The term "Artificial Intelligence" has been used for 70 years, but everyone understands it in different ways. What is AI really?

Artificial intelligence (AI) as of 2023 is one of the most popular, discussed and both controversial and ambiguous terms.

Indeed, on the one hand, numerous experts argue that the level of development of artificial intelligence technologies in the country is the most important criterion for the technological, economic and military-strategic power of the state, an indicator of competitiveness in high-tech markets. Analysts report that tens of thousands of companies are using AI technologies, thousands of startups around the world are developing solutions based on artificial intelligence, and IT giants are competing for the opportunity to buy the most successful developers in this area. The artificial intelligence market is valued by consulting companies at hundreds of billions of dollars. On the other hand, a number of experts say that AI is a myth, no artificial intelligence has yet been created! Moreover, such statements can be heard not only in private conversations. For example, in an article titled "AI doesn't really exist yet," the authors begin their narrative with the categorical statement: "many companies that claim to use AI are deceiving themselves - and so are you."

The authors of the aforementioned publication, citing Julia Julia, vice president Samsung of innovation and co-author, digital assistant Siri cite her following quote: "Today's tools for business use mathematics, statistics, machine learning deep learning, big data getting machines better than in the past. But what is so often called AI is not really related to an artificial form of intelligence. "

Many authors noted that the phrase "Artificial Intelligence" is an unsuccessful term, which nevertheless became common. A number of experts offered their own clarifications of this name, recall, for example, the remark of Professor Konstantin Vorontsov, according to which it would be more logical to decipher AI as an "imitation of intelligence."

The lack of a generally accepted definition complicates the solution of a number of legal and economic issues that were not on the agenda before.

Today, AI defacto enters an area where the "foot of the car did not step" before. How we interpret the terms, including the delineation of legal responsibility in such difficult issues as who is responsible for the accident caused by the unmanned vehicle control system, who is responsible for the decision, if the diagnosis to the patient was mistakenly made by the AI system, who owns the authorship of the melody generated by AI - the list can be continued.

Why is the term that people have been using for almost 70 years still understood by different groups of specialists in different ways? What AI definitions exist and how do they differ?

Why defining AI is so difficult

Here are some of the challenges facing specialists trying to define AI. First, AI is rapidly evolving, more new solutions are emerging, the content of the term is changing, and it is difficult to give an interpretation that is flexible enough to encompass ever new approaches to AI implementation.

Secondly, it should be noted that AI relies on studies of different sciences, both natural and human sciences (computer sciences, statistics, psychology, neuroscience, philosophy), each of which has its own conceptual apparatus, its own views on the subject, and these views do not always coincide. Some philosophers believe that consciousness can exist only within the framework of wildlife, others believe that intelligence and the presence of will, thirst for knowledge and even love are all attributes that AI can acquire at an operatively divided stage of its development.

The contradictions that exist in different disciplines also find their refraction in the topic of AI. Representatives of different schools and teachings (materialists and idealists, supporters of evolutionary theory and its opponents, etc.) cannot reach consensus in the interpretation of the term AI, especially in issues affecting the philosophical aspects of being.

From a philosophical, epistemological point of view, the term intelligence is tied to a biological object, and the question of whether non-living objects can have a feeling (as the primary basis of the structure of consciousness and intelligence) is not resolved. Therefore, most neuroscientists are trying to postpone the question of whether a machine can have intelligence and, accordingly, intelligence for the future.

Different community members set themselves different challenges. For some, AI research is associated with highly scientific goals, for example, how to understand the mechanisms of human intelligence using computer modeling. For others, the goal is to create smart machines that will solve new scientific and practical problems that are not subject to humans today. For others, the task is formulated as a business and consists in using new AI technologies wherever their use carries economic benefits.

Thirdly, the term AI can be called developments of a very different level of complexity, based on different directions of artificial intelligence, different technologies implemented both in a virtual environment and embodied in a physical device (in the form of smart robotics).

Evolution of AI definitions

It is very difficult to give an unambiguous comprehensive and consistent definition of AI, and therefore there are hundreds of private definitions. Let's try to figure out what they have in common and how they differ.

