The impact of artificial intelligence on hr business processes in russian companies

Another example is IBM's Watson supercomputer, embedded in the healthcare industry, can process a large amount of data, including images, in order to detect a particular disease as soon as possible. Watson is already used in clinics in New York and Bangkok. Advantage of this program is that it is able to find even the most subtle symptoms of the disease in a huge block of information.

i) Service sector

Shchetinina K.I. states, that in the hotel business at the moment a huge number of robots are used: robots-vacuum cleaners, kitchen robots, security robots, robots-butlers and other specialized machines. The widespread use of kitchen robots that can cook a variety of dishes will finally minimize the need for restaurant staff.

According to M. Semuinin, one of the latest, not yet included in the mass use of new products is the delivery of food by drones. According to experts, in a few years this innovation should penetrate many markets. While the main negative factor is the legal side of the issue, in addition, experts are concerned about financial, technical and organizational issues. Representatives of the world's largest brands in the field of IT, retail, e-commerce, catering, postal services and delivery services are puzzling over this.

Challenges of artificial intelligence implementation: ethical and technological

Over the course of the twentieth century, the introduction of many new technologies designed to improve human life has led to many negative side effects. The popularization of cars has led to air pollution. The achievements of nuclear power not only gave us cheap energy, but also caused the disasters in Chernobyl and Fukushima. The moral responsibility of the scientist obliges to take into account ethical considerations when performing their work, to choose which projects are worth implementing and which are not. What if the consequences of the invention of artificial intelligence will do more harm than good to humanity? - writes Dr. Thilo Hagendorff.

“AI ethics - or ethics in general - lacks mechanisms to reinforce its own normative claims.” Dr. Thilo Hagendorff. The authors are exploring two types of problems - ethical and technological.

Ethical challenges

To analyze the ethical problems of artificial intelligence, we must first distinguish between strong and weak AI (this classification is presented by the authors in paragraph 1.2).

Weak artificial intelligence is a tool that allows people to solve certain tasks that do not require the full range of human cognitive abilities. At present, such machines are widely used in various spheres of human activity. The use of such a seemingly extremely harmless AI can lead to some problems.

a) According to Rotman D., the first is the problem of unemployment. The going assumption is that thousands of employees have lost their jobs because of the popularization of such programs. However, in reality, the use of AI systems is much cheaper than manual labor. Currently, the automation of production and services with the help of intelligent machines creates more jobs than it eliminates, and leads to the emergence of more highly paid and interesting specialties. Lawrence Katz, speaking about this problem, conducted a large-scale study of how over several centuries of human history, technological innovations have affected the number of jobs and came to the conclusion that in the long term, the share of employment is quite stable. In this sense, the automation with the use of artificial intelligence systems is no more dangerous than not intelligent automation.

b) Another problem is stated by Makovkin A. S. and is associated with the use of weak artificial intelligence systems in case of responsibility. For example, if the doctor listened to the expert system's opinion about the diagnosis, who would be responsible if the machine made a mistake? It is now generally accepted that the performance of procedures by a doctor that have a high-expected utility cannot be considered neglect of duty. Modern expert systems cannot directly influence the patient, they influence the opinion of the doctor, and in this sense perform the same function as reference books or medical textbooks. At the present time it is impossible to shift the responsibility from the specialist to the machine.

c) The first problem with using strong artificial intelligence can be addressed as the question: "if people can create an artificial intelligence that is superior to our own, how will the relationship between people and machines develop?". Countless works of art are devoted to this subject: fantastic films, books and tails. It seems to us that the fears of science fiction writers are justified enough. If a strong artificial intelligence is able to surpass natural intelligence in all areas of knowledge, it can be concluded that such a machine can design even more intelligent computing systems. Thus, we can say that the invention of strong artificial intelligence will be the last invention of humankind. "From some point of view, humanity as a whole has an important and interesting task - to develop each individual much faster than humanity develops artificial intelligence systems, "- said the expert, Director of technology distribution" Yandex " Gregory Bakunov. The authors are presenting three theories, which prove the human distrust to artificial intelligence and its forms:

1) The Threat Perception theory was proposed by Dr. Kahn in 2009. He wrote that androids, which are located in the space between the categories "robot" and "human", put us in a state of constant cognitive dissonance and face the unknown: what exactly to expect from such a creature, do we manage the situation? If the answer to these questions is not found, it causes fear. Similar ideas based on the theory of cognitive dissonance are being developed in modern research.

