Ethical issues IN AI use in hr management: russian experience

However, to date, it has been proven that AI copes with the prognosis much better than doctors. The success of AI in this situation is due to the fact that AI has access to a huge amount of data not only within the country, but also abroad. The ethical problem here is also that decisions are made based on probabilities. The probability appears only when the machine does not see enough data on a particular topic in its database and a situation of uncertainty arises. Some doctors have noted the fact that they try to predict the decision of the system, rather than form their own forecast (Lamanna, 2018). The lack of transparency in building trust in medical AI forecasts is also ethically problematic.

1.8 Ethical problems in the use of artificial intelligence in HR management

1.8.1 Discrimination

One of the most important ethical issues encountered in HR management when using AI tools is discrimination. Discrimination in the labor market is the unequal opportunities of employees with equal productivity, or the unequal attitude towards them on the part of employers, managers, society, and the state (Mazin, 2004). Individual employees and certain groups of people can be subjected to labor discrimination.

In Russia, labor discrimination has been studied and analyzed in detail since 2000. While abroad, discrimination in the labor market has almost always been the object of close attention. There are such types of labor discrimination:

1. Discrimination in wages of some employees or groups of employees in comparison with others.

In any country and in any company, employees who have equal productivity, qualifications and work experience should receive different salaries for performing the same work. But today, there are the most often discriminated against groups of employees based on the level of wages. For example, women compared to men (receive a smaller salary and do not occupy important positions) or Blacks compared to whites.

2. Discrimination in hiring and dismissal from work.

People with problems, such as those released from prison and people with disabilities, are usually subjected to such discrimination. They are hired last, or not at all. If we are talking about reducing the staff, then such people are dismissed first. Unequal employment opportunities may arise due to the age of the employee, now few people accept students without experience. It also refers to a person's race or ethnicity. It's all discrimination.

3. Discrimination in the promotion of those who already work in the organization.

Employees who discriminate against employees (or groups of employees) are almost never able to build a career, because they are reluctant to be promoted through the ranks and are not appointed to responsible positions. It can be said that women, immigrants and national minorities are also more often subjected to such discrimination.

4. Occupational segregation.

There are male and female professions. Professional segregation is observed when, for example, a woman wants to become a firefighter and extinguish houses, because all her life she loved to help people and her father took her to work in the fire department all her childhood. A girl may have excellent physical fitness, but after the interview, they will take a man, even if the motive for his employment is simply to earn money and not to put all your heart into the profession.

5. Discrimination in education and training.

For example, if a person was born in a poor family, he may simply not have enough money to continue studying after school, if he did not score the required number of points on the entrance exam.

These types of labor discrimination are closely related to each other. Discrimination in education contributes to occupational segregation, which perpetuates and deepens unequal access to education. Discrimination against a certain group of people reduces their income and investment opportunities in human capital, which makes it difficult for members of this group to access high-paying professions and strengthens occupational segregation.

If you look at AI systems from a very high point of view, these technologies function as systems of discrimination, they differentiate, rank, categorize, and thus, in some special cases, they discriminate and create social inequality. (Kazimadze, 2017). The authors note the presence of a diversity crisis in artificial intelligence technologies (hereinafter referred to as AI) in areas such as gender and race (West, Whittaker, Crawford, 2019).

While there is concern and public focus on "addressing" diversity in AI through working with data quality, fair models, and inclusive design, many experts insist that more in-depth analysis of workplace culture, power inequalities, harassment, discriminatory hiring practices, and unfair pay that force people to leave or avoid working in the AI sector altogether is needed (West, Whittaker, Crawford, 2019).

The problem of inequality in AI is not just a technical problem. This problem can be solved by using an interdisciplinary approach, involving different stakeholders, but most importantly involving civil society. The term "AI" is used to generalize technologies and systems that mimic the human mind, using a set of techniques such as automatic speech recognition, image recognition, natural language processing, speech generation, etc (Kazimadze, 2017).

An interdisciplinary approach is a way of expanding the scientific worldview within the framework of local worldviews, which implies enriching the knowledge, methodology and language of one scientific discipline at the expense of the knowledge, methodology and language of another scientific discipline, forming a moral responsibility for the results and consequences of professional activity, the level of which is determined by the framework of interacting disciplines.

Discussion on the subject of whether AI improves our life or increases the inequality is sharp as ever (O'Neil, 2017). Systems that use physical appearance as an indicator of character traits or internal state are deeply unreliable. The same applies to AI tools that claim to be able to determine sexuality from a head photo, predict "criminality" from facial features, or evaluate an employee's competence from "microexpressions" (West, Whittaker, Crawford, 2019).

Such systems replicate patterns of racial and gender bias, possibly deepening and justifying historical inequalities. The commercial use of such tools is a matter of great concern» (Olteanu, Castillo, Diaz, Kiciman, 2016).

