Everyone talks about artificial intelligence. Simple words, explain that this

"I want to do II. What should I explore? What languages \u200b\u200bto use? In which organizations to learn and work? "

We appealed for clarifying our experts, and we receive the answers we present to your attention.

It depends on your basic training. First of all, mathematical culture is needed (knowledge of statistics, theory of probabilities, discrete mathematics, linear algebra, analysis, etc.) and readiness to quickly learn. When implementing II methods, programming (algorithms, data structures, OOP, etc.) will be required.

Different projects require possessing different programming languages. I would recommend to know at least Python, Java and any functional language. We use experience with various databases and distributed systems. To quickly explore the best approaches used in the industry, knowledge of English is required.

I recommend learning in good Russian universities! For example, in MIPT, MSU, HSE there are relevant departments. A wide variety of thematic courses is available on Coursera, EDX, UDACITY, UDEMY and other Mooc sites. Some leading organizations have their own training programs in the field of AI (for example, the School of Data Analysis in Yandex).

Applied tasks solved by the methods of AI can be found in a wide variety of places. Banks, financial sector, consulting, retail, e-commerce, search engines, postal services, game industry, security systems industry and, of course, AVITO - all need specialists of various qualifications.

Enhance downgrade

We have a project on FINTEKH associated with machine learning and computer vision, in which his first developer wrote everything on C ++, then the developer came, who all rewrote on Python. So the language here is not the most important thing, as the language is first of all the tool, and it depends on how to use it. Just on some languages, the task is to solve faster, and on others more slowly.

Where to learn, it's hard to say - all our guys learned themselves, the benefit of the Internet and Google.

Enhance downgrade

I can advise from the very beginning to prepare myself to learn a lot. Regardless of what is meant to "do II" - work with large data or neural networks; Development of technology or support and teaching a certainly defined system already developed.

Let's take a Trend Profession of Data Scientist. What does this person do? In general, it collects, analyzes and prepares large data. It is those who grow and train the AI. And what should know and be able to Data Scientist? Static analysis and mathematical modeling - by default, and at the level of free possession. Languages \u200b\u200b- say, R, SAS, Python. It would also be nice to have any development experience. Well, generally speaking, a good date-safist must confidently feel in the database, algorithm, data visualization.

Not to say that such a set of knowledge can be obtained in every second technical university of the country. Large companies in whose priority Development of AI, they understand and develop appropriate training programs for themselves - there is, for example, a school of data analysis from Yandex. But you must make a report that this is not the scale where you come to courses "from the street", but you leave with them ready junior. Plast is big, and go to learn from discipline makes sense when the base (mathematics, statistics) is already covered at least within the university program.

Yes, time will leave decently. But the game is worth the candle, because a good Data Scientist is very promising. And very expensive. There is also another moment. Artificial intelligence is, on the one hand, no longer just an object of an excitement, but a technology published on the coil productivity. On the other hand, the AI \u200b\u200bis still developing. For this development requires a lot of resources, a lot of skills and a lot of money. While this is the level of the highest league. I will now say an obvious thing, but if you want to turn out to move progress on the tip of attacks and do it yourself, kill the Facebook or Amazon company.

At the same time, in a number of areas, technology is already used: in the banking sector, in a telecom, in industrial enterprises-giants, in retail. And there are already needed people who can support it. Gartner predicts that by 2020, 20% of all enterprises in developed countries will hire special employees to train neural networks used in these companies. So still there is a little time to rush it yourself.

Enhance downgrade

AI is now actively developing, and predicted for ten years ahead is difficult. For the next two or three years, the approaches on the basis of neural networks and calculations based on the GPU will dominate. The leader in this area is Python with the JUPYTER interactive environment and the NUMPY, SCIPY libraries, TensorFlow.

There are many online courses that give a basic idea of \u200b\u200bthese technologies and general principles of AI, for example, Andrew NG course. And in terms of training this topic, Russia is now more efficiently effective independent training or in the local interest group (for example, in Moscow I know about the existence of at least a couple of groups where people share experiences and knowledge).

Enhance downgrade

Enhance downgrade

To date, the fastest progressive part of artificial intelligence is, perhaps, neural networks.
Study of neural network and AI should begin with the development of two sections of mathematics - linear algebra and probability theory. This is a mandatory minimum, unshakable pillars of artificial intelligence. Applicants wishing to comprehend the foundations of AI, when choosing a university, in my opinion, should pay attention to faculties with a strong mathematical school.

The next step is to study the issues of the issue. There is a huge number of literature, both educational and special. Most publications on the topic of artificial intelligence and neural networks are written in English, but Russian-speaking materials are also published. Useful literature can be found, for example, in the publicly available digital library Arxiv.org.

If we talk about the directions of activity, then here you can allocate training for applied neural networks and the development of completely new neural network options. Bright example: There is such a very popular specialty now - "Data System" (Data Scientist). These are developers who, as a rule, study and preparing some sets of data for teaching neural networks in specific, applied areas. Summarizing, emphasize that every specialization requires a separate path of preparation.

Enhance downgrade

Before proceeding with narrow-profile courses, you need to study a linear algebra and statistics. Immersion in AI would advise to start with the textbook "Machine training. Science and art of building algorithms that extract knowledge from data, "this is a good allowance for beginners. At Coursera it is worth listening to the introductory lectures to K. Vorontsov (it emphasizes that they require good knowledge of the linear algebra) and the Machine Learning course of Stanford University, who reads Andrew NG, Professor and Chapter Baidu AI Group / Google Brain.