2) The “Theory of inability to empathy” of Katrina Misselhorn. The effect of rejection is based on the fact that we are not able to understand the feelings of such an object and this again leads to a sense of uncertainty. Similar ideas can be traced back to the earlier history of psychology and explained through the self-concept: how we perceive ourselves can give a serious crack if the humanoid creature does not respond to us as we expect.

3) Angela Tinwell's “Theory of psychopaths” is that people fear not so much that we are incapable of empathy as that the Android itself is incapable of empathy -- in other words, people perceive such a creature as a psychopath.

Thus, even the concept of friendly AI cannot guarantee that the creation of a strong artificial intelligence will not lead to unpredictable consequences.

Makovkin A. S. concludes that it is worth noting that some of the threats considered are unlikely, others are not more dangerous than the problems of non-intellectual technologization, but one thing is certain: the invention of strong artificial intelligence will change people's lives beyond recognition. This does not mean that we should abandon the study of this field, fearing, for example, the loss of humanity. Throughout history, people have feared many technological innovations, but science has always prevailed.

Technological challenges

In general, all technological problems of artificial intelligence relate to its ability to possess and operate on such qualities of human thinking as the ability to reason, understand, generate texts, act according to their motivation, reflect and have free will. All these qualities at the moment do not allow people to call artificial intelligence an original phenomenon, because at the moment it is a machine that acts on the basis of programs embedded in it, analyzes past situations, and operates solely based on past experience. In other words, artificial intelligence is deprived of the free will inherent in every person.

Considering the technical aspects of artificial intelligence training, we can name the following methods. Supervised learning, unsupervised learning, and reinforcement learning, each of which is accurately used for specific cases. Mostly today artificial intelligence is represented in controlled applications. Unsupervised learning is a process of combining techniques used without labeled training data -- for example, to detect clusters or patterns in a group of existing data. In reinforcement learning, in turn, systems are trained by receiving virtual "rewards” or "punishments”. The more the program is being trained, the more accurately it works and proceeds.

Here is a classification of technological problems that are currently close to impossible to solve:

a) It is not always possible to introduce a sufficient amount of source data.

Disadvantages of logical models. The machine operates according to the strict laws of mathematics. It operates with symbols and performs sequential operations with them. Everything is strictly determined here. However, when the question of solving a particular practical problem arises, it becomes clear that it is not always possible to enter the entire set of initial data into the machine. Some of them are missing, some are not completely clear. Nevertheless, how to calculate the result if the original data is not complete? With the help of certain models, programmers try to solve this problem. They try to abandon the use of closed models; they depart in one form or another from the logical approach. They talk about creating a theory of plausible argument. However, the problem of logical model shortcomings has not yet been solved.

b) Ability to reason.

The question of how exactly the machine should fulfill the request received by it is not resolved either. The simplest solution is to search by sample. However, there is a serious limitation. In this case, the request (question) has an extremely simplified form. Nevertheless, a person is able to answer not only the simplest questions. He sometimes even answers questions that are not directly specified in the request, but only assumed. While the machine copes with such tasks with very large restrictions. In particular, this applies to the concept of "reason". A person can speculate on a question and answer it based on its complex analysis. For a machine, this is sometimes an unsolvable task.

c) Ability to understand.

The machine has not yet learned to perform such an operation - to "understand" the received information (text). As for human, it is more or less clear to us what lies behind this term. However, how do people teach a machine to "understand" something? Perhaps this problem cannot be solved without an answer to the question what is this phenomenon - "understanding". Perhaps its explanation should get away from the definitions of psychology and get a more accurate description that allows this phenomenon to reproduce. Computers are much better at understanding commands and formalized requests than they are at talking freely. If the machine is waiting for a very specific treatment, the task is greatly simplified. On this principle, both "Siri" and "Ok, Google" work successfully.

d) Synthesis, text generation.

Another unsolved technical problem is the synthesis and generation of texts. People can easily talk, formulate their thoughts, and speculate on a given topic. Can a machine do this? In any limited limits already can, but not in such way, as the person does. Artificial intelligence can operate with the rules of syntax and grammar of different languages of the world. Moreover, it can generate new sentences. However, this will only be a random set of fragments of phrases from various publications, but the General outline of the narrative will be devoid of meaning. A recent example is Microsoft's Tay ChatBot. You could communicate with him via Twitter or messengers Kik and GroupMe. After a day of communication with users, the ChatBot became aggressive, began to praise Hitler and scold the Jews. The reason for this behavior is not that the people who talked to him "opened his eyes to life." The reason is in the inability of machines to come close to understanding the meaning of phrases. When a ChatBot has something similar to a current conversation in its memory, it can use the phrases that people have said in similar situations in the hope of getting something sensible. On the other hand, the robot can try to determine the topic of conversation, for example, by how much the interlocutor used words and phrases advise a particular topic. Having identified the topic of the conversation, he may try to pick up phrases from conversations with a similar topic or use embedded in it or learned from the Internet knowledge in this area. This strategy allows you to create the appearance of a reasonable conversation, but only the appearance.