Timnit Gebru, a researcher in the field of algorithm ethics at Microsoft, emphasizes that deep learning can change the insurance market, in which minorities and other vulnerable groups can be discriminated against only because car accidents occur in densely populated areas, where these groups mainly live (Gebru, Morgenstern, 2018).

Social networks, which were created to connect people with each other, have become trusted places to share personal information, photos, and leisure time. Everything from politics to religion is discussed on social media. Problems in the field of employment are currently one of the most important and acute problems of humanity in connection with the ongoing economic crises, mass unemployment, which threaten to worsen the quality of life of people and lower their standard of living (Kabaykina, 2020). In the Constitution of the Russian Federation, Article 19 stipulates: “The State guarantees equality of human and civil rights and freedoms regardless of gender, race, nationality, language, origin, property and official status, place of residence, attitude to religion, beliefs, membership in public associations, as well as other circumstances. Any form of restriction of the rights of citizens on the grounds of social, racial, national, linguistic or religious affiliation is prohibited” (The Constitution of the Russian Federation. Chapter 2. Human and Civil rights and freedoms. Article 19.).

At the legislative level, they are trying to make sure that there is no employment based on gender, age or nationality. But at the moment, the Russian labor market, which is free in its form, is influenced by many local restrictions, prejudices and cultural stereotypes.

Among other things, the use of the AI creates certain ethical risks. Because AI technologies are relatively young, their objectivity is not fully proven. It is not clear to what extent the AI is able to assess the candidate and whether the facts on which the AI was based are credible. Another ethical problem is that AI learns from human biases and mistakes. Due to the fact that society has only recently begun to shift towards equality, the environment in which the AI learns is not yet sufficiently tolerant. A good example of this is that Amazon dropped its recruitment algorithm in 2018 due to discrimination against women. Moreover, the AI cannot function without a person at this time. As mentioned earlier, the individual makes a significant contribution to AI education. The data on which the AI learns is generated by a person, and only a person can test the objectivity of the AI. The ethical question is whether a person can keep up with the AI and whether the AI can become objective without a person. The main problem in the use of AI technology is the absence of regulatory rules and laws. However, the regulatory process is already under way. On January 1, 2020, the Illinois Act came into force to regulate video interviews using AI. From now on, the employer is obliged to warn candidates about the use of AI technologies. This is the first attempt to regulate the use of AI technologies in labour law.

1.8.2 Transparency

The use of AI is growing in all business processes, thereby attracting more and more attention from society and the state concerned about the ethical problems associated with the use of AI. Analytical intelligence is embedded in the complex algorithms of the artificial intelligence system. The complexity of the algorithms can lead to a lack of transparency in the AI decision-making process (Munoko, Brown?Liburd, Vasarhelyi, 2020). Therefore, one of the main of these problems is the lack of transparency in how the tools of many AI vendors work. AI tools are associated with a "black box" without a clear explanation of their inner workings. Thus, ignorance and misunderstanding of the functioning of machine learning algorithms can lead to unconscious bias in decision-making, for example, when hiring new employees (Zielinski, 2020). Matisse Hollister, an associate professor of organizational behaviour at McGill University in Montreal, has been studying the use of artificial intelligence in the workplace for several years. She noted that initially, the secrecy and opacity of AI tools worked as a way to sell a product. Currently, the opacity of AI attracts a lot of attention and pressure from society.

This problem requires appropriate control from the state. For example, the creation of laws and regulations on the control of the use of AI in human resources. So, the state of Illinois signed a ground-breaking law regulating the use of AI in video interviews. Companies are required to notify candidates that they are using new technologies to analyse video interviews, and employees are also required to explain to candidates how AI works. The candidate, in turn, can give consent or refuse. Further, New Jersey, Washington, and New York have introduced relevant legislation to regulate the use of AI in hiring, pay, and other personnel-related decisions. Thus, developers pre-check the AI tools for bias. Appropriate regulatory bodies are also being created to investigate the use of artificial intelligence tools in the workplace. For example, the Electronic Privacy Information Center (EPIC) monitors companies that use AI tools. EPIC, a Washington-based public interest research center, accused HireVue of violating international and national standards of fairness, transparency, and accountability when using AI-based interviewing tools. EPIC believes that the uncontrolled use of AI harms job candidates who are exposed to “opaque and unprovable" hiring decisions.

To address this problem, AI tool makers are starting to work hard to increase the transparency and fairness of their AI products. For example, Kevin Parker, CEO of HireVue claims that HR managers for video interviews include a more detailed screen of what to expect from job candidates to give them more information about the role of AI in the evaluation process. Another solution to increase the transparency of AI algorithms is to create a GIF that will be available to show consumers how the artificial intelligence system accepts various factors and makes its recommendations on them. But here comes the risk that candidates can play with their own pre-evaluation systems.