The bulk is written on Python, then go R, Lua.

If we talk about educational institutions, it is better to go to courses at the departments of applied mathematics and computer science, there are suitable educational programs. To check its abilities, you can take part in the Kaggle competitions, where large global brands offer their cases.

Enhance downgrade

In any case, before proceeding with projects, it would be nice to get theoretical basis. There are many places where you can get a formal degree of a master in this area, or improve your qualifications. So, for example, Scolteh offers master programs in the areas of "Computational Science and Engineering" and "Data Science", which includes courses "Machine Learning" and Natural Language Processing. You can also mention the Institute of Intellectual Cyber \u200b\u200bSystems of Niya Mafi, Faculty of Computational Mathematics and Cybernetics of Moscow State University and the Department of Intellectual Systems MFT.

If formal education is already available, there are a number of courses on various MOOC platforms. For example, EDX.ORG offers artificial intelligence courses from Microsoft and Colombian University, the last of which offers a micro-master program for moderate money. I would like to emphasize that you can usually get the knowledge yourself and free, payment is only for the certificate if it is needed for your resume.

If you want to "deeply plunge" in the topic, a number of companies in Moscow offer weekly intensives with practical classes, and even offer equipment for experiments (for example, newProlab.com), however, the price of such courses from several tens of thousand rubles.

From companies that are engaged in the development of artificial intelligence, you probably know Yandex and Sberbank, but there are many other different sizes. For example, this week, the Ministry of Defense opened in Anapa Military Innovation Technopolis Era, one of which is the development of AI for military needs.

Enhance downgrade

Before studying artificial intelligence, it is necessary to solve the principal question: to take a red tablet or blue.
Red tablet - become a developer and plunge into the cruel world of statistical methods, algorithms and permanent comprehension of unknown. On the other hand, it is not necessary to immediately throw in the "Rabbit Nora": you can become a managers and create AI, for example, as a project manager. These are two fundamentally different ways.

The first is perfectly suitable if you have already decided that you will write algorithms for artificial intelligence. Then you need to start from the most popular destination today - machine learning. To do this, you need to know the classical statistical methods of classification, clustering and regression. It will also be useful to also get acquainted with the basic measures to assess the quality of the solution, their properties ... and all that will fall on the way.

Only after the base is mastered, it is worth shifting more special methods: decision making trees and ensembles of them. At this stage, it is necessary to deeply immerse yourself in the main ways of building and learning the models - they are hiding for the barely decent words of Beguing, Busting, Filia or Blendding.

Immediately it is worth knowing the methods of controlling the retraining of models (another "Ing" - overfitting).

And finally, at all, the Jedic level is to obtain highly specialized knowledge. For example, for deep learning it will be necessary to master the main architectures and algorithms of the gradient descent. If there are interesting tasks of processing a natural language, then I recommend to study recurrent neural networks. And the future creators of algorithms for the processing of pictures and video should be thoroughly deepening in the sweeping neural networks.

The last two mentioned structures are bricks of popular architectures today: Connecting Networks (GAN), relational networks, combined networks. Therefore, learning them will be worth it, even if you do not plan to learn a computer to see or hear.

A completely different approach to the study of AI is the "blue tablet" - begins with the search. Artificial intelligence gives rise to a bunch of tasks and integers: from the managers of II projects to data engineers who can prepare data, clean them and build scalable, loaded and fault-tolerant systems.

So, with the "managerial" approach, you first need to evaluate your abilities and bemarkund, and only then choose where and what to learn. For example, even without a mathematical warehouse of the mind, you can engage in design II interfaces and visualizations for smart algorithms. But get ready: after 5 years, artificial intelligence will begin to troll you and call "humanitarian".

Major ML methods are implemented as ready-made libraries available for connecting in different languages. The most popular languages \u200b\u200bin ML today are: C ++, Python and R.

There are many courses in both Russian and English, such as Yandex data analysis school, SkillFactory and Otus courses. But before investing time and money in specialized training, I think it is worth "to penetrate the topic": watch open lectures on YouTube with DataFest conferences over the past years, to undergo free courses from Coursera and Habrahabra.

Artificial Intelligence (AI, Artificial Intelligence, AI) is a science of creating intelligent technologies and computer programs.

Artificial intelligence is closely related to the task of understanding human intelligence using computer technologies. At the moment, it is impossible to say exactly which computational methods can be called intelligent. Some intelligence mechanisms are open to understanding, the rest are not. At the moment, programs are used in the programs that are not found in humans.

Artificial intelligence has a scientific direction that studies the solution of human intellectual activity. Artificial intelligence is aimed at performing creative tasks in the area, knowledge of which is stored in the intelligent program system - the database of knowledge.

With these knowledge, the program mechanism works - task Raster. Then the person gets an idea of \u200b\u200bthe result of the program through the intelligent interface. The result of an artificial intelligence program is the recreation of intellectual argument or reasonable action.

One of the main properties of artificial intelligence is the ability to self-study. First of all, it heuristic training - Continuous training of the program, the formation of the learning process and its own goals, analysis and awareness of their training.