In Table 1.1, we presented ten AI definitions proposed by different authors over the past 50 years and identified a key part of the definition.

Difference in emphasis in the definition of the term "artificial intelligence"

It is interesting to note that most of the authors mentioned in Table 1.1 are based on the statement that AI is some area of ​ ​ human activity, for example, the field of science or engineering, while another part of experts emphasizes that AI is a certain property of machines (meanwhile, when the question of the definition of what human intelligence is, the definition primarily affects the questions of precisely human abilities determined by mental intellectual activity _.

Note that these are fundamentally different approaches. Indeed, AI, as a science, can deal with purely theoretical issues. The answer to the question "What is AI as science?" Assumes primarily the designation of a list of research areas, scientific schools, directions, etc.

The set of publications over a period of time determines the range of topics that scientists are working on, which focus on one or another scientific problem. In particular, analyzing the share of articles in the field of various disciplines, one can judge the change in the focus of scientific research in the field of artificial intelligence at different periods of time (see Figure 1.1).

Fig. 1.1. Proportion of arXiv preprints presented in the main Al categories by category. Source: arXiv

For example, as can be seen from the figure, in 1998 almost 75% of all publications were in computer linguistics, and after 20 years only 15% of articles from the mentioned source were assigned to it. The category of "computer vision and pattern recognition," which was in its infancy in the early 20th century, grew many times and took a dominant position in 2018.

As we noted, in addition to the definitions of AI associated with scientific activity (Table 1.1), there are formulations that proceed from the fact that AI is interpreted as a set of properties, abilities of intelligent machines. Here the question is raised in a more practical plane, that is, in this case, first of all, we are talking about the implemented machine intelligence in the form of programs running on this or that hardware and showing certain abilities of artificial intelligence.

If science creates the theoretical foundations of the products of the future, then, speaking of the properties of AI, we are forced to more clearly distinguish between the set of capabilities of intelligent machines that exist at this point in time and those that can theoretically be created in the future.

Why modern AI is weak

Approaching the fork between the concepts of "existing AI" and "hypothetical AI," it is necessary to mention the division of artificial intelligence into the so-called strong or general AI (AGI, Artificial General Intelligence) and narrow, applied or weak AI (ANI, Artificial Narrow Intelligence).

Examples of projects on the basis of which we argue how exactly weak AI differs from strong AI are constantly changing as we gain wider opportunities as part of new implementations of artificial intelligence systems. Relatively recently, speaking of the difference between the first and second, the authors pointed out that narrow (weak) AI is an intelligence capable of performing a specific intellectual function, or their limited set is sometimes better than a person. At the same time, exceeding the capabilities of a person in a narrow area, such a system, as a rule, does not have intellectual abilities in other areas, unlike a person who is trained to solve problems in various fields.

However, in 2023 - with the advent of multimodal ^[2] - this difference does not sound so clear. Indeed, AI still does not have the ability to learn in as wide a range of areas as humans. But, for example, the multimodal neural network Gato (from Isp. Cat) is able to perform 604 types of tasks, including creating descriptions for images, conducting a dialogue, laying blocks using roborooka, playing arcade games, while performing more than half of these tasks better than the average person.

The emergence of each new promising opportunity gives enthusiasts a reason to say that another milestone has been passed on the way to creating strong AI. At the same time, skeptics continue to say that AI will never have the breadth of a person's cognitive abilities, will not have emotions, self-awareness and goal-setting.

These reasoning is often justified by the fact that a machine commensurate in intelligence with a person must have goals and motives similar to those of a person. The goals and motives of human behavior depend on many factors. Some are conditioned by our instincts, such as wanting sex, food and shelter. A person's behavior and goals are influenced by emotions - such as fear, anger, or jealousy.

Some of our goals and motives are of a social nature, for example, people are exposed to collective concepts such as "successful person."

"A revolutionary turn in the understanding that only the brain can be individual, and the mind is an exclusively collective phenomenon, will radically change the trajectory of research on general (strong) artificial intelligence. His researchers will have to find new, fundamentally different architectures and algorithms, focused not only on neural, but on sociocognitive hypersets "(quote source).