1.8.3 Data privacy

Security and confidentiality issues are another challenge in the implementation of AI technologies. The AI algorithms use large amounts of data that need to be securely protected. The Information Technology Security Division should develop appropriate guidance on AI technologies that address not only data security aspects but also legal and ethical issues. AI tools such as chat bots should not store any personal data or confidential information during the processing of staff requests. This information should be transmitted through secure data transmission protocols and be encrypted to prevent data theft by third parties. Only data that does not contain personal information and confidential information about employees can be used to train AI. Finally, companies must be sure that they have the necessary tools to control HR data. The absence of such control can lead to erroneous algorithms and, consequently, mismanagement decisions. Therefore, when working with HR systems in which AI technologies are present, employees should have authorized access to information.

1.8.4 Replacing employees with Artificial intelligence

The difficulty of integrating AI into existing infrastructure is also a problem for different companies in implementing AI technologies. The difficulty of integration is that modern AI solutions for personnel management are narrowly focused. Before making a decision on integration into the infrastructure of AI-solutions, it is necessary to pre-test them in work taking into account the culture and business model of the company.

In addition, the maintenance of various AI solutions is costly and unpredictable. It is not yet easy to find specialists in the labour market with the necessary education and skills to develop, fine-tune and deploy AI technologies. In addition, although modern technology equipment is reliable, there is always the possibility of accidents and malfunctions. There is also the possibility of hacker attacks. Therefore, AI technologies require in-depth study, continuous analysis and updating.

Concerns about rising unemployment due to the introduction of AI in organizations also remain an important issue. The World Economic Forum produced the Future of Jobs Report, which predicts that by 2022 the number of jobs will have fallen by about 75 million, while around 133 million new jobs will have been created by the AI (Economic Forum. 2018). These jobs will involve preparing machines for effective work in the digital community, as well as analyzing and training AI in human skills such as critical thinking, empathy and problem solving. In the next few years it is unlikely that HR departments will be automated, but AI can fully take over HR-related tasks such as searching for candidates (63%), attracting applicants (47%), and training staff (47%) (Bit,2018).

To summarize the advantages of using AI technologies, it is worth noting that in 2019 KPMG conducted a study that showed that 88% of respondents who had already invested in AI considered the investment justified. 35% of respondents focused on training and 33% of respondents focused on analytics (Bolton, Dongri, Saran, & Ferr, 2019). With regard to priority business indicators affected by the application of AI instruments, particular attention should be paid to productivity. Developers of the Zoom.ai chat bot (virtual meeting planning and simplification of interaction between employees at remote work) have conducted research and found that users who actively use chat bot usually save up to 25 hours per month, and companies save on average about $16,000 per employee with a salary of $100,000. The value of this tool for business and HR becomes apparent when these indicators are analyzed on the scale of hundreds or thousands of corporate users (Eubanks, 2019). In addition, the use of chat bots simplifies and speeds up the transition to new systems as communication with them can replace the use of manuals. An example of the influence of AI on the speed of the introduction of new technologies in the company is the chat bot WalkMe discussed earlier. This indicator is difficult to measure, but it is certainly important. In addition, the use of AI tools makes it possible to carry out some of the non-urgent or less important tasks without human participation, and to concentrate the personnel managers on urgent and more important tasks. In this way, all the existing tasks of the HR-division are systematically solved, and the HR-head performs its work more efficiently.

Artificial intelligence tools also help to support people development, which increases their job satisfaction and consequently their loyalty to the organization. AI tools are also capable of analyzing the mood of employees, helping managers prevent the dismissal of valuable employees. All the above factors confirm the usefulness of artificial intelligence tools in the management of human resources. AI technologies enable companies to free HR specialists from routine tasks, optimize and automate many HR processes, increase productivity and speed of data processing.

However, companies need to be realistic about the appropriateness of using AI technologies. Before implementation, it is necessary to evaluate the return on investment, as well as to develop AI regulations that will address technical, legal and ethical aspects of the company's activities.

management artificial intelligence ethical

2. Methodology

2.1 Research strategy and the rationale

After analyzing previous studies, we found that research on the use of artificial intelligence in HR management in Russian companies from the ethical side is not enough. The introduction of artificial intelligence systems is happening rapidly. To begin with, we decided to determine what kind of artificial intelligence tools Russian companies use in HR business processes and how these tools help Hr managers in their work. The next step was to determine what ethical problems Russian companies face when using artificial intelligence in HR business processes, how companies solve these problems, and what measures are taken to prevent ethical problems. The last stage of the findings is to make recommendations. The research purpose of this study is research with elements of descriptive and comparative. Therefore, our type of research is qualitative data analysis, axial coding of interviews. The most important method of data collection is an in-depth interview. Interviewees can ask clarifying questions and change the sequence of questions or even questions in case to find out important information for the research (Hague P., Hague N., & Morgan, 2004).

