Artificial Intelligence in Human Resources management

First Author1[0000-1111-2222-3333] and Second Author2[1111-2222-3333-4444]

1
Princeton University, Princeton NJ 08544, USA

Abstract: At a high level, computer science (AI) could be a technology that enables computers to find out
from and make or recommend actions supported previously collected data. In terms of human resources
management, computer science are often applied in many alternative ways to streamline processes and
improve efficiency. There is a considerable gap between the promise and reality of AI in human resource
(HR) management.
This paper identifies some challenges in using data science techniques for HR tasks: complexity of HR
phenomena, constraints imposed by small data sets, accountability questions associated with fairness and
other proper and legal constraints, and possible worst employee reactions to management decisions.
It then proposes practical responses to those challenges supported three overlapping principles—causal
reasoning, randomization and experiments, and employee contribution—that would be both economically
efficient and socially appropriate for using data science within the management of employees.

Keywords: data analysis, human capital, human resource ethics, hiring and recruitment, information
systems, decision-making tools

Introduction:

The speed with which the business rhetoric in management moved from big data (BD) to machine learning
(ML) to computer science (AI) is staggering. The match between the rhetoric and reality may be a different
matter, however. Most companies are struggling to form any progress building data analytics capabilities: 41%
percent of CEOs report that they're not in any respect prepared to form use of recent data analytic tools, and
only 4 percent say that they're “to an oversized extent” prepared (IBM 2018). “AI” conventionally refers to a
broad class of technologies that allow a computer to perform tasks that normally require human cognition,
including decision-making. Our discussion here is narrower, specializing in a sub-class of algorithms within AI
that rely principally on the increased availability of knowledge for prediction tasks. sure enough, there are major
advances within the domains of pattern recognition and tongue processing (NLP) over the last several years.
Deep learning using neural networks has become increasingly common in some data-rich contexts and has
brought us closer to true AI, which represents the flexibility of machines to mimic adaptive human decision
making. Nevertheless, with relevancy the management of employees, where the promise of more sophisticated
decisions has been articulated loudly and infrequently, few organizations have even entered the massive data
stage. Only 22 percent of firms say they need adopted analytics in human resources (LinkedIn 2018), and the
way sophisticated the analytics are in those firms isn't in any respect clear.
The application which can have an effect of AI to HR problems put forward very different challenges. they vary
from practical to conceptual, including the very fact that the character of information science analyses when
applied to people has serious conflicts with criteria societies typically see as important for creating
consequential decisions about individuals. Consider the following:
 a primary problem is that the complexity of HR outcomes, like what constitutes being a “good
employee.” There are many dimensions to it construct, and measuring it with precision for many jobs is
sort of difficult: performance appraisal scores, the foremost widely-used metric, are roundly criticized
for problems of validity and reliability yet as for bias, and plenty of employers doesn’t find it suitable
altogether (Cappelli and Tavis 2017). Any reasonably complex job is interdependent with other jobs
and so individual performance is tough to disentangle from group performance (Pfeffer and Sutton
2006).
 the information sets in human resources tend to be quite small by the standards of information science.
the amount of employees that even an outsized company may have is trivial 3 compared to the quantity
of purchases their customers make, for instance. Moreover, many outcomes of interest are rarely
observed, like employees fired for poor performance. Data science techniques results were very poor
when predicting relatively rare outcomes.
  The outcomes of human resource decisions (such as who gets hired and fired) have such serious
consequences for people and society that concerns about fairness – both procedural and distributive
justice - are paramount. Elaborate legal frameworks constrain how employers must set about making
those decisions. Central to those frameworks is that the concern with causation, which is often absent
from algorithm-based analyses.
 Employment decisions are subject to a spread of complex socio-psychological concerns that exist
among employees, like personal worth and standing, perceived fairness, and contractual and relational
expectations, that affect organizational outcomes likewise as individual ones. As a result, having the
ability to clarify and also to justify the practices one uses is way more important than in other fields.
 Finally, employees are capable of gaming or adversely reacting to algorithmic based decisions. Their
actions, in turn, affect organizational outcomes. as an instance these concerns, consider the utilization
of an algorithm to predict who to rent. As is typical in problems like these, the applying of machine
learning techniques would create an algorithm supported the attributes of employees and their job
performance within the current workforce. whether or not we could demonstrate a causal relationship
between sex and job performance, we would well not trust an algorithm that says hire more white men
because job performance itself could also be a biased indicator, the attributes of this workforce is also
distorted by how we hired within the past (e.g., we hired few women), and both the system and social
norms would create substantial problems for us if we did act thereon.
Uwe Hohgrawe, explains that “we as humans see the data ahead people and use our intelligence to draw
conclusions. Machines don't seem to be intelligent, but we will make them appear intelligent by feeding them
the proper information and technology.
3 Top Applications of AI in HR

