects (computer graphics systems).

Training and self-study. This current area of ​​AI includes models, methods and algorithms that focus on the automatic accumulation and generation of knowledge using data analysis and synthesis procedures. This area includes the recently appeared data mining systems and patterns of searching patterns in computer databases (Knowledge Discovery).

Pattern recognition. This is one of the earliest directions of AI, in which the recognition of objects is carried out on the basis of the use of a special mathematical apparatus that provides for the assignment of objects to classes, and classes are described by sets of certain values ​​of attributes.

Games and machine creativity. Computer creativity covers composing computer music, poetry, intelligent systems for the invention of new objects. Creating intelligent computer games is one of the most developed commercial areas in the field of software development. In addition, computer games provide a powerful arsenal of various tools used for learning.

Software systems AI. Tools for developing intelligent systems include:

- special programming languages ​​focused on processing symbol information (LISP, SMALLTALK, REFAL);

- logic programming languages ​​(PROLOG), knowledge representation languages ​​(OPS 5, KRL, FRL);

- integrated software environments containing tools for creating AI systems (KE, ARTS, GURU, G2);

- shells of expert systems that allow you to create application ES, without resorting to programming (BUILD, EMYCIN, EXSYS Professional, EXPERT).

New computer architectures. This direction is connected with the creation of computers of non-von Neumann architecture oriented on the processing of symbolic information.

Intellectual robots. Creating intelligent robots is the ultimate goal of robotics. At present, programmable manipulators with a rigid control circuit, called the first generation robots, are mainly used. Despite the obvious success of individual developments, the era of intelligent autonomous robots has not yet come. The main constraints in the development of autonomous robots are unsolved problems in the field of knowledge interpretation, computer vision, adequate storage and processing of three-dimensional visual information.

Classification signs of artificial intelligence systems

The Intelligent Information System (IIS) is based on the concept of using a knowledge base to generate algorithms for solving applied tasks of various classes depending on the specific information needs of users.

Figure 1.1 shows the classification of IIS, the signs of which are the following intellectual functions:

• communication skills - the way the end user interacts with the system;

• solving complex, poorly formalized tasks that require the construction of the original solution algorithm, depending on the specific situation, characterized by uncertainty and dynamism of the initial data and knowledge;

• ability to self-study - the ability of the system to automatically extract knowledge from accumulated experience and apply it to solve problems;

• adaptability - the ability of the system to develop in accordance with objective changes in the field of knowledge.

Figure 1.1 - Classification of intelligent information systems.

Each of the listed signs conditionally corresponds to its own IIS class. Different systems may have one or more signs of intelligence with varying degrees of manifestation.

The use of AI to enhance the communicative abilities of information systems has led to the emergence of systems with an intelligent interface, among which are the following types.

Intelligent databases . They allow, in contrast to traditional databases, to provide a selection of the necessary information, which is not present explicitly, but rather derived from the set of stored data.

Natural language interface. It is used to access intelligent databases, contextual search of documentary textual information, voice input of commands in control systems, machine translation from foreign languages. To implement the NL interface, it is necessary to solve the problems of morphological, syntactic and semantic analysis, as well as the problem of synthesizing utterances in natural language. In morphological analysis, recognition and verification of the spelling of words are carried out. Syntactic control involves decomposing input messages into individual components, checking compliance with the grammatical rules of the language and, if necessary and possible, eliminating errors. Semantic analysis provides the meaning of the message. Synthesis of statements - the task, inverse analysis - is to convert information into a representation in natural language.

Hypertext systems . Used to implement keyword searches in text data databases. In intelligent hypertext systems, the search engine first works with the keyword knowledge base, and then with the text itself. Similarly, the search for multimedia information, including, in addition to text, graphic information, audio and video images.

Context help systems . They belong to the class of knowledge dissemination systems. Such systems are, as a rule, documentation supplements. Contextual help systems are a special case of hypertext and EYa-systems. In them, the user describes the problem, and the system specifies it and searches for recommendations related to the situation.

Cognitive graphics systems . Focused on communication with the user of IIS through graphic images that are generated in accordance with changes in the parameters of the simulated or observed processes. Cognitive graphics make it possible to visually present a variety of parameters characterizing the phenomenon being studied, frees the user from analyzing trivial situations, contributes to the rapid development of software and increase the competitiveness of the developed IMS. The use of cognitive graphics is relevant in monitoring and operational management systems, in training and training systems, and operational decision-making systems operating in real time.