We chose this type of interview because the interviewees can ask clarifying questions to complete their answer. We also use this method, because each organization has an individual approach to research in the field of AI in HR, to find out what AI programs companies use in HR management and to identify ethical conflicts when using AI. High-quality interviews are the tools needed to answer the research questions of the current study: “How do HR managers in Russian companies understand the ethical problems when using AI systems and what measures are taken to solve them?” The empirical base of the study includes semi-structured interviews with HR managers working with artificial intelligence systems. The interviews were collected in April 2021. A total of 15 interviews with an average duration of 40 minutes were collected. A numbered list with the following descriptions represents a research strategy for achieving results:

1. The determination of the companies to be included in the list of studies.

a. Companies are required to provide goods or services in the territory of the Russian Federation.

b. Companies linked to similar kind of activity (they have one control center)

S. Companies have a large number of staff (no less than 1,000 people)

d. Companies have a huge resource base

e. Companies have a high level of social security

f. Companies enter the markets of developing countries

g. Companies cover several regions of the country or are located internationally (they always have a large number of end users, so demand exceeds supply)

2. Identification of informants for the study.

3. Preparation of the interview protocol with a list of questions.

4. Pilot interview. Thanks to the results obtained from the interview, we will be able to finalize the protocol, correct errors, and shortcomings.

5. Address to informants.

6. Transcribing all Interviews.

7. Interview coding.

8. Content Analysis. The in-depth analysis of representatives' responses and identification of relevant information.

2.2 Criteria for selecting informants and companies

The HSE partner database was used to select respondents. Respondents for the study were selected through a mailing list: a letter was sent to the company requesting an interview on the topic "Ethical problems of using AI in human resources management: The Russian experience". The sample contains 15 representatives of companies specializing in the provision of services and goods on the territory of the Russian Federation. Companies use artificial intelligence in their HR management business processes. The companies are connected by a similar type of activity (they have the same control center). The companies also have a large staff (at least 1000 people). Companies have a huge resource base and a high level of social security. Companies enter the markets of developing countries and cover several regions of the country or are located internationally (they always have a lot of end users, so demand exceeds supply) All respondents are specialists who are hiring new employees. These specialists must work in these companies for at least 2 years. Interviews were conducted at will, which may impose some restrictions on the sample. Therefore, we used a snowball sample for respondents in individual companies. This method is considered effective because it uses referral networks of already selected respondents (Cooper and Schindler (2006: 457)). This method provides high completeness and reliability of the answers, since the interviewer comes “on the recommendation". Since all employees sign non-disclosure documents and in general, it is not supported in the culture of organizations to give interviews on such topics. All our interviews from respondents are confidential.

2.3 Research tools and analysis logic

To conduct the research, a guide was developed, divided into several thematic blocks. The guide includes the following 6 blocks: basic information about the respondent and his / her work experience, the use of artificial intelligence in work, discrimination of artificial intelligence systems, collection, and storage of personal data, how well the respondent understands the work of AI, questions about the regulation of digital technologies and digital products.

Block 1.

Includes questions related to basic information about the respondent and work experience.

1. Who do you work for? Briefly describe your responsibilities. Tell us briefly about your work experience.

Block 2.

Questions about the use of the AI system in HR business processes, how they help and what problems they solve.

2. What AI systems do you use in HR business processes?

3. How do these systems help you and what is their effectiveness?

Block 3.

Questions about possible discrimination of candidates by artificial intelligence, what solutions exist to the problem of discrimination of AI candidates, what measures should be taken to avoid cases of discrimination.

4. Do you think AI helps human rights or makes the situation worse?

5. Were there any cases at work when the system discriminated against candidates or employees based on any criteria?

6. How did you solve this problem? What measures do you take to avoid discrimination?

Block 4.

Questions about the storage and use of personal data, the confidentiality of personal data, for what purposes the company uses the data.

7. Does the AI system collect and store personal data? For what purposes do you use this data?

8. How do you guarantee the confidentiality of all data collected by the AI?

Block 5.

How well does the respondent understand the work of AI, how can AI be useful in the work, how transparent is the work of AI for the respondent and future candidates.

9. Do you think digital products should be understandable?

10. How transparent is the AI system you use in your work? In other words, when the system makes a decision, it makes a recommendation, etc. Do you know how and why she made this decision?

11. Do you check the criteria that the system uses for accuracy and correctness?

12. Do you notify candidates that you use AI systems in your work?

13. Do you explain to the candidates what factors the AI system takes into account and how it makes its recommendations on these factors?

Block 6.

This section contains questions about respondents ' views on whether strict regulation of the development of technology and entrepreneurship is necessary, as well as the respondent's opinion on the development of digital products.

14. Does strict regulation hinder the development of technology and entrepreneurship?

15. What do you think is the current stage of development of Ai systems used in HR management?

16. What do you think in the future Ai systems will completely replace the HR manager?

The analysis of the interview was carried out by the axial coding method. There were 6 main categories: "AI systems", "Pros and Cons of AI", "Discrimination of AI candidates", "Storage of personal data", "Understanding of AI work" and "General opinion about AI and its prospects". Below is a tree of codes with the main categories and sub-categories that I used during the analysis of the interview.