Among the various applications of AI within the human resources sector, a number of the primary
changes HRprofessionals should expect to determine involve recruitment and onboarding, employee
experience, process improvement, and also the automation of administrative tasks.

1. Recruitment and Onboarding

Recruitment and Onboarding While many organizations are already starting to integrate AI technology
into their recruiting efforts, the overwhelming majority of organizations aren't. In fact, Deloitte’s 2019
Global Human Capital Trends survey found that only 6 percent of respondents believed that they'd the
best-in-class recruitment processes in technology, while 81 percent believed their organization’s
processes were standard or below standard. For this reason, there are tremendous opportunities for
professionals to adapt their processes and reap the advantages of using this advanced technology.

During the recruitment process, AI may be wont to the advantage of not only the hiring organization but
its job applicants, as well. as an example, AI technology can streamline application processes by
designing more user-friendly forms that employment applicant is more likely to finish, effectively
reducing the amount of abandoned applications. While this approach has made the role of the human
resources department in recruitment much easier, AI also allows for less complicated and more
meaningful applications on the candidate’s end, which has been shown to boost application completion
rates.

Additionally, AI has played a very important role in candidate rediscovery. By maintaining a database of
past applicants, AI technology can analyze the present pool of applicants and identify people who would
be a decent appropriate new roles as they open up. instead of expending time and resources looking for
fresh talent, HR professionals can use this technology to spot qualified employees more quickly and
simply than ever before. Once hiring managers have found the simplest suitable their open positions, the
onboarding process begins.

With the assistance of AI, this process doesn’t need to be restricted to straightforward business hours—a
huge improvement over onboarding processes of the past. Instead, AI technology allows new hires to
utilize human resources support at any time of day and in any location through the utilization of chatbots
and remote support applications. this alteration not only provides employees with the power to travel
through the onboarding process at their own pace, but also reduces the executive burden and typically
ends up in faster integration.

2. Internal Mobility and Employee Retention

Internal Mobility and Employee Retention In addition to improvements to the recruitment process, HR
professionals may also utilize computer science to spice up internal mobility and employee
retention. With the use of personalized feedback surveys and employee recognizing systems, HR
departments can make employee engagement and job satisfaction more accurately today than ever
before. this is often incredibly beneficial considering how important it's to grasp the needs of employees,
however there are several key organizational benefits to having this information, as well.

According to a recent report from the Human Resources association, some AI software can evaluate key
indicators of employee success so as to spot those who should be promoted, thus driving internal
mobility. This has the potential to significantly reduce talent acquiring costs and improve employee
retention rates.

This technology isn't limited to identifying opportunities to market from within, however; it can even
predict who on a team is possibly to quit. Having this data as soon as possible allows HR professionals to
deploy retention efforts before it’s too late, which may strategically reduce employee attrition.

3. Automation of Administrative Tasks

One of the key benefits of leveraging computer science in various human resources processes is really
the identical because it is in other disciplines and industries: Automating low value, easily repeatable
administrative tasks gives HR professionals longer to contribute to strategic planning at the
organizational level. This helps the HR department to develop strategic business partner within their
organizations.
 Smart technologies can automate processes like the administration of advantages, pre-screening
candidates, scheduling interviews, and more. Although each of those functions is vital to the success of a
corporation, closing the tasks involved in such processes is usually time-consuming, and therefore the
burden of those duties often implies that HR professionals have less time to contribute to serving their
employees in additional impactful ways.

