Literature Review on the use of Artificial Intelligence in Human Resources Field

Università degli Studi di Milano-Bicocca

Dipartimento di Scienze Umane per la Formazione “Riccardo Massa”

Corso di Laurea Magistrale in Formazione e Sviluppo delle Risorse Umane

Literature Review on the Use of Artificial Intelligence in

Human Resources Field

Relatore:

Prof. Fabio Mercorio

Correlatore:

Dott. Alessandro Castelnovo Tesi di Laurea Magistrale di:

Matteo Della Valle

Matricola:

900304

Anno Accademico 2023-2024

Contents

Contents i

List of Tables iii

1 INTRODUCTION 1

1.1 Contributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

1.2 Thesis Outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

2 RESEARCH METHODOLOGY 7

2.1 Selection of Sources . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

2.1.1 Scientific Journals and Papers . . . . . . . . . . . . . . . . . . . . . . 7

2.1.2 Web Articles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

2.2 Criteria for Evaluating Articles . . . . . . . . . . . . . . . . . . . . . . . . . 8

3 RELATED WORKS 11

3.1 Ethics of AI-Enabled Recruiting and Selection: A Review and Re-

search Agenda . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

3.2 A Survey on XAI and Natural Language Explanations . . . . . . . . . . . 13

3.3 Fairness, AI & Recruitment . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

4 THE ARTIFICIAL INTELLIGENCE ACT 15

4.1 The AI Act, A Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

4.2 The AI Act, The Official Regulation . . . . . . . . . . . . . . . . . . . . . . 19

4.3 The High-Risk AI Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

i ii

CONTENTS

4.4 Fairness, Explainability, Privacy And Human Oversight: How The AI

Act Modifies These Themes? . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

5 FAIRNESS 31

5.1 Fairness in HR Context . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31

5.2 Opportunities in the Use of AI in HR to Promote Fairness . . . . . . . . . 32

5.3 Threats in the Use of AI in HR to promote Fairness . . . . . . . . . . . . . 38

5.4 Paper Scores and Considerations . . . . . . . . . . . . . . . . . . . . . . . . 43

6 EXPLAINABILITY 49

6.1 The Problem of ’Black Box’ Algorithms in HR Techniques . . . . . . . . . 50

6.2 Explainability Under a Legal Perspective . . . . . . . . . . . . . . . . . . . 55

6.3 Paper Scores and Considerations . . . . . . . . . . . . . . . . . . . . . . . . 57

7 PRIVACY 61

7.1 Privacy Concerns in Voice, Body and Emotional Analysis . . . . . . . . . 62

7.2 Privacy Concerns in the Analysis of Social Media and Web Scraping . . . 65

7.3 Privacy Concerns in Big Data and HR Analytics . . . . . . . . . . . . . . . 68

7.4 Paper Scores and Considerations . . . . . . . . . . . . . . . . . . . . . . . . 72

8 HUMAN OVERSIGHT 75

8.1 The Importance of ’Human in the Loop’ in HR Decisions . . . . . . . . . . 76

8.2 Human Oversight Under a Legal Perspective . . . . . . . . . . . . . . . . . 80

8.3 Paper Scores and Considerations . . . . . . . . . . . . . . . . . . . . . . . . 83

9 CONCLUSION 87

9.1 Unconsidered Points and Future Perspectives . . . . . . . . . . . . . . . . . 89

List of Tables

5.1 This table

Overview of Papers with Scores and Orientations.

summaries various papers along with their respective scores and orien-

tations in the context of the research. . . . . . . . . . . . . . . . . . . . . . 45

6.1 This table summaries various

Overview of Papers with Scores.

papers along with their respective scores in the context of the research. . 58

7.1 This table summaries various

Overview of Papers with Scores.

papers along with their respective scores in the context of the research. . 73

8.1 This table summaries various

Overview of Papers with Scores.

papers along with their respective scores in the context of the research. . 84

iii

Chapter 1

INTRODUCTION

In recent years, the global business landscape has been profoundly transformed by the

digital revolution, which has introduced a series of technological innovations destined

to revolutionise business operational processes [84]. In this context, modern artificial

intelligence (AI) systems are emerging as fundamental tools to optimise a wide range

of business activities. Among the various sectors benefiting from these technologies,

recruitment and personnel selection in human resources represent an area where AI can

offer significant and revolutionary contributions. As argued by Eubanks [39], modern

AI-based recruitment and selection systems allow for the automation of tasks tradition-

ally performed by humans, doing so more quickly, efficiently, and at a lower cost [39].

Rathore [187] also contends that one of the largest investments made by organisations

is directed towards hiring and selection, in terms of time, money, and the difficulty of

choices to be made. Therefore, these systems can make corporate selection processes

more efficient and rapid [187].