1. AI systems

a. Clever Stuf

b. Sever.ai

c. Yva.ai

d. HRM

e. PolyAnalist

f. Websoft HCM

g. Document recognition systems

h. Personnel screening

i. Video analytics

j. Recommendation systems

k. Chatbots

2. Pros and cons of AI

a. Pros

b. Cons

3. Discrimination employees and against AI candidates

a. Have not encountered

b. Have encountered

i. Discrimination criteria

ii. Ways to solve the problems

iii. Methods of preventing discrimination

4. Understanding and transparency how AI works

a. Understanding how AI works

i. Full understanding

ii. Inaccurate understanding

iii. No understanding of the work

b. Transparency how AI works

i. Full transparency

ii. Inaccurate transparency

iii. No transparency of the work

5. Personal Data

a. Storage of personal data

b. Use of personal data

c. Confidentiality

6. General opinion about AI and its prospects

a. AI opinion

b. AI perspective

The category "AI Systems" includes all artificial intelligence systems that were mentioned by the informants during the interview. This category should also be considered in conjunction with the category of Pros and Cons of AI, thanks to which, there is an understanding of what are the pros and cons of different AI systems.

To understand what possible cases of discrimination are possible when using AI, the category "Discrimination of AI candidates"was introduced for each type of AI system. Respondents described whether they had experienced AI discrimination in the course of their work, what measures had been taken, if any, and how to prevent AI discrimination.

The category" Understanding the work of AI "includes respondents' knowledge of how the AI system works, understanding of checking the accuracy and correctness of the AI work, and how transparent and understandable the AI system used in the work is to the respondent.

The last category "General opinion about AI and its prospects" includes questions that relate to the respondent's personal opinion about the use of AI and other digital products in work, as well as what prospects for the development and regulation of AI the respondent sees.

The first stage of encoding was built on the basis of existing AI systems. If during the coding process there were AI systems that were not previously described, they were also added. At the first stage, AI systems were highlighted, as well as the pros and cons of a particular AI system, which the respondent described.

At the second stage of coding, AI discrimination cases were encoded. In the course of describing the discrimination cases, the respondents also described how this problem was solved and what methods are available to prevent discrimination of candidates by AI.

Another stage of coding was the block "Understanding the work of AI", which is no less important. In the course of the interview, the respondents told how clear the AI system is for them and how well they know why the system made a particular decision.

The next stage of coding was to provide information about the storage, privacy and use of personal data that is collected during the operation of the AI. Such sub-categories as personal data storage, confidentiality, and how exactly the company uses this data were highlighted.

The last step was to encode the employee's personal opinion about the work of AI and its future prospects. At the first stage, the general opinion about the work of AI was noted, and then the sub-category “AI Perspectives” was highlighted. Respondents told how they feel about the strict regulation of technology and entrepreneurship development, digital products and the work of AI.

2.4 Limitations

The high level of confidentiality in companies, in this regard, at the moment there are not so many HR specialists who agree to give interviews. In addition, in some companies, HR managers must be HR professionals to have access to data and provide reliable information about AI processes.

3. Results

3.1 What AI systems do different companies use

During the interview, the experts named various artificial intelligence systems that help them in their work. Among the special systems called PolyAnalist, Yva.ai., Sever.ai., Clever Stuff, HRM system, Websoft, and HCM. However, not all respondents mentioned the exact name of the system, some said that they use neural networks, robotic voysbots, as well as their own document recognition system. Also, some companies use not one, but several artificial intelligence systems at once, which perform different work in the selection of personnel:

“На этапе скрининга и подбора кандидата на открывающуюся вакансию в нашей компании мы используем систему Clever Stuff; для дальнейшей обработки резюме, первичного обзвона и проведения интервью мы используем Sever.ai.; а для общего HR-менеджмента действующих сотрудников мы используем Yva.ai.”

“Мы используем Нейросети, а также роботизированые войсботы.”

Most companies implement tools such as a document recognition system, personnel screening, video analytics, a recommendation system, and chatbots. Most of the employees said that they use chatbots.

“Главным инструментом в нашей компании на сегодняшний день является система распознавания документов. Данный инструмент упрощает работу HR сотрудников, он автоматически анализирует текст, превращает его в данные и переносит всю информацию в учетные системы.”

“Мы используем кадровый скрининг, он оптимизирует работу отдела кадров, экономит их время и позволяет быстро закрывать срочные вакансии. Нельзя не отметить значительную экономию финансов, потому что данная услуга стоит значительно меньше, чем если оплачивать труд сотрудника.”

“Одним из эффективных инструментов в целях экономии времени являются чат-боты. Чат-боты умеют не только собирать информацию, но и проводят предварительный отбор, отвечая на вопросы кандидатов в онлайн режиме. Также у нас чат-боты занимаются адаптацией новых сотрудников.”

“Наверное, одним из самых важных инструментов является рекомендательная система. Она определяет интересы сотрудника и в соответствии формирует предложения для конкретного сотрудника. Например, система может предложить сотруднику отпуск, премию и все, что важно получить сотруднику на основании его истории в данный период времени.”