Scientific direction studying artificial intelligence began to emerge for a long time ago:

• philosophers thought about the knowledge of the inner world of man
• psychologists studied human thinking
• mathematics were engaged in calculations

Soon, the first computers were created, which allowed to perform calculations overtaking by human speed. Then the scientists began to ask the question: where the boundary of the capabilities of the computers and can they achieve the human level?

Alan Turing is an English scientist, the pioneer of computing technology, wrote the article "Can the car think?", Where the method described that will help determine at what point the computer can be compared with a person. This method got the name - test Turing.

The essence of the method is that the person first answered the questions of the computer, then the questions of another person and at the same time not knowing who exactly asked him questions. If, when answering a computer questions, a person has not suspected that this is a car, then the passage of the Turing test can be considered successful, as well as the computer is an artificial intelligence.

Thus, if the computer shows similar with human behavior in any natural situations and is able to support a dialogue with a person, then we can say that this is an artificial intelligence. Another alleged definition method is an intellectual machine, this is its ability to work and the ability to feel.

There are many different approaches to the study and understanding of artificial intelligence.

Symbolic approach

The character approach became the first in the digital era of machines. After creating the language of symbolic calculations of the Lisp, its authors have begun to implement the intellect. Symbolic approach Use weak formalized views. So far, only a person is able to perform intellectual work and the work associated with the work of the task. The work of computers in this direction is biased and in fact cannot be carried out without human participation.

Symbolic calculations helped to create rules for solving tasks in the process of performing a computer program. However, it was possible to solve only the simplest tasks, and when any complex task appears, it is necessary to connect a person again. Thus, such systems do not allow them to be called intellectual, as their capabilities do not allow to solve emerging difficulties and improve already knowing the ways to solve problems for solving new ones.

Logic approach

The logical approach is based on modeling the arguments and the application of the logical programming language. For example, programming language Prologue is based on a set of logical output rules without rigid consistent actions to achieve results.

Agent-oriented approach

An agency-oriented approach is based on methods of helping intelligence to survive in the environment to achieve certain results. The computer perceives its environment and affects it using the set methods.

Hybrid approach

The hybrid approach includes expert rules that can be created by neural networks, and generating rules through statistical training.

Modeling reasoning

There is such a direction in the study of artificial intelligence as modeling reasoning. This area includes creating symbolic systems, for setting tasks and solving them. The task must be translated into a mathematical form. At the same time, she still has no algorithm for solving due to complexity. Therefore, modeling of reasoning contains proof theorems, decision making, planning, prediction, etc.

Natural language processing

Another important direction of artificial intelligence is natural language processingwhere the analysis and processing of texts on a person understandable to humans is done. The purpose of this area is the processing of a natural language for independent acquisition of knowledge. The source of information may be the text entered into the program or received from the Internet.

Presentation and use of knowledge

Engineering knowledge is the direction of acquiring knowledge from information, their systematization and further use for solving various tasks. With the help of special databases, expert systems receive data for the process of finding solutions to the tasks.

Machine learning

One of the main requirements for artificial intelligence is the possibility of a machine to independently learning without the intervention of the teacher. Machine learning includes objective recognition tasks: recognition of symbols, text and speech. This also includes computer vision associated with robotics.

Biological modeling of II

There is such a direction as quasibiological paradigmwhich otherwise is called Biocomputting. This direction in artificial intelligence studies the development of computers and technologies using living organisms and biological components - biocomputers.

Robotics

The area of \u200b\u200brobotics is closely related to artificial intelligence. The properties of artificial intelligence are also necessary for robots to perform many different tasks. For example, for navigating and defining your location, study of items and planning your movement.

Areas of application of artificial intelligence

Artificial intelligence is created to solve problems from various areas:

• Intelligent systems for education and recreation.
• Synthesis and recognition of text and human speech are used in customer service systems.
• Image recognition systems are used used in safety systems, with optical and acoustic recognition, medical diagnostics, target definition systems.
• The computer games are used to calculate the game strategy, imitation of character behaviors, finding the path in space.
• Systems of algorithmic trade and decision-making.
• Financial systems for advice and financial management.
• Robots used in industry to solve complex routine tasks: robots for patient care, robots Consultants, as well as business-friendly human activities: Robots Rescuers, robots Miners.
• Management of human resources and recruiting, viewing and ranking candidates, predicting the success of employees.
• Recognition and spam filtering systems in email.

This is not all areas where artificial intelligence can be applied.

Now the creation of artificial intelligence is one of the important tasks of a person. However, there is no single point of view on what can be considered intellect, but what is impossible. Many questions arouse disputes and doubts. Is it possible to create an intellectual mind that will understand and solve problems of people? Mind, not devoid of emotions and with the abilities inherent living organism. So far there is no time when we see it.

What is this artificial intelligence? Undoubtedly, many have heard of cars capable of managing their movement without human help, speech recognition devices, such as Apple's Siri, Amazon's Alexa, Google's Assistant and Microsoft's Cortana. But this is not all the possibilities of artificial intelligence (AI).

AI was "open" for the first time in the 1950s. Over the years, it was expected by ups and falls, but at the present stage of human development, artificial intelligence is considered as a key technology of the future. Thanks to the development of electronics and the appearance of faster processors, an increasing number of applications begins to use AI. Artificial intelligence is an unusual program technology with which every engineer must familiarize themselves. In this article we will try to describe this technology to describe this technology.