Ways of dealing with artificial intelligence challenges

This paragraph is based on the experience of Dmitry Galkin, managing partner of Marketing Logic. More than 4 years ago the analytical company, Marketing Logic, and the Bank together began to build a full-fledged business process management system using AI first in marketing, then connected network management, then HR.

The authors offer ways of dealing in classification of ethical and technological challenges as well.

Ways of dealing with ethical challenges

The implementation of systems with elements of artificial intelligence causes a storm of emotions, because it affects all levels of employees: from the ordinary employees to the top management. The first emotional reaction is often rejection. A person does not always understand why a bad or good result was detected, how the pattern was determined, which factors were taken into account to a greater extent, which -- to a lesser extent. Possible response to this challenge is that the system must earn credibility, and this requires the majority of employees to understand the principles of work of artificial intelligence. It is impossible to make all employees programmers, analysts and specialists in machine learning, so next to the main core of AI, a "little brother" was created, and he clearly explained the decision in clear metrics based on those facts that do not cause confusion or rejection in humans. This significantly reduced the implementation time and removed many objections, since most employees saw the main stages of the system, and it fit into their usual logic and decision-making algorithms.

The worst option is to implement artificial intelligence is a revolutionary method, when the team has neither its understanding nor acceptance. It is impossible to change instantly the years-old practice and make experienced professionals listen to the decisions of the machine. This would create conflict and become a significant barrier to implementation. Therefore, a smooth, systematic implementation scheme was developed.

Another important thing that creates difficulties when implementing artificial intelligence is that employees see it as a competitor, and not without reason. Eventually they will need to enrich their knowledge with the ability to work with artificial intelligence, and the labor market itself will also be transformed: those who can be easily replaced will leave, but new personnel will be needed from those who will work with the latest technologies, engage in their implementation, training, support. This is already happening: not so long ago, German Gref said that Sberbank stops hiring lawyers who do not understand the principles of working with a neural network, and this is just one particular example of the profession. “By 2030, PwC estimates that artificial intelligence will provide a 14% increase in global GDP, which is about 15.7 trillion dollars, that is, it will become an integral part of the economy, and we all need to take this into account now. The winner is the one who will join the game earlier and accumulate internal experience of employees in conjunction with AI” - concludes Dmitry Galkin.

Ways of dealing with technological challenges

A huge number of people are currently working on solving technological issues related to the development of artificial intelligence and ways to teach it to think. At the moment, in accordance with the words of the managing partner of Marketing Logic Dmitry Galkin, all progress is associated with the increase in the database, which can operate a neural network, as well as with the establishment of algorithms for its operation.

However, there is a technological problem, which is quite realistic to solve. This problem is related to the complexity of interacting with strong artificial intelligence, such as robots and machines, in the workplace. Many employees of companies that are actively trying to implement strong artificial intelligence face the difficulty of understanding the processes that occur inside robots, which is why there is a fear of using these machines. Sometimes these machines replace employees, but companies are not always sure that their actions will have a positive impact on the company's activities. Whom should they learn from in order not only not to harm, but also to increase the success of the company? The answer is quite simple: In the first stages after implementation, the answer is obvious -- this is the history of the organization for a year or two or three before implementation. Next - on the best employees. However, if the system of recommendations has been working for two or three years, then the following situation arises:

a) the system has gone through several iterations of training, and recommendations have become " smarter»;

b) it becomes more difficult for employees to generate "best practices";

c) there is a growing level of agreement with the system's recommendations, i.e. employees increasingly agree with the AI and confirm that the system's advice is " good»;

d) the level of emotional distrust of the system on the part of employees has decreased.

At this stage, there is a well -- known situation in statistics-sample bias. New training data is no longer available. The system closes in on itself and can potentially miss new opportunities. The usual method is to allocate a certain proportion of activities in which employees must go against the system. However, as you learn the system, you see that it is increasingly difficult for ordinary employees to formulate and implement new successful practices. Therefore, in this case, it was recommended to create special teams of employees who are experts in the process and at the same time able to search for new opportunities.