Modern companies must necessarily embrace these innovations to maintain compet-

itiveness and improve the efficiency of their operations [96]. The introduction of AI

systems into the recruitment and selection process allows for the automation and re-

finement of many traditionally manual activities, reducing not only the associated time

and costs but also enhancing the quality of hiring decisions. In a sector characterised

by the so-called "war for talent," the integration of AI systems into the recruitment and

1 2

CHAPTER 1. INTRODUCTION

selection process has become essential. By automating and refining many traditionally

manual tasks, these systems not only reduce time and costs but also improve the qual-

ity of hiring decisions. Embracing this approach enhances operational efficiency and

speed compared to traditional HR methods [205] and is crucial for staying competitive

in today’s talent-driven landscape [58, 75].

Indeed, AI systems can perform numerous crucial tasks in supporting human re-

sources work. Among these, screening and filtering resumes represent one of the pri-

mary areas of application: AI can quickly analyse resumes, extracting key information

such as skills, work experience, education, and qualifications, and compare them with

the requirements of open positions. This allows for the more efficient and accurate

identification of suitable candidates [54, 187].

Another essential task that AI can perform is conducting preliminary interviews

through chatbots or video interview systems [187]. Chatbots can interact with candi-

dates, asking standardised questions and collecting responses for first evaluation. Video

interviews, analysed by AI, allow for the assessment not only of the content of responses

but also of body language and tone of voice, providing a more comprehensive under-

standing of the candidate [54, 64, 96]. Another important tool related to interaction

with potential candidates is instant messaging, conducted precisely by chatbots, which

allows for the rapid resolution of doubts and questions from candidates [187].

The application of AI systems in HR processes has now become a well-established

practice in many companies worldwide, and this trend is expected to grow rapidly in

the coming years [54].

However, while AI holds the promise of significantly enhancing HR processes, it

also comes with certain fears. An analysis by Ore and Sposato [96] of a Fortune 500

multinational reveals that although AI offers numerous advantages, it raises significant

ethical concerns. On the positive side, AI can enhance data analysis for screening, im-

prove candidate-position matching, and increase decision-making efficiency [63, 83]. It

can also elevate the candidate experience by providing timely and consistent feedback,

which is particularly valuable in large companies where communication with all appli-

3

CHAPTER 1. INTRODUCTION

cants is challenging [36]. Moreover, AI can strengthen employer branding by better

meeting candidate needs and improving interactions, thereby increasing awareness of

brand values [88].

Conversely, there are notable concerns regarding the ethical use of AI in HR, such

as reliability, accuracy, data privacy, and the risk of discrimination. Additionally, there

is apprehension that AI could replace human recruiters in the selection process [73].

A significant issue is the potential for AI systems to perpetuate biases, leading some

experts to question whether AI can be fully trusted without human oversight.

These grey areas and ethical dilemmas must be carefully addressed to ensure that

the adoption of AI in HR not only enhances efficiency and decision quality but does

so in an ethical and transparent manner [2]. Therefore, to use AI systems responsibly,

it is essential to incorporate principles of fairness, explainability, privacy, and human

oversight, as will be discussed throughout this thesis.

One of the main ethical dilemmas concerns fairness in AI-supported decision-making

processes. Moreover, many authors underline the difficulty in accessing data that are

not victim of historical bias. AI algorithms can perpetuate or amplify existing biases in

training data, leading to discriminatory decisions based on factors such as race, gender,

age, or other protected characteristics [200]. It is essential for companies to implement

measures to monitor and mitigate these biases, ensuring that AI systems are designed

and used fairly [9].

Another critical aspect is the explainability of AI algorithms. Often, these systems

operate as "black boxes," making it difficult for human decision-makers to understand

the reasons behind certain decisions or recommendations [6]. The lack of explainability

can lead to a decrease in candidates’ trust in algorithmic decisions if decision-makers and

algorithm programmers do not justify them. Furthermore, the GDPR allows citizens to

ask the decision-maker the reason why a particular decision was made, the process that

led to that decision, and naturally, if the professional is unable to adequately answer

this question, it not only jeopardises the candidate’s trust but also violates the law [30,

178]. Therefore, it is crucial to develop and implement an explainable AI system to

4

CHAPTER 1. INTRODUCTION

ensure that the decision-maker themselves can trust the outputs it provides.

Privacy stands for another significant challenge. AI systems that use large amounts

of personal data must be designed to protect individuals’ privacy [33]. Through this

work, we investigate the most pressing privacy issues of recent years, examining how

modern artificial intelligence systems can pose a threat to the security and protection

of citizens’ data and identity. We pay particular attention to the European context,

characterised by stringent privacy protection regulations, as governed by the GDPR

(General Data Protection Regulation).

Finally, human oversight remains a fundamental element in the use of AI for re-

cruitment and selection. Despite the automation and efficiency that AI can offer, it

is essential for humans to maintain a supervisory role to ensure that final decisions

are fair, ethical, and in line with company values; the so-called "human in the loop"

process as discussed by van den Broek et al. [10] is indeed essential for the proper

functioning of a selection algorithm; decisions made must necessarily be verified and

controlled step by step by the individuals who handle them [10]. Human supervision

helps identify and correct any errors or biases that AI may introduce, ensuring that the

recruitment process remains under human control. Within this work, we will delve into

the detailed analysis of these issues, examining the ethical and practical implications of

AI use in recruitment and selection, and providing recommendations on how to address

these challenges to ensure responsible and ethical use of AI in HR.