Using AI software to automate administrative tasks can reduce this burden. for example, a study by
Eightfold found that HR personnel who utilized AI software performed administrative tasks 19 percent
more effectively than departments that don't use such technology. With the time that's saved, HR
professionals can devote more energy to strategic planning at the organizational level.

Preparing For the Future of Human Resources Management

While it's clear that computer science will still positively shape the sphere of human resources management
within the coming years, HR professionals should even be tuned in to the challenges that they could face. The
most common concerns that HR leaders have focus totally on making AI simpler and safer to use. In fact, the
foremost factor preventing people from using AI at work are security and privacy concerns. Additionally, 31
percent of respondents in Oracle’s survey expressed that they might rather interact with somebody's within the
workplace than a machine. Moving forward, HR professionals will must be prepared to handle these concerns
by staying on top of trends and technology as they evolve and alter. “People should remember the ethical and
privacy questions when using this technology,” Hohgrowe says. “In human resources, [AI] can use sensitive
information to make sensitive insights.” For instance, employees want their organizations to respect their
personal data and enkindle permission before using such technology to assemble information about them.
However organizations also want to feel shielded from data breaches, and HR professionals must take the
acceptable security measures into consideration. To prepare for the long run of human resources management,
professionals should take the mandatory steps to find out about current trends within the field, additionally as
lay a powerful foundation of HR knowledge that they'll depend upon because the profession evolves.

Literature review

Below, we address each of those challenges separately at each stage of what we call the AI Life Cycle:
Operations – Data Generation – Machine Learning – deciding. We depend upon key ideas from Evidence-Based
Management (EB Mgmt) - a theory driven causal analysis of “small data” (Barends and Rousseau 2018; Pfeffer
and Sutton 2006; Rousseau 2014). We then suggest how, given these constraints, we would make progress
within the application of machine learning tools to HR. Specifically, we concentrate on the role of causal models
in machine learning (Pearl 2009, 2018). Establishing causation is central to concerns about fairness, which are
fundamental to creating decisions about employees, and machine learning-based algorithms typically struggle
thereupon challenge. We also suggest that randomization will be useful as a call process (Denrell, Fang, and Liu
2015; Liu and Denrell 2018), given its perceived fairness and therefore the difficulty that person analysing will
otherwise have in making fair and valid decisions. We base our arguments on knowledge of up to date practice
additionally as on interactions with practitioners.

The AI Life Cycle:

Fig. 1. depicts a conventional AI Life Cycle: Operations, Data Generation, Machine Learning, and Decision-Making

“Operations” constitute the phenomenon of interest, like how a company hires employees. one among the
explanations for the interest in applying data science tools to human resources is because HR performs such a
lot of operations and then much money is involved in them. within the US economy as an entire, roughly 60
percent of all spending is on 5 labor. in commission industries, the figure is way higher (MLR 2017). Below are
the foremost common operations in human resources with corresponding prediction tasks for workforce
analytics:

HR operation Prediction task

Recruiting – identifying possible candidates and Are we securing good candidates?
persuading them to apply
Selection – choosing which candidate should receive Are we offering jobs to those who will be the best
job offers employees?
On-boarding - bringing an employee into an Which practices cause new hires to become useful
organization faster?
Training What interventions make sense for which
individuals, and do they improve performance?
Performance management – identifying good and Do our practices improve job performance?
bad performance
Advancement – determining who gets promoted Can we predict who will perform best in new roles?
Retention Can we predict who is likely to leave and manage the
level of retention?
Employee benefits Can we identify which benefits matter most to
employees to know what to give them and what to
recommend when there are choices, and what are the
effects of those benefits (e.g., do they improve
recruiting and retention)?