Within the terminology described above, all existing AI systems today are classified as weak AI. It should be noted that the terms "weak" and "narrow" reflect different aspects of the concept. The word "weak" is not entirely appropriate when we, for example, talk about programs that beat a person in one game or another, the term "narrow" is more suitable here, where AI does not reach the level of a person, the term "weak" is more appropriate.

Is strong AI possible

Unlike weak, strong AI, or general-purpose intelligence, is a hypothetical machine of the future that will be able to understand the world at a level comparable to a person's level of understanding of it, and learn to perform the full range of intellectual tasks that a person can perform (in addition to the concept of human-level AI, they also use the concept of "super intelligence" or "superintendent," which, by the way, will not necessarily have a social structure and be anthropomorphic).

Such a division helps to terminologically separate two different concepts and reveal one of the main reasons for the lack of consensus in the definition of AI. Indeed, some specialists (more humanitarians) tend to perceive the term AI as a strong AI, as an intellectual machine endowed with the properties of consciousness, awareness, thinking, sensual perception of reality, expression of will. Another part of the community, engaged in the development of AI technologies and various kinds of applications based on it, is used to calling artificial intelligence exactly what they are dealing with (that is, weak AI). And in this regard, the understanding of AI solutions under the term "weak AI" does not cause them to be rejected. Despite the fact that "strong AI" and "weak AI" significantly drown the concept, people mainly use the term AI, suggesting that, depending on the context, it is clear what is at stake.

Despite the fact that we noted the dignity of dividing into strong and weak AI, this approach has some limitations: in particular, we cannot, within the framework of this terminology, reflect the process of "strengthening" (developing) weak AI. Obviously, those solutions that we classify as weak AI are improving are becoming more intelligent, but until AI is equal in capabilities to human, it will remain weak by definition. In this case, researchers engaged in the construction of strong AI will receive solutions in the category of "weak AI" at all intermediate stages within the framework of the described terminology.

You cannot build a strong AI by gradually increasing the power of known solutions based on weak AI and integrating them with each other. To build strong AI, you need a qualitative leap, a new architecture. As they say, "you cannot reach the moon by transplanting into increasingly tall trees."

Stressing that strong AI is a speculative construct of the future, the importance of this concept should not be underestimated. It is wrong to believe that this is a term that only science fiction writers operate on. The concept of "strong AI" is a kind of goal, an ideal model that developers can focus on in order to create intelligence that will become an assistant to a person in solving any problems. The question "Is the task of building a strong AI achievable?," Apparently, will remain the subject of debate among specialists until such a system is built. At the same time, it is obvious that if such intelligence is created, then due to the exponential development of computer technologies, on the basis of which, we believe, it will be created, and with insignificant improvement in time of human intelligence, strong AI should soon overtake human intelligence in capabilities, turning into the so-called "superintendent" (Fig. 1.2).

Fig. 1.2. Forecasts of the growth of the level of intelligence of machines and humans. Source: www.scoro.com

One scientist, the idea that machines would once reach the level of human intelligence, seems ^[3] Others are confident in the onset of the hour "x," when the intellects of a person and a car will equal, and even name approximate dates of the event, starting from which cars will become smarter than a person in all respects. For example, Raymond Kurzweil called 2045 as the most likely date for the emergence of the superintendent.

Predictions of the transition to the stage of super-intelligence can also be found in the works of Nick Bostrom, who predicts a sharp acceleration in the growth of AI capabilities after reaching human level (Figure 1.3.).

Fig. 1.3. Forecast of the dynamics of the growth of the level of intelligence of machines. Source: Nick Bostrom

The question of the attainability of strong intelligence remains open. However, humanity knows many statements about the impossibility of a particular technological innovation. Philosophers like Dreyfus once argued that "computers can never play chess" (Dreyfus, 1972).

It should also be mentioned such an important function of artificial intelligence as a person's knowledge of himself. Indeed, as we create a computational version of an intelligent system in a computer, we begin to understand the nature of our own intelligence much better.