3.2 Pros and cons of the artificial intelligence system in the selection of personnel

In the course of communication with the specialists of the companies, we were also interested in what pros and cons they can note in the work of AI. First, the respondents noted that thanks to AI, the time for hiring an employee is reduced: the AI system is designed in such a way that it allows you to make a large number of decisions in a short time. Moreover, the AI system is quite easy to use for specialists, since the mechanisms of its operation are quite transparent. In many companies, according to respondents, there are special programs for adaptation and training of work in the system.

“Алгоритмы крайне эффективны: они могут понимать много решений за небольшой промежуток времени это позволяет ускорить и упростить принятие некоторых решений.”

“Все просто и прозрачно, что понятно как система работает.”

“Она не требует знаний программирования совсем.”

Moreover, as one of the advantages of AI, an important quality was highlighted - the ability to analyze large amounts of information, which is almost impossible to do with the help of a person, or it will take a very long time.

“Искусственный интеллект может проанализировать тысячи таких команд, там, где человек не справляется”

“Clever Stuff помогает нашим HR менеджерам в автоматизированном режиме осуществлять парсинг резюме кандидатов по 15 площадкам и порталам”

An interesting plus, which was told by the respondents who hold the post of head of the company, is the reduction in the number of employees of the HR department. The managers explained that the implementation of the AI system can reduce the number of employees, as well as save money.

“Позволяет сократить штат HR-департаментов.”

“Сокращает время на закрытие позиций, сокращает траты на подбор и найм персонала, увеличивает конверсию найма.”

As for the disadvantages of the AI system, there are not as many of them as there are pluses. However, some of them can significantly affect the operation of the system. Respondents still noted that AI is not able to completely replace the work of a person, so an employee is still necessary. However, sometimes the people who program the AI system themselves make a mistake in selecting the criteria, and this can also affect the choices that the AI makes. The AI system requires constant monitoring and training, so that there are fewer selection errors each time.

“Искусственный интеллект не способен полноценно заменить человека в этой сфере.”

“Но в этом есть опасность, поскольку если в систему заложена ошибка это влияет одновременно на много решений, тогда как если ошибается человек это может повлиять только на одно решение.”

“ИИ до сих пор глуповат и делает иногда глупые ошибки с точки зрения человека.”

3.3 Cases of human discrimination by the artificial intelligence system

No less important during the interview was to find out whether the respondent had any cases when the AI discriminated against a person on a particular occasion. In most interviews, respondents noted that discrimination is possible in the models used. For example, discrimination based on gender or age. Since the data used in the models already initially discriminate against society. It is worth noting that the algorithms reveal a bunch of hidden dependencies, and it is not always possible to say what exactly the system has highlighted. The respondents also stressed that at the moment this is a difficult problem, since if you exclude discriminating characteristics such as gender and age from the model, they can still be determined by other parameters.

«Дискриминация, конечно, есть, например, у нас есть данные о том, сотрудник мужчина или женщина и может быть дискриминация по полову признаку. Там алгоритмы выявляют кучу скрытых зависимостей и не всегда можно сказать, что конкретно он выделил. Пол человека можно определить по куче других параметров, это зависит от продолжительности отпуска, по количеству детей. Например, женщин с детьми больше, чем мужчин, т.к мужчины разводятся, а с детьми остаются женщины. Из таких зависимостей можно выделить, мужчина или женщина.»

“Насчёт половой дискриминации, то тут очень сложно говорить. Чтобы правильно понять, как это все работает нужно осознать то, что модель учиться на данных, которые создал человек. Поэтому если в данных присутствуют какие-то когнитивные искажения людей, то модель будет также эти когнитивные искажения использовать.”

“Результат из работы при равных входных данных всегда одинаковый, если система статична и это даёт нам широкие возможности, поскольку люди могут принимать разные решения в зависимости от настроения, степени усталости. Алгоритмические системы постоянны и если в них ошибка, то они будут дискриминировать последовательно.”

Only a couple of respondents did not meet with cases of discrimination, especially since the respondents noted that the data from the AI is just a recommendation, and the final decision is left to the person.

“Скорее нет. В любом случае личное общение никто не отменял. Эти данные лишь рекомендации, но не финальное решение.”

“Нет, подобных кейсов не было совсем. Все системы, основанные на ИИ, которые мы применяем, либо вообще не берут во внимание факты по которым можно дискриминировать человека, либо дают обезличенные выводы по целым группам сотрудников или людей без указания на конкретного человека, специально чтобы избежать дискриминации.”

Of course, there are certain ways to solve the problem of discrimination when using AI. One option, the respondents say, may be to remove some characteristics so that they are not used in the classification and processing of the results in the future. Therefore, to prevent discrimination on the part of AI, as respondents note, it is equally important to understand the criteria by which the system makes a decision, as well as to personally check why the system chose certain candidates. You can also set coefficients for characteristics to reduce possible discrimination. All respondents noted that the data from the AI is a recommendation, and the final decision is left to the employee.