It is no coincidence that the four principles just presented are focal points within

the AI Act, where the responsible use of AI is not only a benefit for individuals but

also a requirement for companies. This is particularly true for the HR sector, which,

as we will see, is classified as high-risk by the 2024 regulation.

1.1 Contributions

This work aims to provide readers with clear and structured guidance on the critical

topics of fairness, explainability, privacy, and human oversight in the context of AI in

5

CHAPTER 1. INTRODUCTION

HR. To achieve this, the study surveys relevant literature, organising each thematic

chapter with a summary table that lists the examined articles along with their assigned

scores.

The scoring system is designed for helping readers distinguish the varying levels

of depth and focus within the articles. Importantly, these scores are not intended to

indicate the scientific significance of the articles but rather to serve as an intuitive guide

for understanding the complexity and coverage of the discussed topics.

The adopted methodology integrates both academic research and public opinion

sources, offering a holistic perspective of the topics. By employing a scoring system,

the work seeks to ensure a clear and accessible classification of the articles, helping

readers easily navigate and understand the four central areas of responsible AI in HR.

1.2 Thesis Outline

The thesis is divided into eight main chapters, organised as follows:

• This part will discuss the research methodology, including the criteria

Chapter 2:

for selecting sources and the scoring system applied to the various papers analysed

in the survey, in relation to the different topics of the thesis.

• This section will identify the key reference works that are related to

Chapter 3:

the thesis, highlighting their contributions to the field and their significance for

the research.

• This section will provide an in-depth analysis of the new regulation

Chapter 4:

introduced by the 2024 AI Act, with a focus on high-risk systems, including those

used in Human Resources. It will also explore how the four main topics of this

thesis — fairness, explainability, privacy, and human oversight — are addressed

within the EU legislation.

• This part will delve into the concept of fairness, discussing its var-

Chapter 5:

ious interpretations in the context of artificial intelligence and decision-making

6

CHAPTER 1. INTRODUCTION

processes. It will then analyse the positions of different authors on this topic,

highlighting the division between opportunities and threads regarding the full

implementation of AI in HR practices. Additionally, tables will be presented,

assigning scores to the articles examined in the survey.

• Focusing on the concept of explainability, this section will discuss

Chapter 6:

the challenges posed by ’black box’ algorithms in HR decision-making. It will also

examine the legal requirements imposed on AI systems in terms of explainability

and the implications for transparency in automated processes. As in the previous

part, tables will be included to score the articles analysed in this context.

• This part will address privacy issues, particularly in relation to the

Chapter 7:

European GDPR of 2016. It will explore which HR techniques are most suscepti-

ble to privacy breaches and the challenges involved in safeguarding personal data

in AI-driven recruitment and selection processes. Tables will also be presented

here to score the relevant articles discussed.

• This section will explore the importance of maintaining human

Chapter 8:

oversight in decision-making processes that are increasingly automated. It will

discuss the legal and regulatory frameworks that govern human involvement in AI

systems, emphasising the need for a balanced approach between automation and

human judgement. As in the previous sections, tables will be included to score

the articles examined.

• The final part will summarise the findings of the thesis, reflecting on

Chapter 9:

the implications of AI in HR practices. It will also discuss the results presented

in the tables from the previous sections, offering recommendations for ensuring

that AI is used responsibly and ethically in this field.

Chapter 2

RESEARCH METHODOLOGY

This chapter outlines the research methodology adopted for the thesis, which focuses

on the analysis of articles and essays published in scientific journals and on web pages.

The primary objective is to explore the theme of artificial intelligence in human resource

practices, with a specific focus on four key topics: fairness, explainability, privacy, and

human oversight.

2.1 Selection of Sources

2.1.1 Scientific Journals and Papers

To understand the perspective of the international scientific community on the afore-

mentioned topics, articles published in relevant and high-impact scientific journals and

papers were selected. These sources were chosen based on their relevance, citation

impact, and recognised authority in the fields of AI and HR.

The selection process involved identifying leading journals in the domains of AI,

computer science, and HR management. Extensive searches were conducted in vari-

ous academic databases using keywords related to fairness, explainability, privacy, and

human oversight in AI. The inclusion criteria involved the article’s publication date

to ensure the inclusion of recent and relevant studies, the number of citations as an

7 8

CHAPTER 2. RESEARCH METHODOLOGY

indicator of influence, and the journal’s impact factor to ensure high-quality and cred-

ible sources. By focusing on these parameters, the research aimed to incorporate a

diverse and comprehensive collection of scholarly works that reflect the current state of

academic discourse on these critical issues.

2.1.2 Web Articles

To gain an understanding of public opinion on these topics, articles published on web

pages such as ProPublica, Harvard Business Review, and other well-known platforms

were studied. These articles often address concrete business cases and offer a practical

and applied viewpoint on the use of AI in the workplace.

The selection of web articles aimed to capture a broad spectrum of perspectives from

various stakeholders, including industry experts, practitioners, journalists, and the gen-