Note also that strong intelligence, like the superintendent, does not have a clearly defined task, and, as human intelligence, forms goals and ways to achieve them as it develops. At the same time, the superintendent may have a value system within which clearly defined tasks aimed at achieving these values ​ ​ can be set by the agent himself.

By purpose, strong artificial intelligence is positioned as a universal means of solving the urgent problems of mankind, that is, it does not have specific predetermined goals, unlike weak intelligence. And since the urgent tasks of mankind largely depend on who sets them, this gives rise to various kinds of distrust of society about the humanistic orientation of such AI. Which further divides the concepts of strong and weak AI, puts on the agenda new issues from the field of ethics and security of communication with intelligent machines. The question has two sides - whether a person can solve issues of ethics and security when communicating with AI and whether a person can create an AI that will help solve issues from the field of ethics and security of communication between people.

How intelligence relates artificial and living

Some authors tend to believe that at some stage there will be a physical fusion of the substrate of human intelligence (human brain) and machine. As of 2023, it is impossible to imagine the physical integration of two foreign bodies - a silicon chip and brain neurons. However, technologies for chips implanted in the brain are already being developed, for example, to restore vision. No matter how fantastic the project of merging the machine and the person looks, there is a weighty foundation under it. A healthy level of egocentrism encourages people to use the machine to prolong and expand their own capabilities and range of sensations, perhaps even more actively than they seek to create a certain subject, more perfect than the person himself.

Commenting on the reasons for the presence of contradictions in the interpretation of the term AI, it should also be noted that they can be found in human psychology - it has long been observed that people tend to deny the presence of intelligence in the behavior of machines, after they learn exactly how the mechanism of "intellectuality" of their decision is implemented.

This trend can be traced in Figure 1.4, which shows four groups of artificial intelligence technologies from mature developments (long mastered by a mass user) to promising ones.

Fig.1.4. Examples of AI applications distributed by the degree of complexity of the solved problems and the level of maturity of solutions. Source: Lux Research

Indeed, tasks such as optical character recognition, spam filtering, production defect control are increasingly less associated with artificial intelligence, while prospective developments at an earlier stage of research, such as automatic software development or personal robots, can more often claim in the mass consciousness to be examples of artificial intelligence.

Artificial intelligence is endowed with a collection of abilities that connect with the front edge of computer science and beyond. It is no coincidence that the well-known expression "Artificial intelligence is something that has not yet been done," which received the conditional name "Tesler's Theorem," appeared.

In support of the thesis, we can cite a curious dialogue, known since the creation of pioneer AI programs, in which an expert is asked if a machine is intelligent that can read a newspaper and do a "squeeze of events," offering the reader only the most important theses. After a positive response from the expert, he is invited to answer the same question again, explaining to him that the program only chooses from the text the headings in bold... This comic story reflects a person's tendency to identify the concept of artificial intelligence with abilities beyond the already known algorithms.

Another dilemma follows from the example considered: what do we consider the criterion for the intellectuality of a machine - its ability to fulfill a given task or how the mechanism for solving this problem is implemented?

We also note that it is very difficult to define AI, which is created as a semblance of natural intelligence, without knowing the full picture of the mechanisms of work, the exact localization of the latter, and even without an unambiguous definition of the term "human intelligence" itself. The difficulty of reaching consensus in determining AI is also explained by the lack of such in the interpretation of the concept of "human intelligence."

Indeed, you can find a lot of different interpretations of the concept of "human intelligence." The authors of one of the studies admit that... "it seems that there are almost as many definitions of intelligence as experts who were asked to define this term." So, according to the interpretation of the explanatory dictionary of the Russian language edited by D.N. Ushakov, "intelligence is the mind, reason, thinking ability of a person (as opposed to will and feelings)." And, according to the definition of the AllWords Dictionary, 2006, this is "the ability to use memory, knowledge, experience, understanding, reasoning, imagination and judgment to solve problems and adapt to new situations." According to The American Heritage Dictionary 2000, this is the ability to acquire and apply knowledge.

One of the specialists in the field of AI - the CEO of the Deep Learning Partnership - in his report "Towards a general theory of intelligence," comparing artificial and living intelligence, suggests starting from the interpretation of human intelligence as a combination of its different abilities (Fig. 1.5), and analyze which of them are available to the machine and to what extent.