Artificial intelligence is defined

AI is a sacrifice of computer science, which includes more reasonable use of computers and electronic components, imitating the human brain. Intellect is the ability to acquire knowledge and experience and apply them to solve problems. AI is especially useful when analyzing and interpreting data arrays and extracting actual information from it. Of the information comes an understanding that can be applied to making decisions or any kind of action.

Areas of research

Artificial intelligence is a wide technology with a multitude of possible applications. Usually it is divided by letters. We will make a small review of each of them:

• Solving common tasks - no specific algorithmic solution. Tasks with uncertainty and ambiguity.
• Expert Systems - software that contains the basis of knowledge of rules, facts and data obtained from several individual experts. The database may be requested to solve problems, diagnose diseases or consultation.
• Processing of a natural language (NLP) is used to analyze texts. Voice recognition is also part (NLP).
• Computer vision - analysis and understanding of visual information (photography, video and so on). An example is machine vision and face recognition. Used in "autonomous" cars and production lines.
• Robotics - creating smarter, adaptive and "independent" robots.
• Games: AI plays a great game. Computers are already programmed to play and win in chess, poker and in go.
• Machine training - procedures allowing the computer to study on the basis of input data and comprehend the results. Neural networks make up the basis of machine learning.

How artificial intelligence works

Conventional computers use algorithms for solving problems. The instruction of instructions leads to step-by-step action to obtain results. Traditional forms of artificial intelligence are based on the knowledge bases and logical output mechanisms that use various mechanisms to work with the knowledge base through the user interface. Useful results are obtained by some of the methods listed below:

• Search: Search algorithms use a database of information collected in graphs or trees. Search is the main method of artificial intelligence.
• Logic: deductive and inductive reasoning is used to determine the truth or falsity of statements. This includes both the logic of statements and the logic of predicates.
• Rules: Rules are a series of instructions "if", which can be found to determine the result. The rules based systems are called expert systems.
• Probability and statistics: Some tasks can be solved, and solutions are located, thanks to the use of standard mathematical theory of probability and statistics.
• Lists: Some information types can be saved in lists that become available for search.
• Other forms of knowledge are schemes, frames and scenarios, which are structures encapsulating various types of knowledge. Search methods are looking for answers on relevant requests.

Traditional or inherited methods of AI, such as search, logic, probability and rules, are considered the first wave of artificial intelligence. These methods are still used and well perceive knowledge and reasoning, especially for a narrow circle of tasks. In the first wave, there are no human training features and abstraction solutions. These qualities are now available in the second wave of artificial intelligence, thanks to neural networks and machine learning.

Neural networks

Today, most research and development of AI are based on the use of neural networks or artificial neural networks (INS). These networks consist of artificial neurons imitating neurons in the human brain, which are responsible for our thinking and training. Each neuron is a node of a complex relationship that binds many neurons with others by means of synapses. Ins imitates this network.

Each node has several suspended inputs, as well as the output and installation of the threshold (drawing above). Such nodes are usually implemented in software, although hardware emulation is also possible. The typical scheme consists of three layers - the inlet layer, hidden (processing or training layer) and the output layer:

Some mechanisms use the opposite distribution to provide feedback, which changes the weight of the input of some nodes as new information has been received.

Machine learning and deep training

Machine training is a method for learning a computer to recognize images. A computer or device is "learned" with an example, and then special programs are launched to compare input with the trained value. As a rule, there are huge amounts of data for software training. Machine training programs are intended for automatic study, as they receive more knowledge and experience thanks to new materials.

Neural networks are commonly used for machine learning, but other algorithms can also be used. Then the software can change itself by improving recognizability based on new input data. Now some machine learning systems can independently recognize images without learning, and then modify themselves to further improve.

Deep learning is an extended case of machine learning. It also uses neural networks called deep neural networks (SCS). They include additional hidden levels of computing for further improvement of their capabilities. Mass learning is required. Programmers can increase productivity by playing with interconnect weights. SCS also require matrix processing. However, it should be noted that the SCS use statistical weights, so the results, say, in visible recognition, may be not 100%. In addition, debugging such systems is a very painstaking job.

Machine learning and deep learning are widely used to analyze large data arrays, as well as in computer vision and speech recognition. They can also be applied in other areas, such as medicine, jurisprudence and finance.

Artificial Intelligence Software

For programming, the AI \u200b\u200bcan use almost any programming language, but some languages \u200b\u200bhave certain advantages. Profile languages \u200b\u200bdesigned specifically for AI include Lisp and Prolog. Lisp, one of the oldest high-level languages, processes lists. Prolog is based on logic. Today, C ++ and Python are popular. There is also special software for the development of expert systems.

Several major users of AI provide development platforms, including Amazon, Baidu (China), Google, IBM and Microsoft. These companies offer pre-trained systems as a start point for some common applications, such as voice recognition. Processor suppliers, such as NVIDIA and AMD, also offer specific support.

Hardware for artificial intelligence

Starting artificial intelligence software on a computer usually requires high speed and a large amount of memory. However, some simple applications can work on an 8-bit processor. Some of the modern processors are more than suitable, and several parallel processors can be an ideal solution for certain applications. In addition, special processors have been developed for some applications.

Graphics Processors (GPU) are an example of focusing architecture and set of instructions for specified use to optimize performance. For example, NVIDIA special processors for independent driving cars and AMD graphic processors. Google has developed its own processors to optimize their search engines. Intel and Knupaath also offer software support for their advanced processors. In some cases, special logic in ASIC or FPGA can implement a specific application.