“Мы можем выкинуть часть данных о человеке, то, что мы не хотим использовать для классификации.”

“Например, проверка этих рекомендаций лично.”

“Можно задавать веса критериям и добавлять новые критерии, проверять.”

3.4 Understanding and transparency of the work of artificial intelligence by company employees

All respondents noted that all AI systems should be clear and transparent. In practice, most respondents know what criteria are used to select AI systems, and how these systems make decisions on the selection and evaluation of personnel. As for the internal part, for example, on the basis of which mathematical models the system makes a decision, employees do not know, and they do not need this information. The employees also said that their companies have training and adaptation so that working with AI is as clear and transparent as possible.

“В большей степени прозрачная. Есть описание алгоритма работы.”

“Алгоритм формирования выводов системой относительно понятен, они рассчитываются по определенной формуле на основе более чем 40 параметров.”

“В основном очень важно знать на каких основаниях модель приняла то или иное решение.”

Only a few respondents said that they know how to use the AI system, but do not know what factors are used to make decisions by this system. Since the AI system is strictly classified in the company.

“Есть обучение, но система не полностью прозрачна для сотрудников HR, мы умеем ей пользоваться, но не знаем, как именно она работает.”

“Сотрудники HR департамента проходят обучение, где объясняется, как пользоваться системой ИИ, но система остается не полностью прозрачной. Я не совсем понимаю, как работает система и многие критерии, принимающиеся в учет инструментами ИИ от сотрудников скрыты. Специалисты data scientists разбираются в этом гораздо лучше эйчарщиков.”

As for checking the criteria of the AI system for accuracy and correctness, almost all respondents said that they do not check the criteria, and once again stressed that the results of the system are of a recommendatory nature.

“Нет. Результаты работы системы носят рекомендательный характер, и только специалист принимает решение, прислушиваться к ним или нет.”

“На данный момент мы не проверяем критерии системы ИИ на точность, результаты системы носят исключительно рекомендательный характер на данный момент. Финальное решение у нас всегда остаётся за живым человеком из команды нр, как правило менеджером по подбору персонала.”

All respondents said they warn candidates about using AI systems when selecting candidates. But no explanations are given to candidates about the work of AI tools, and the factors and criteria for decision-making by AI systems are not explained. All AI tools remain opaque to candidates. All respondents refer to the fact that this is confidential information of the company.

“Разумеется, мы предупреждаем кандидатов о том, что в работе мы используем системы ИИ. Наша компания использует видео аналитику, и каждый кандидат перед интервью получает подробную инструкцию. Кандидатам мы не объясняем, на какие факторы система обращает внимание, т.к это конфиденциально и важно знать только сотрудникам HR отдела, как ИИ делает рекомендации по этим факторам, мы не объясняем также.”

“Да, конечно, мы предупреждаем об использовании ИИ каждого кандидата, если после отправки резюме они получили отрицательный ответ и после интересуются, в чем была причина отказа. У нас технология занимается подбором резюме, как она работает наши HR менеджеры знают, но мы не озвучиваем, из чего принимаются решения системой.”

3.5 User's personal data: collection, storage, confidentiality, and use

As for the storage of personal data, the respondents noted that they are under reliable "protection". Technical organizational measures are used to protect data, both in the cloud and on servers: data encryption, support for 2-contour deployment - an external contour with depersonalized data and an internal contour with full information composition, and integration with Identity Management systems. Most of them said that in order to access them, you need to write an application and go through some procedures, since it is impossible to get them via the Internet. Also, not all systems store data. According to the employees, the system can collect personal data, and then store only the information that is necessary in an anonymous format.

“Только одна система собирает, но не хранит именно личные данные сотрудников. После того как она их собрала (разрешение на сбор данных дает или не дает сам сотрудник компании который ею пользуется) система делает анализ полученных данных и удаляет их, оставляя только полученные выводы из данных, но даже они хранятся сугубо на серверах компании и никуда за периметр внутренней системы компании не выходят.”

“Все, что связано с личными данными хранится в системах, где нет даже интернета. Ты даже не можешь иногда с некоторыми чувствительными данными поработать с компьютера, который находится в офисе, тебе надо подключаться к другому серверу, которые находится в офисе.”

As for privacy, when answering the question about personal data, many employees referred to the law that regulates the storage, use and transfer of personal data of users.

“По поводу конфидента есть закон о персональных данных и там все жестко, но если компания не крупная, то на него забьют, если компания вроде мегафона, то за ними очень следят и они легко могут быть пойманы на фз”

“В соответствии со 152-ФЗ, техническими мерами исключается распространение персональных данных.”

Respondents said that using data-linguistics, different data about employees is collected. It is a great practice to collect data from chat logs and task traker (programs where employees ' tasks are conducted). Thus, data is collected about what these people think, how they talk, how they communicate with each other, how they interact, and data about what tasks these people solve, how they work with certain tasks, what is the speed, which people are more productive, which are less productive. This data is used to diagnose individual employees and teams as a whole.