Fig. 1.5. Types of human intellectual abilities according to Mark Vital

Table 1.2 presents the estimated characteristics of AI capabilities at the time of publication (2019) by how much AI approached human intelligence in various parameters.

Table 1.2. Assessing the capabilities of AI in comparison with human intelligence ^[4]

Note that if in Fig. 1.5 we are talking about the abilities of a certain averaged person, then in the case of Table 1.2 the total abilities of the machines created today are implied, and not the properties of a particular AI. This is obvious, at least, due to the fact that so far not a single project has been implemented by a person in which the created AI would simultaneously possess all of the listed abilities indicated in Table 1.2.

As follows from the example described above, in different contexts, the authors, arguing about what AI is, use this concept, both characterizing the totality of the abilities of various AI machines, and as the abilities of a particular AI implementation. Indeed, when we say that AI has surpassed a person in the game of chess, we mean a specific program, and when it comes to, for example, that AI will displace a person in many non-creative professions to a certain extent, we talk about AI as a combination of technologies.

Note that the abilities of the average person or the abilities of a person when we talk about a person (as "humanity"), acting as the author of all the achievements of creative thought, will differ significantly.

With regard to artificial intelligence, these differences are even more striking. Unlike people who have a common physical structure of the body and a set of basic abilities, machines endowed with intelligence can act as a variety of objects that are not comparable in capabilities.

By AI, we can mean relatively simple programs operating in the virtual world, and the most complex robots embodied in an artificial device that have a lot of intelligent subsystems (or an entire network of coupled devices). Moreover, intelligent functions can be secondary, built into a variety of machines, including those not positioned as intelligent.

Therefore, the concept of AI as the average intelligence of an average machine simply does not exist.

Interpretations and classifications of the concept of AI

Summarizing the above approaches to AI interpretation, we present them in the form of a diagram (Fig. 1.6).

Fig. 1.6. Examples of the use of the term AI in different contexts. Source: Authors

Here, along the abscissa axis, the stages of development of any project that develops from the stage of scientific theory to the practice of its technological implementation and is further implemented as a product based on this technology are postponed. The ordinate axis indicates a time reflecting a longer historical process - the process of developing AI as a science and industry.

In these coordinates, you can arrange a number of concepts that in different contexts can be called the term AI and denote different things. For example, AI as a scientific direction, AI as a collection of technologies, AI as the sum of the abilities of available AI products at the moment in time, AI as the abilities of a specific intellectual machine, AI as the totality of the abilities of the intelligence of the future, when it equals or surpasses human intellectual abilities.

The figure clearly shows that AI is an umbrella term that describes a set of different kinds of solutions that differ significantly in technology, complexity, implementation stage, purpose, etc. That is, we will emphasize once again that, depending on the context, we can mean different things by AI.

The classification of definitions allows us to analyze the reasons for the multiplicity of definitions and understand the content of the term. Consider one such classification, first proposed by Stuart Russell and Peter Norvig. The authors note that various definitions of artificial intelligence rely on one of four abilities: act as a person, think as a person, think rationally and act rationally. Examples of definitions in each category are given in Table 1.3.

The interpretations at the top of the table are related to the thought process, at the bottom - to behavior. The left part comes from comparison with man, the right - from the principles of rationality. Each of the four directions sets its own field of research.

Table 1.3. Classification of AI definitions by Stuart Russell and Peter Norwig

The definition of AI, repelled by the criterion of "the ability to think like a person," sets the direction in which scientists try to build AI by studying the structure and mechanisms of the human brain, analyzing a person's thought process.

The criterion "ability to act as a person" essentially became the basis of ^[6] and his like, the basic principle of which can be expressed by the phrase "in order to be considered intellectual, the program must demonstrate behavior indistinguishable from human actions."

The definition, which takes as a basis the ability to "think rationally," corresponds to the direction associated with the creation of computer programs that could reason logically. And the definition, which comes from declaring the ability of machines to "act rationally," assumes the possibility of creating intelligent agents that can optimally achieve (under given restrictions) the set goal (for example, a business goal).

