Knowledge engineering is a set of methods, models and techniques aimed at creating systems designed to find solutions to problems based on existing knowledge. In fact, this term is understood as methodology, theory and technology, covering methods of analysis, extraction, processing and presentation of knowledge.
The essence of artificial intelligence lies in the scientific analysis and automation of intellectual functions inherent in man. At the same time, the complexity of their machine implementation is common to most problems. The study of AI made it possible to make sure that behind the solution of problems lies the need for expert knowledge, that is, the creation of a system that can not only memorize, but also analyze and use expert knowledge in the future; it can be used for practical purposes.
History of the term
Knowledge engineering and the development of intelligent information systems, in particular expert systems, are closely related.
At Stanford University in the USA in the 60-70s, under the leadership of E. Feigenbaum, aDENDRAL system, a little later - MYCIN. Both systems received the title of expert because of their ability to accumulate in computer memory and use the knowledge of experts to solve problems. This area of technology received the term "knowledge engineering" from the message of Professor E. Feigenbaum, who became the creator of expert systems.
Approaches
Knowledge engineering is based on two approaches: knowledge transformation and model building.
- Transformation of knowledge. The process of changing expertise and the transition from expert knowledge to its software implementation. The development of Knowledge Based Systems was built on it. Knowledge representation format - rules. The disadvantages are the impossibility of representing implicit knowledge and different types of knowledge in an adequate form, the difficulty of reflecting a large number of rules.
- Building models. Building AI is considered a type of simulation; building a computer model designed to solve problems in a particular area on an equal basis with experts. The model is not able to imitate the activity of an expert at the cognitive level, but it allows you to get a similar result.
Models and methods of knowledge engineering are aimed at the development of computer systems, the main purpose of which is to obtain the knowledge available from specialists and then organize it for the most effective use.
Artificial intelligence, neural networks and machine learning: what's the difference?
One of the ways to implement artificial intelligence is neuralnetwork.
Machine learning is an area of AI development aimed at studying methods for building self-learning algorithms. The need for this arises in the absence of a clear solution to a specific problem. In such a situation, it is more profitable to develop a mechanism that can create a method for finding a solution, rather than looking for it.
The commonly used term "deep" ("deep") learning refers to machine learning algorithms that require a large amount of computing resources to operate. The concept in most cases is associated with neural networks.
There are two types of artificial intelligence: narrowly focused, or weak, and general, or strong. The action of the weak is aimed at finding a solution to a narrow list of problems. The most prominent representatives of narrowly focused AI are the voice assistants Google Assistant, Siri and Alice. In contrast, strong AI abilities allow it to perform almost any human task. today, artificial general intelligence is considered a utopia: its implementation is impossible.
Machine learning
Machine learning refers to the methods in the field of artificial intelligence used to create a machine that can learn from experience. The learning process is understood as the processing of huge data arrays by the machine and the search for patterns in them.
The concepts of Machine learning and Data science, despite their similarity, are still different and each cope with their own tasks. Both instruments are included in the artificialintelligence.
Machine learning, which is one of the branches of AI, is algorithms based on which a computer is able to draw conclusions without adhering to rigidly set rules. The machine looks for patterns in complex tasks with a large number of parameters, finding more accurate answers, unlike the human brain. The result of the method is an accurate prediction.
Data science
The science of how to analyze data and extract valuable knowledge and information from them (data mining). It communicates with machine learning and the science of thinking, with technologies for interacting with large amounts of data. The work of Data science allows you to analyze data and find the right approach for subsequent sorting, processing, sampling and information retrieval.
For example, there is information about the financial expenses of an enterprise and information about counterparties that are interconnected only by the time and date of transactions and intermediate banking data. Deep machine analysis of intermediate data allows you to determine the most costly counterparty.
Neural networks
Neural networks, being not a separate tool, but one of the types of machine learning, are able to simulate the work of the human brain using artificial neurons. Their action is aimed at solving the task and self-learning based on experience gained with minimizing errors.
Machine learning goals
The main goal of machine learning is considered to be partial or complete automation of the search for solutions to various analytic altasks. For this reason, machine learning should give the most accurate predictions based on the data received. The result of machine learning is the prediction and memorization of the result with the possibility of subsequent reproduction and selection of one of the best options.
Types of machine learning
Classification of learning based on the presence of a teacher occurs in three categories:
- With the teacher. Used when the use of knowledge involves teaching the machine to recognize signals and objects.
- Without a teacher. The principle of operation is based on algorithms that detect similarities and differences between objects, anomalies, and then recognize which of them is considered dissimilar or unusual.
- With reinforcements. Used when a machine must perform tasks correctly in an environment with many possible solutions.
According to the type of algorithms used, they are divided into:
- Classical learning. Learning algorithms developed more than half a century ago for statistical offices and carefully studied over time. Used to solve problems related to working with data.
- Deep learning and neural networks. Modern approach to machine learning. Neural networks are used when generation or recognition of videos and images, machine translation, complex decision-making and analysis processes are required.
In knowledge engineering, ensembles of models are possible, combining several different approaches.
The benefits of machine learning
With a competent combination of different types and algorithms of machine learning, it is possible to automate routine business processes. The creative part - negotiating, concluding contracts, drawing up and executing strategies - is left to people. This division is important, because a person, unlike a machine, is able to think outside the box.
Problems of creating AI
In the context of creating AI, there are two problems of creating artificial intelligence:
- The legitimacy of recognizing a person as a self-organizing consciousness and free will and, accordingly, for recognizing artificial intelligence as reasonable, the same is required;
- Comparison of artificial intelligence with the human mind and its abilities, which does not take into account the individual characteristics of all systems and entails their discrimination due to the meaninglessness of their activities.
The problems of creating artificial intelligence lie, among other things, in the formation of images and figurative memory. Figurative chains in humans are formed associatively, in contrast to the operation of a machine; in contrast to the human mind, the computer searches for specific folders and files, and does not select chains of associative links. Artificial intelligence in knowledge engineering uses a specific database in its work and is not able to experiment.
The second problem is learning languages for the machine. Translation of text by translation programs is often carried out automatically, and the final result is represented by a set of words. For correct translationrequires understanding the meaning of the sentence, which is difficult for AI to implement.
The lack of manifestation of the will of artificial intelligence is also considered a problem on the way to its creation. Simply put, the computer has no personal desires, as opposed to the power and ability to perform complex calculations.
Modern artificial intelligence systems have no incentives for further existence and improvement. Most AIs are motivated only by a human task and the need to complete it. In theory, this can be influenced by creating a feedback between a computer and a person and improving the computer's self-learning system.
Primitiveness of artificially created neural networks. Today, they have advantages that are identical to the human brain: they learn based on personal experience, they are able to draw conclusions and extract the main thing from the information received. At the same time, intelligent systems are not able to duplicate all the functions of the human brain. The intelligence inherent in modern neural networks does not exceed the intelligence of an animal.
Minimum effectiveness of AI for military purposes. The creators of artificial intelligence-based robots are faced with the problem of AI's inability to self-learn, automatically recognize and correctly analyze the information received in real time.