“С помощью data-лингвистики мы получаем интереснейшие данные про то, о чём эти люди думают, как говорят, как общаются друг с другом, как взаимодействуют с командой.”

“Мы вытаскиваем оттуда вот эти данные, анализируем их и так мы можем посмотреть, как и какие задачи решаются этими людьми, как они работают с определенными задачами, какая продуктивность сотрудников, какая командная роль у определенных сотрудников и т.д.”

Conclusion

The purpose of this study was to describe what AI systems are used by HR managers in Russian companies, to describe what ethical problems employees face when using AI in Russian companies, and to give recommendations. During the interview, we found out that many companies use no single AI system, but several at once, for example, at the screening and selection stage, they use Clever Stuf to select candidates based on formal criteria, which helps to quickly close a vacancy while reducing costs and working hours of specialists. To process resumes, make initial phone calls, and conduct interviews, companies use Sever.ai that saves employees ' time. For general HR management, the company's current employees use Yva.ai, to identify employees at the initial stage of burnout. The HRM system of the company is used not only for communication of employees, but also for the analysis of vacation provision, planning of employee development, adaptation of new employees, as well as a real-time module of motivation and reporting. Thus, any decision is made through analysis and significantly saves time in HR processes. From the interview, we learned that many HR managers use the PolyAnalist system for data processing, this system does not require programming knowledge. We also learned about the Websoft HCM system from the interview. This system automates routine processes, saves a complete history for each candidate, conducts analytics for each recruitment project, and accompanies new employees during the adaptation process. Using this system, HR managers increase efficiency and save time on routine processes. Companies also use systems such as document recognition, which automatically analyzes the text, turns it into data, and transfers all the information to accounting systems, which again saves employees time. Some companies use HR screening, which optimizes the work of the HR department, saves their time and allows them to quickly close urgent vacancies. During the interview, we found out that companies use video analytics, thereby saving employees ' time, since this tool uses machine vision algorithms, conducts video monitoring and analyzes data without human intervention. Recommendation systems determine the interests of an employee and, in accordance with them, form proposals for a particular employee, which increases the effectiveness of employees. The last tool, but no less important, is chatbots. Respondents noted that chatbots are an effective tool to save time. They collect information and help new employees adapt. The big advantages of all these AI systems used in Russian companies by HR managers are: first, it saves employees 'time, since such systems can quickly process a huge amount of data in a short period of time; Second, it reduces companies' spending on HR processes, since there is no need to hire a large staff of people; Third, it increases the efficiency of the company's HR management, since decision-making takes place through analysis.

We found out during the interview that there are indeed ethical problems that HR managers face when using AI systems in Russian companies. One of the first problems that the respondents confirmed was the problem of discrimination. AI and other smart technologies harm stakeholders who are already a miniginalized part of society (Noble, 2018). In most interviews, respondents noted that discrimination is possible in the models used. For example, discrimination based on gender or age. It is worth noting that the algorithms reveal a bunch of hidden dependencies, and it is not always possible to say what exactly the system has highlighted. Thus, this confirms the first hypothesis, which suggests that when using AI systems in HR business processes in Russian companies, employees face the problem of discrimination. As it turned out, this issue cannot be solved until the end, since discrimination can be initially embedded in the data and new correlations constantly appear. But still, companies understand the full responsibility for the use of AI systems, so the systems are constantly being improved, discriminating factors are removed, different coefficients are put on the criteria, and the most important thing is that the decision is left to the employee.

The second ethical problem is the transparency of AI systems, ignorance and misunderstanding of the functioning of machine learning algorithms. The transparency of the AI system means that employees are aware and understand how the artificial intelligence system takes various factors and makes its recommendations on them. The complexity of algorithms can lead to a lack of transparency in the AI decision-making process (Munoko, Brown-Liburd, Vasarhelyi, 2020). In most companies, but not all, employees said that they understand from what the AI system makes a choice, according to what criteria decisions are made by AI systems. It is worth noting that all companies undergo training and adaptation. But HR managers do not know all the mechanisms of the AI system completely, as it turned out, HR management employees do not need this knowledge, the IT department has this information. During the interview, we found out that HR managers notify candidates that they are using new technologies for analysis, but employees do not explain to candidates what factors the AI system takes into account and how it makes its recommendations on these factors. The state of Illinois has signed a law regulating the use of AI in video interviews. Companies are required to notify candidates that they are using new technologies to analyze video interviews, and employees are also required to explain to candidates how AI works. Therefore, the hypothesis 2 that when using AI systems in HR business processes in Russian companies, employees face the problem of opacity of AI tools has been confirmed. But it is worth noting that we also found out that in almost all companies, employees do not check the criteria of AI systems for correctness and accuracy, which means that if there is an error in the models, it will be quite difficult to notice it and not immediately.

...