The authors of said classification indicate that an important characteristic that should be reflected in the definition of the term AI is the autonomy of this intelligent machine. It is this property of AI machines that will allow them not only to find solutions to the problems that a person poses, but also to identify, formulate and solve problems whose existence is unknown to a person.

The classification of AI systems by degree of autonomy and adaptability can be found in the PWC report (Table 1.4). The approach offers three types of AI in terms of autonomy and adaptability and separates the concept of AI from the concept of "automation."

Table 1.4. Classification of AI systems by degree of adaptability and autonomy. Source: PWC

The issue of AI autonomy is closely related to the need to control the work of artificial intelligence and determine the permissible risks when delegating certain tasks to AI systems. In this regard, the classification of AI systems by the degree of autonomy and the need to control the latter is of some ^[7].

The limitation of the autonomy of a number of intelligent machines is explained by the fact that fully autonomous machines can make an irreparable mistake. And the more serious the possible consequences of this error, the more a person seeks to retain control over decision-making (Fig. 1.7).

Fig. 1.7. Where modern AI systems can be more efficient than humans. Source: Adapted from Massachusetts Institute of Technology

In the upper left corner there is an area where cars are more efficient than a person. Here, a person only slightly complements and adjusts the car. For example, in spam filtering or keyword selection. These are areas where the machine is effective, and the possible error in terms of the level of negative consequences is small. On the contrary, in the lower right corner there is an area where making a decision without a person is impossible. Here a person only half-senses the tip of AI, and the decision is made by himself. As, for example, when making medical diagnoses.

Speaking about the development of AI systems and using the designations of Figure 1.7, we can say that the zone where "a person complements the machine" will continue to grow, and the zone where "a machine complements a person" will continue to shrink, which to some extent can be called the process of gradually displacing a person from the field of systems for solving intellectual problems and making decisions.

AI systems can be implemented in the form of complex humanoid robots, such as, for example, in the ATLAS project of Boston Dynamics, however, this is a very narrow direction.

There are a huge number of AI solutions for implementing practical tasks (increasing labor productivity, improving the quality and degree of customization of various kinds of products and services, etc.). Therefore, the introduction of AI acts rather as a multifaceted process of intellectualization of the artificial environment surrounding a person, which is becoming more connected, more intellectual, more convenient for a person in terms of the concept of "convenience of human functioning" in a particular society at a certain point in its development.

In some ways, the implementation of AI in an enterprise can be considered as one of the elements of digitalization, in which more and more complex tasks are transferred to the computer - for example, recognizing goods on shelves or analyzing incoming letters from support services.

Prospects for weak and strong AI

Existing AI-based solutions to date are incapable of doing many things that seem simple to four-year-olds but can compete with the best in narrow tasks. The fact that in many respects, including in terms of universality, AI lags far behind human capabilities does not diminish the importance of the successes achieved on the basis of AI. Despite the fact that we operate with the concept of "narrow intelligence," this "narrow intelligence," penetrating numerous applications, becomes a wide phenomenon and is already interpreted by many analysts as a stage in the evolution of data processing processes, in which human participation is minimized (transferred to tasks not mastered by AI means) (Fig. 1.8).

"partially [[Image:Скриншот 20-03-2023 201708.jpg|840px|thumb|Fig. 1.8. Evolution of the data process. Source: Deloitte Research ^[8]

Obviously, the final part of the diagram in Figure 1.8 is a picture of the future, which is feasible only when approaching the concept of strong AI. Here lies another contradiction in the views on the goals and prospects of the development of AI. Scientists who aim to build strong AI talk about the need to find new approaches in AI construction, noting that the extensive path of development of weak (narrow) AI is a lateral, if not dead-end, path of direction development. At the same time, the percentage of scientists involved in projects to build strong AI is significantly inferior in number to groups of developers engaged in scaling and introducing existing AI technologies into all new solutions in various sectors of the economy. Therefore, we can say that the purpose of building strong intelligence, as a phenomenon that justifies the original concept of the term AI, in most projects is replaced by a more obvious goal - maximum commercialization of narrow AI.