Activity and current status

Artificial intelligence was once considered an exotic software designed for special needs. The requirement of high-speed computers with a large amount of memory limited its use. Today, thanks to super fast processors, multi-core processors and cheap memory, the AI \u200b\u200bhas become more popular. Google search engines that we all use daily are based on artificial intelligence.

To date, the emphasis is undoubtedly made on neural networks and deep machine learning. While recognizing voice and self-propelled vehicles are still in the spotlight, other key applications appear, such as face recognition, unmanned navigation, robotics, medical diagnosis and finance. The development also contains advanced military applications (for example, autonomous weapons).

The future of AI looks promising. According to Orbis Research, by 2022, an increase in the global market for artificial intelligence with a cumulative annual growth rate of more than 35% is expected. The International Data Corporation (IDC) is also positive, stating that the costs of artificial intelligence are expected to increase to $ 47 billion in 2020, compared with 8 billion in 2016.

Many people have a logical question - will the artificial intelligence of people of some professions replace the artificial intelligence, and what will it be for professions? The answer sounds as follows - "Perhaps only some". Most likely, computers based on artificial intelligence will help improve the performance of some professions, increasing productivity, efficiency and decision-making. However, some workplaces in the industry will still be lost, since a lot of development receives robotics, but the replacement of man by cars will lead to the creation of new jobs related to servicing these machines.

Another question defined by many people can be an artificial intelligence dangerous for humanity? AI was smart, but not so smart. Its main purpose will be the analysis of data, solving problems and decision-making based on available information and distilled knowledge. People still dominate, especially when it comes to innovation and work. However, it is difficult to predict the future. At least, at this stage of development, there are no smart robots, not yet ...

The concept of artificial intelligence (AI or AI) combines not only technologies to create intelligent machines (including computer programs). AI is also one of the directions of scientific thought.

Artificial Intelligence - Definition

Intelligence - This is a mental component of a person who has the following abilities:

• adaptable;
• learning by accumulation of experience and knowledge;
• the ability to apply knowledge and skills to managing the environment.

Intellect unites all human abilities to the knowledge of reality. With the help of it, a person thinks, remembers new information, perceives the environment and so on.

Under the artificial intelligence is one of the directions of information technology, which is engaged in the study and development of systems (machines), endowed with the possibilities of human intelligence: learning ability, logical reasoning, and so on.

At the moment, work on artificial intelligence is carried out by creating new programs and algorithms that decisive tasks just as a person does.

Due to the fact that the definition of AI evolves as this direction develops, it is necessary to mention the AI \u200b\u200bEFFECT. The effect is understood under it that creates an artificial intelligence that has achieved some progress. For example, if the AI \u200b\u200bhas learned to perform any actions, then critics are immediately connected, which prove that these successes do not testify to the presence of thinking from the car.

Today, the development of artificial intelligence is in two independent areas:

• neurokaberetics;
• logical approach.

The first direction involves the study of neural networks and evolutionary calculations from the point of view of biology. The logical approach implies the development of systems that imitate intelligent high-level processes: thinking, speech and so on.

The first work in the field of AI began to lead in the middle of the last century. Pioneer studies in this direction has become Alan TuringAlthough certain ideas began to express philosophers and mathematics in the Middle Ages. In particular, at the beginning of the 20th century, a mechanical device was presented capable of solving chess tasks.

But truly this direction was formed by the middle of the last century. The emergence of work on AI was preceded by research on the nature of man, ways to know the surrounding world, the possibilities of the mental process and other fields. By that time, the first computers and algorithms appeared. That is, a foundation was created on which a new direction of research was originated.

In 1950, Alan Turying published an article in which I was asked about the possibilities of future machines, as well as whether they were able to bypass a person in terms of rationality. It was this scientist who developed the procedure called later in his honor: Turing Test.

After the publication of the works of the English scientist, new research in the field of AI appeared. According to Tyurring, only the car that is impossible to distinguish from a person can be recognized as it is possible. At about the same time, when the scientist appeared, the concept called Baby Machine originated. It provided for the progressive development of AI and the creation of machines, the thinking processes of which are first form at the level of the child, and then gradually improve.

The term "artificial intelligence" originated later. In 1952, a group of scientists, including Turing, gathered at the American University of Dartmund to discuss issues related to AI. After that meeting, the active development of cars with the possibilities of artificial intelligence began.

A special role in creating new technologies in the field of AI played military departments that actively financed this direction of research. Subsequently, work in the field of artificial intelligence began to attract large companies.

Modern life puts more complex tasks in front of the researchers. Therefore, the development of AI is carried out in principle of other conditions, if we compare them with what happened during the origin of artificial intelligence. The processes of globalization, the action of attackers in the digital sphere, the development of the Internet and other problems - all this puts complex tasks before scientists, the solution of which lies in the field of AI.

Despite the successes achieved in this area in recent years (for example, the emergence of autonomous equipment), the voices of skeptics, which do not believe in the creation of truly artificial intelligence, and not a very capable program. A number of critics fears that the active development of AI will soon lead to a situation where the machines will completely replace people.

Directions of research

Philosophers have not yet come to a common opinion about what the nature of human intelligence, and what is its status. In this regard, in scientific works dedicated to the AI, there are many ideas telling what tasks artificial intelligence solves. There is also no unified understanding of the issue, which car can be considered reasonable.