Self-learning intellectual systems are based on the methods of automatic classification of situations from real practice, or on methods of teaching by examples. Examples of real situations constitute the so-called training sample, which is formed during a certain historical period. The elements of the training sample are described by a set of classification features.

The strategy of “learning with the teacher” implies the task of a specialist for each example of his belonging to a particular class of situations. When learning without a teacher, the system should independently allocate classes of situations. Critical education is intermediate between the first two. It is assumed that it is only possible to evaluate the correct operation of the network and indicate the desired direction of study. This situation is often found in systems associated with optimal control.

In the process of learning, an automatic construction of generalizing rules or functions describing the belonging of situations to classes, which the system will later use in interpreting unfamiliar situations, is carried out. In turn, the knowledge base is automatically formed from generalizing rules, which is periodically adjusted.

Self-learning systems built in accordance with these principles have the following disadvantages:

• relatively low adequacy of knowledge bases to real problems arising due to incompleteness and / or noise of the training set;

• low degree of explanability of the results;

• superficial description of the problem area and a narrow focus of application.

Neural networks are a classic example of technology based on examples. Neural networks are a generic name for a group of mathematical algorithms that have the ability to learn from examples, “later recognizing” the features of the patterns and situations encountered. Thanks to this ability, neural networks are used in solving problems of signal and image processing, pattern recognition, and also for prediction.

In case-based systems, the KB contains descriptions of specific situations (precedents). The search for a solution is based on analogies and includes the following steps:

• obtaining information about the current problem;

• comparison of the obtained information with the values ​​of attributes of precedents from the knowledge base;

• selection of a precedent from the knowledge base closest to the problem under consideration;

• adaptation of the selected precedent to the current problem;

• verification of the correctness of each solution obtained;

• entry of detailed information about the decision received in the BR.

Precedents are described by a set of attributes by which quick search indexes are built. In case-based systems, a fuzzy search is launched to produce a set of permissible alternatives , each of which is evaluated by a certain confidence coefficient. The most effective solutions are adapted to real-world situations using special algorithms. Precedent-based systems are used to disseminate knowledge and in contextual help systems.

Information repositories are subject-oriented, integrated, time-bound, immutable collection of data used to support management decision-making processes. Objectivity means that the data is grouped into categories and stored in accordance with the areas that they describe, and not with the applications that they use. Data is integrated into the repository in order to meet the requirements of the enterprise as a whole, and not a separate unit. The attachment of data to time expresses their “historicity”, i.e. the time attribute is always explicitly present in the data warehouse structures. Immutability means that once you get to the repository, the data does not change.

Technologies for extracting knowledge from data warehouses are based on statistical analysis and modeling methods aimed at finding models and relationships that are hidden in the aggregate data. These models can be further used to optimize the activities of the enterprise or company.

To extract meaningful information from data warehouses, there are special methods (OLAP analysis, Data Mining or Knowledge Discovery) based on the use of mathematical statistics methods, neural networks, inductive methods for constructing decision trees, etc.

Adaptive Information Systems

The need for adaptive information systems arises when the problem areas they support are constantly evolving. In this regard, adaptive systems must meet a number of specific requirements, namely:

• adequately reflect the knowledge of the problem area at each time point;

• be suitable for easy and quick reconstruction when the problem environment changes.

Adaptive properties of information systems are provided through the intellectualization of their architecture. The core of such systems is a constantly evolving model of the problem area, supported in a special knowledge base, the repository. The system kernel controls the processes for generating or reconfiguring software.

In the process of developing adaptive information systems, original or typical design is applied. The original design involves the development of an information system from a “clean slate” based on the formulated requirements. The implementation of this approach is based on the use of computer-aided design systems, or CASE-technologies (Designer2000, SilverRun, Natural Light Storm, etc.).

In a typical design, a typical development is carried out to the peculiarities of the problem area. To implement this approach, the tools of component (assembly) design of information systems (R / 3, BAAN IV, Prodis, etc.) are used.

The main difference between the approaches is that when using the CASE-technology, each time when the problem area changes, software is generated in general, and when using assembly technology, the modules are configured and only in rare cases is their processing.