Obviously, most investments today are related to projects that pay off in the foreseeable future (Chart 1.9) and are aimed mainly at expanding the number of developments in the field of narrow (weak) AI. At the same time, it is also obvious that projects to create a common (strong) intelligence require much greater investment in development and imply the emergence of products in a much more distant future (in the absence of consensus on the prospects and achievability of applications of the level of strong AI).

Fig. 1.9. Assessment of the dynamics of costs and revenues in projects based on weak and strong AI. Source: Authors

The development of solutions based on narrow AI, although it is a less ambitious task, nevertheless requires huge resources. Therefore, no matter how close existing AI solutions come to the concept of strong AI, they represent the most advanced edge of engineering. Modern AI technology, requiring high-performance computing, powerful IT infrastructure, advanced scientific developments and related personnel, defacto has begun to act as some measure of the technological maturity of companies, industries and entire states.

Many experts agree that AI is the locomotive of the last wave of convergence of the most complex complementary DARQ technologies (an abbreviation for four technologies: Distributed ledger technology, Artificial intelligence, Extended reality and Quality computing () quantum computing (Fig. 1.10), which lead humanity into the digital world.

Fig. 1.10. Stages of convergence of infocommunication technologies.

Chinese experts attach even greater importance to the role of AI and in one of the studies (Fig. 1.11) represent AI as the pinnacle of evolution not only of infocommunication, but also of technological progress, implying that AI, as a technology, incorporates and is based on the totality of technological experience of mankind.

Fig. 1.11. AI as the pinnacle of humanity's technological experience. Source: CAICT

We reflected different accents and different approaches in the interpretation of the term AI. The complexity of the unambiguous definition of the concept of AI and its development paths gives rise to problems in the definition of derived terms such as "AI-solution," "AI-project," "AI-company," "AI-market." Which, in turn, leads to difficulties in trying to rating AI projects, AI companies and, most importantly, in quantifying the artificial intelligence market due to the fact that different authors include a different set of technologies and services in the AI project.

Commenting on the question why there is no generally accepted interpretation of the boundaries of the AI project, let's give an analogy - we will try to answer the question of where a person's intelligence is localized and how to distinguish the system that forms it. The most common response is the neocortex region (the new cortex are new areas of the cortex that in lower mammals are only outlined, and in humans make up the bulk of the cortex), which is responsible for human intellectual activity. A broader view of the problem leads to the idea that other, more ancient parts of the brain cannot be discarded when talking about the work of intelligence. Further, it becomes obvious that vision must be included in the system under consideration, as "the part of the brain placed on the periphery," after which you have to remember the hearing organs, taste receptors that convey information to the brain, without which it is impossible to form ideas about the outside world - the basis of intelligence. Adding tactile sensations to the sensors, we come to the conclusion that consideration of intelligence is impossible in isolation from the human body. And recognizing that intelligence is a product of human communication in society, it is necessary to further expand the field of its formation. In the case of artificial intelligence, the boundaries of the phenomenon are even more blurred. How much of an intelligent machine is artificial intelligence?

Is the processor that participates in the calculation process part of AI? And the server? And if we are talking about a cloud in which calculations are made? These questions, which at first glance seem to be secondary, become very relevant when we try to determine, and then quantify such concepts as the AI market.

At the same time, it should be noted that the lack of global consensus on the definition of AI does not interfere with the existence of consensus within individual technologies, individual communities, refined context, or yet - within the framework ^[9], adopted in a particular analytical company.

In various contexts, the term AI acquires its content - as a concept, as a field of science, as a marketing term, as a product of IT companies, as a product of software companies, as a product of vendors producing chips, as a measure of the development of a technology company or even an industry, as a character in works of fantastic literature, etc.

After reviewing the problems in the definition of the term AI, it would be wrong not to give a definition that seems less contradictory to us.

AI is an umbrella term that, depending on context, can define a number of concepts, including such as the field of science, the field of engineering, technology and machines, and the abilities of these machines to solve problems and perform actions that were performed only on the basis of human intelligence before the creation of AI. And also actions that were not performed at all on the basis of natural intelligence. For example, designing the structure of proteins or predicting particles in a hadron collider.

Notes