Today, the development of artificial intelligence technologies is in two directions:

1. Downward (semiotic). It provides for the development of new systems and base bases that imitate high-level mental processes of the type of speech, expressions of emotions and thinking.
2. Ascending (biological). This approach involves conducting research in the field of neural networks through which the intellectual behavior models are created from the point of view of biological processes. On the basis of this direction, neurocomputers are created.

Determines the ability of artificial intelligence (machine) to think just like a person. In general understanding, this approach provides for the creation of AI, whose behavior does not differ from human actions in the same, normal situations. In fact, the test of Turing suggests that the car will be reasonable only if, when communicating with it it is impossible to understand who says: a mechanism or a living person.

Books in the genre fiction offer another method of assessing the possibilities of AI. This artificial intelligence will become in the event that he will feel and can create. However, this approach to definition does not withstand practical applications. Already, for example, cars are created that have the ability to respond to environmental changes (cold, heat and so on). At the same time, they can not feel the way man does.

Symbolic approach

Success in solving tasks is largely determined by the ability to flexibly approach the situation. Machines, in contrast to people, interpret the obtained data in a single way. Therefore, only a person participates in solving problems. The machine conducts operations based on written algorithms that exclude the use of several abstraction models. Program flexibility can be achieved by increasing the resources involved during solving problems.

The above disadvantages are characteristic of the symbolic approach used in the development of AI. However, this direction of the development of artificial intelligence allows you to create new rules in the process of calculating. And the problems arising from a symbolic approach are able to solve logical methods.

Logic approach

This approach involves the creation of models imitating the reasoning process. It is based on the principles of logic.

This approach does not provide for the use of rigid algorithms that lead to a specific result.

Agent-oriented approach

It uses intelligent agents. This approach suggests the following: Intellect is a computing part by which the goals are achieved. The car plays the role of an intelligent agent. It knows the environment using special sensors, and interacts with it by means of mechanical parts.

An agency-oriented approach focuses on the development of algorithms and methods that allow machines to maintain performance in various situations.

Hybrid approach

This approach involves combining neural and symbolic models, due to which the solution of all tasks associated with the processes of thinking and calculations is achieved. For example, neural networks can generate a direction in which the operation of the machine is moving. And static learning provides that basis by which tasks are solved.

According to experts' forecasts Gartner.However, by the beginning of the 2020s, almost all manufactured software products will use artificial intelligence technologies. Also, experts suggest that about 30% of investments in the digital sphere will be on the AI.

According to analysts Gartner, artificial intelligence opens up new opportunities for cooperation of people and cars. In this case, the process of displacing a person AI cannot be stopped and in the future it will accelerate.

In company Pwc. It is believed that by 2030, the volume of the global gross domestic product will grow by about 14% due to the rapid introduction of new technologies. Moreover, approximately 50% of the increase will ensure an increase in the efficiency of production processes. The second half of the indicator will be an additional profit obtained by implementing AI to products.

Initially, the effect of the use of artificial intelligence will receive the United States, since in this country the best conditions for the operation of machines on the AI \u200b\u200bare created. In the future, they will be ahead of China, which will remove the maximum profit by introducing such technologies into products and its production.

Experts companies Saleforce. It is claimed that the AI \u200b\u200bwill increase the yield of small businesses by about 1.1 trillion dollars. And this will happen by 2021. Partly to achieve the specified indicator will succeed at the expense of the implementation of the solutions offered by the AI, in the system responsible for communicating with clients. At the same time, the efficiency of production processes will improve due to their automation.

The introduction of new technologies will also create an additional 800 thousand jobs. Experts note that the specified indicator levels the loss of vacancies that occurred due to the automation of processes. According to the forecast of analysts based on the results of the survey among companies, their costs for the automation of production processes by the beginning of the 2020s will increase to about 46 billion dollars.

In Russia, work in the field of AI is also underway. For 10 years, the state has financed more than 1.3 thousand projects in this field. Moreover, most of the investment has gone to the development of programs not related to the conduct of commercial activities. This shows that the Russian business community is not yet interested in introducing artificial intelligence technologies.

In total, about 23 billion rubles invested in Russia in Russia. The amount of state subsidies is inferior to the volume of funding from the field of AI, which demonstrate other countries. In the United States, about $ 200 million allocate for these purposes.

Mainly in Russia from the state budget allocate funds for the development of technologies of AI, which are then applied in the transport sector, defense industry and in safety-related projects. This circumstance indicates that in our country it is more often investing in directions that allow you to quickly achieve a certain effect of invested funds.

The above study also showed that in Russia a high potential is now accumulated for training specialists who can be involved in the development of technologies of AI. In the last 5 years, approximately 200 thousand people passed training in the directions associated with AI.

Technologies AI develop in the following directions:

• solving problems that allow us to bring the possibilities of AI to human and find ways to integrate them into everyday life;
• the development of a full-fledged mind, by means of which tasks facing humanity will be solved.

At the moment, researchers are focused on developing technologies that solve practical tasks. While scientists did not approach the creation of a full-fledged artificial mind.

Many companies are engaged in developing technologies in the field of AI. "Yandex" for more than one year applies them in the work of the search engine. Since 2016, the Russian IT company has been studying in the field of neural networks. The latter change the nature of the search engines. In particular, neural networks compare the user-entered request with a certain vector number that most fully reflects the meaning of the task. In other words, the search is not conducted by the word, namely, in fact, the information requested by man.