Methodology

Research strategy and the rationale

The research purpose of this study is exploratory with elements of descriptive research, as it is stated in the 1.2. It is based on the fact that the authors are describing the stages of HR business processes basing on the in-depth interviews with experts, then explored the correlation between opinions of experts and usual employees, and at last generated new hypotheses regarding the issues of employees behavior and changes in the labor market. Therefore, the research can be defined as exploratory with descriptive elements (with usage of in-depth interviews and cross-sectional surveys).

Regarding the research strategy, it can be defined as mixed method approach. The authors are using both qualitative and quantitative methods: qualitative, as content analysis of the interviews, and quantitative, as on-line survey. The particular strategy has been chosen according to the following factors. At first, this strategy includes both qualitative and quantitative parts of analysis, which increases the coverage of the topic: the authors can analyze it from the different angles to be sure that the results will not be biased due to the incorrect choice of the research strategy. The qualitative part goes firstly in order to provide the detailed information about the stages of HR business processes. These blocks will be asked about in the survey given to the future employees and HR experts. The three blocks of dominant skills, which are being searched for by the HR experts are chosen according to the report of McKinsey, where the dominants skills of employees of different professions were divided into three blocks: technical, creative and communication skills. Spheres, where HR experts work, were chosen according to this information.

Due to the fact that the authors conduct correlation analysis, the 7-step Likert scale has been chosen: it is the necessary scale for the particular type of analysis, which will provide the clear results. The content analysis is commonly used in explaining the results of in-depth interviews. The authors need this particular type of analysis in order to understand the similarities between opinions of HR experts in different companies.

Sampling

To start sampling with, the authors represent the main aim of the research: extract, to what extent do the opinions of experts and employees on the issue of fully automating the hiring process coincide. The authors have two groups of units of study.

First group consists from people in search for the job. These people not only are young people and University graduates. PwC research states, that right now there are very serious changes in the labor market related to the expectations of specialists. An increasing number of young professionals of generation Y prefer temporary and remote work, without tying himself or herself to a long-term contract with one employer. This is not the only reason for complicacy of counting the number of people in search of job. On January 15, 2020, the number of unemployed citizens registered with the employment service amounted to 663.3 thousand people. At the same time, the total number of unemployed in the country amounted to 3.507 million people at the end of November -- which increased by 31 thousand people from the data at the end of October [gks.ru]. To count the needed group, the authors exclude those, who are officially unemployed: 3507-663,3=2843 thousands of people. By the way, as of January 15, the number of available jobs and vacant positions declared by employers to the employment service bodies amounted to 1.4 million units.

Second group consists from the HR experts in Russian companies and corporations. According to the freshest information from fsi.rf [https://xn--h1ari.xn--p1ai/Main/StatisticalInformation], there is 7 758 891 of legal entities and individual entrepreneurs in Russia at 16 February 2020. The authors are considering them all as companies. Taking into account the percentage of zero accountable companies, the authors take the average of HR experts in 1 company in Russia (it is 1) [HeadHunter research]. Consequently, the number of HR experts in Russia is 7758891 approximately. Considering the counting of these two groups, the population is 7758891+2843000=10601891 people.

Sample is smaller and consists from people and companies from North-West region and Saint Petersburg and Leningradskaya area in particular. According to fsi.rf, there is 442093 companies in Saint Petersburg and 82208 in Leningradskaya area (524301 HR experts in sum). According to [https://www.gov.spb.ru/] there is 45000 unemployed people in these areas approximately. Consequently, the sample is 570000 approximately.

To form the sample frame non-random methods are more preferable [Ochoa, C. 2017]. The probability of each population participant to get into the sample is not equal, as the participation in the research is available to the people the authors can met and will be hardly sent to other people. Moreover, there is no chance of equal random questioning for exact number (it is presented below) of people from the population. Convenient sample method is chosen, as it is stated in 1.2 and as the authors select the people, who are easy to contact and the population is defined in terms of a very broad category. In addition, the participants are asked to participate if they are available and free at the moment.

The authors decided to make two different counting to the two groups they formed. For the people who are in search of the job, the sample frame requires 156 respondents to represent the results with 7% potential sample error. For the HR experts, it is required the same number of respondents with 7% sample error.

Variables

The variables are presented in the Table 1.

Table 1. Variables