In 2016. Yandex Launched service "Zen"which analyzes user preferences.

The company ABBYY. Recently a system appeared Compreno.. With the help of it, it is possible to understand the text written in the natural language. Other systems based on artificial intelligence technologies are also relatively recently published on the market.

1. Findo. The system can recognize human speech and is engaged in searching for information in various documents and files using complex requests.
2. Gamalon. This company presented the system with the ability to self-educate.
3. Watson. IBM computer, using a large number of algorithms in the process of finding information.
4. Viavoice. Human speech recognition system.

Large commercial companies do not bypass the side of achievements in the field of artificial intelligence. Banks actively introduce similar technologies to their activities. With the help of AI based systems, they carry out operations on stock exchanges, lead property management and perform other operations.

The defense industry, medicine and other spheres are implementing object recognition technologies. And companies involving computer games use AI to create another product.

Over the past few years, a group of American scientists is working on a project. NeilIn which researchers offer a computer to recognize what is depicted in the photo. Specialists suggest that in this way they will be able to create a system capable of self-study without external intervention.

Company VisionLab. introduced its own platform Luna.which can realize the individuals choosing them from a huge cluster of images and videos. This technology is used today large banks and network retailers. With LUNA, you can compare the preferences of people and offer them the relevant goods and services.

The Russian company works on such technologies N-Tech Lab. At the same time, its specialists feed on to create a system recognition system based on neural networks. According to the latest data, Russian development is better coping with tasks than a person.

According to Stephen Hawking, the development of artificial intelligence technologies will lead to the death of humanity. The scientist noted that people because of the introduction of AI will begin to gradually degrade. And in a natural evolution, when a person for survival needs to constantly fight, this process will inevitably lead to his death.

In Russia, the issue of introducing AI positively consider. Alexey Kudrin once stated that the use of such technologies would allow about 0.3% of the WFP to reduce the cost of ensuring the work of the state apparatus. Dmitry Medvedev predicts the disappearance of a number of professions due to the introduction of AI. However, the official stressed that the use of such technologies will lead to the rapid development of other industries.

According to experts of the World Economic Forum, by the beginning of the 2020s in the world due to the automation of the production of jobs, about 7 million people are lost. The introduction of AI with a high probability will cause the transformation of the economy and the disappearance of a number of professions related to data processing.

Experts McKinsey. It is stated that the process of automating production will be more active in Russia, China and India. In these countries, in the near future, up to 50% of workers will lose their places due to the introduction of AI. Their place will occupy computerized systems and robots.

According to McKinsey, artificial intelligence will replace professions providing for physical labor and processing information: retail, hotel staff and so on.

By the middle of the current century, experts of the American company believe, the number of jobs around the world will decrease by about 50%. Persons of people will take cars capable of conducting similar operations with the same or higher efficiency. At the same time, experts do not exclude the option in which this forecast will be implemented before the specified period.

Other analysts note the harm that robots can apply. For example, McKinsey experts pay attention to the fact that robots, unlike people, do not pay taxes. As a result, due to the decline in revenues to the budget, the state will not be able to support infrastructure at the same level. Therefore, Bill Gates proposed to introduce a new tax on robotic techniques.

AI technologies increase the efficiency of companies by reducing the number of errors made. In addition, they allow you to increase the speed of operations to the level that the person cannot achieve.

The essence of artificial intelligence in the format of questions and answers. The history of creation, research technology, is artificial intelligence with IQ and is it possible to compare it with human. I answered questions professor of Stanford University John McCarthy.

What is artificial intelligence (AI)?

Artificial intelligence is an area of \u200b\u200bscience and engineering, engaged in creating machinery and computer programs with intelligence. It is associated with the task of using computers to understand human intelligence. At the same time, artificial intelligence should not be limited only by biologically observed methods.

Yes, but what is intelligence?

Intellect - the ability to come to a solution with the help of calculations. Intellect of different types and levels have people, many animals and some cars.

Is there no definition of intelligence that does not depend on its correlation with human intelligence?

So far there is no understanding, what types of computational procedures we want to be called intellectual. We know far from all the mechanisms of intelligence.

Is the intelligence of an unequivocal concept, to the question "Does this car intelligence?" It was possible to answer "yes" or "no"?

Not. AI studies have shown how to use only some of the mechanisms. If only well-studied models are required to perform the task, very impressive results are obtained. Such programs have a "small" intelligence.

Is the artificial intelligence attempt to imitate human intelligence?

Sometimes, but not always. On the one hand, we learn how to make machines solve problems, watching people or for the work of our own algorithms. On the other hand, research researchers use algorithms that are not observed in humans or require much greater computational resources.

Computer programs have IQ?

Not. IQ is based on the pace of development of intellect in children. This is an attitude of age, in which the child usually gains a certain result, to the age of the child. This assessment appropriately applies to adults. IQ correlates well with various indicators of success or failure in life. But the creation of computers that can dial a high score in IQ tests will be poorly connected with their utility. For example, the child's ability to repeat the long sequence of numbers is well correlated with other intelligent abilities. It shows how much information a child can remember at a time. In this case, the deduction in the memory numbers is a trivial task even for the most primitive computers.

How to compare human and computer intelligences?

Arthur R. Jensen, a leading researcher in the field of human intelligence, as an "heuristic hypothesis" argues that ordinary people have the same intelligence mechanisms and intellectual differences are associated with "quantitative biochemical and physiological conditions". These include the speed of thinking, short-term memory and the ability to form accurate and recoverable long-term memories.

Regardless of whether Jensen's point is right in relation to human intelligence, the situation in the AI \u200b\u200bis the opposite.

Computer programs have a large supply of speed and memory, but their abilities correspond to intellectual mechanisms that program developers understand and can invest in them. Some abilities that children usually do not develop before adolescent age are introduced. Others who own two-year children are still missing. The point is even more aggravated by the fact that cognitive sciences still cannot accurately determine what human abilities are. Most likely, the organization of intellectual mechanisms of AI is relatively different from those in humans.

When a person manages to solve the task faster than a computer, this suggests that the developers lack the understanding of the intelligence mechanisms necessary to effectively perform this task.

When did the study begin?

After World War II, several people began to work independently on intellectual machines. English Mathematics Alan Turing may have been the first of them. He read his lecture in 1947. Turing One of the first to decide that the AI \u200b\u200bis best investigated by programming computers, and not designing machines. By the end of the 1950s there were many researchers AI, and most of them founded their work on programming computers.

Is the goal to put the human mind into a computer?

Human mind has a lot of features, it is hardly possible to imitate each of them.

What is Turing Test?

In Article A. Alan Tyurin, 1950, "Computer and Mind" discussed conditions for the hardware of the machine by intelligence. He argued that if the car could successfully pretend to be a man before a reasonable observer, then you, of course, should consider it reasonable. This criterion will satisfy most people, but not all philosophers. The observer must interact with a machine or man through an I / O tool to eliminate the need to simulate the external or human vote. The task of both cars and a person is to make the observer consider themselves to be a man.

Turing Turing is one-sided. The machine, successfully passing the test, should definitely be considered reasonable, even if it does not have knowledge of people sufficient to imitate them.

The book of Daniel Dennet "BrainChildren" contains an excellent discussion of the Turing test and its various parts that were implemented successfully, that is, with restrictions on knowledge of the observer about the AI \u200b\u200band the subject of discussion. It turns out that some people are quite easy to convince that a fairly primitive program is reasonable.

Is the goal of achieving the human level of intelligence?

Yes. The ultimate goal is to create computer programs that can solve problems and achieve goals as well as a person. However, scientists conducting research in narrow regions put much less ambitious goals.

How far is the artificial intelligence from the achievement of the human level? When will it happen?

The human-level intelligence can be achieved by writing a large number of programs, and collecting extensive knowledge bases on the facts in languages \u200b\u200bthat are used today to express knowledge.Nevertheless, most researchers of AI believe that new fundamental ideas are needed. Therefore, it is impossible to predict when the human level intelligence will be created.

Is the computer machine that can be intellectual?

Computers can be programmed to simulate any type of machine.

The speed of computers allows them to have intelligence?

Some people think that they are required both faster computers and new ideas. Computers and 30 years ago were fast enough. If we only knew how to program them.

What about the creation of a "children's car", which could improve by reading and learning on their own experience?

This idea was repeatedly offered since the 1940s. In the end, it will be implemented. Nevertheless, the AI \u200b\u200bprograms have not yet reached the level allowing a lot of what a child is learning during vital activity. Existing programs are not well understood well to learn a lot by reading.

Are the theory of computability and computational complexity to the keys to the AI?

Not. These theories are relevant, but do not affect the fundamental problems of the AI.

In the 1930s, mathematical logic Kurt Gödel and Alan Turing found that there are no algorithms that would guarantee the solution of all tasks in some important mathematical regions. For example, answers to questions in Spirit: "Is the proposal of the first order of the theorem" or "whether the polynomial equation has integer solutions in other variables." Since people are able to solve the tasks of this kind, this fact was proposed as an argument in favor of the fact that computers are inherently unable to do what people do. Roger Penrose says this. However, people cannot guarantee solutionsarbitrary Tasks in these areas.

In the 1960s, scientific programmers, among whom were Steve Cook and Richard Carp, developed the theory of NP-complete tasks. Tasks in these areas are solvable, but, apparently, their solution requires the time growing exponentially with the dimensionality of the problem. The simplest example of the NP-complete task area is the question: what statements logic statements are fulfilled? People often solve problems in the field of NP full tasks at times faster than it is guaranteed by the main algorithms, but cannot solve them quickly in the general case.

For AI, it is important that when solving tasks algorithms were the same effective as human mind. The definition of subdomains in which there are good algorithms, is important, but many programs that decisive AI tasks are not related to easily identifiable sublicas.

The theory of the complexity of common classes of tasks is called computational complexity. Until now, this theory has not interacted with AI as much as possible to hope. Success in solving problems with people and programs AI apparently depends on the properties of tasks and methods for solving problems that neither researchers of complexity, nor the AI \u200b\u200bcommunity can determine exactly.

Also relevant is the theory of algorithmic complexity, designed independently of each other Solomonov, Kolmogorov and Chaytti. It defines the complexity of a character object as the shortest program that can generate it. Proof of the fact that the candidate program is the most short or close to that, is an intractable task, but the presentation of objects by generating short programs can sometimes clarify the situation, even if you cannot prove that your program is the shortest.