Each of those operations involves administrative tasks, each affects the performance of the organization in
important ways, and every includes specific offices, job roles, written instructions and guidelines to execute
additionally because the actual activities and interactions of all parties. These operations produce volumes of
knowledge, within the kind of texts, recordings, and other artifacts. As operations move to the virtual space,
many of those 6 outputs are within the type of “digital exhaust,” which is trace data on digital activities (e.g.
online job applications, skills assessment) which will be accustomed build recruiting algorithms.

Human resource information systems, applicant tracking systems, digital exhaust, and other markers are very
important inputs for the “data generation” stage. Typically, this input should be extracted from multiple
databases, converted to a typical format, and joined together before analysis can occur.

By “machine learning” (ML) we discuss with a broad set of techniques that may adapt and learn from data to
form algorithms that perform better and better at a task, typically prediction. Within business contexts, the
foremost common application of machine learning technologies has been “supervised” applications, during
which an information scientist creates a machine learning algorithm, determines the foremost appropriate metric
to find its accuracy, and performing training for the algorithm using the training sample. a number of the
foremost commonly used prediction algorithms, like logistic regression and random forest infer the result
variable of interest from statistical correlations among observed variables. The accuracy of preliminary models
is assessed on the event sample until it stabilizes at some acceptable level. the ultimate model is run on the test
sample; the accuracy of the predictions on this sample is that the ultimate indicator of the model’s quality.

For hiring, as an example, we would see which applicant characteristics are related to better job performance
and use that to pick out candidates within the future. “Algorithmic management,” How to use algorithms to
guide incentives and other tools for “contacting” platform workers and contractors within the contract (Lee et al
2015), is used for regular employees (e.g., Netessine and Yakubovich 2012). At present, this can be principally
the case in making recommendations. IBM, for instance, uses algorithms to advise employees on what training
be for them to require, supported the experiences of comparable employees; the seller Quine uses the career
progression of previous employees to create recommendations to client’s employees about which career moves
be for them. Vendors like Benefit focus develop customized recommendations for employee benefits, much
within the same way that Netflix recommends content supported consumer preferences or Amazon recommends
products supported purchasing or browsing behavior. 7 These algorithms differ in some important ways from
traditional approaches employed in HR. In applied psychology, the sector that historically focused the foremost
attention on human resource decisions, research on hiring, say, would test separate explanatory hypotheses
about the link between individual predictors and job performance. The researcher picks the hypothesis to look at
and therefore the variables with which to look at it. This process produces lessons for hiring, one test at a time,
e.g., the link between mental testing scores and job performance, then in another exercise, the connection
between education and job performance, so forth.

Machine learning, in contrast, generates one algorithm that creates use of the many variables. The variables
might not be within the cannon of the theoretical literature related to the subject, and also the researcher isn't
hypothesizing or indeed even examining the connection between anybody variable and therefore the outcome
being predicted. Indeed, one amongst the attractions of ML is its investigation of non-traditional factors because
the goal is to make a higher prediction instead of advancing the speculation of the sector within which the
researcher relies by providing evidence on particular hypotheses.

“Decision-making,” the ultimate stage, deals with the way during which we use insights from the machine
learning model in everyday operations. within the area of human resource decisions, individual managers may
have more discretion now in how they use empirical evidence from data science and other models than they did
within the heyday of the nice corporations when hiring and other practices were standardized across a complete
company. Managers today typically have the choice of ignoring evidence about predictions, using it as they see
fit, and generating their own data about actions like hiring within the type of interviews they structure
themselves.

Conclusion: While general-purpose AI continues to be a protracted shot in any domain of human action, the
speed of progress towards specialized AI systems in health care, industry, social media, advertising and
marketing is considerable. Far less progress has been made in issues round the management of employees even
on the primary step of the AI path, which is decisions guided by algorithms. We can find four reasons for this:
complex HR phenomena, acquiring data challenges HR operations, fairness and legal conditions, and employee
reactions to AI-management. Causal reasoning is that the ABCs relevant to addressing these challenges across
the stages of the AI Life Cycle. Because the creation of algorithms relies on association instead of causation, an
absence of notions of causation makes it rather more difficult to make the datasets needed for analysis: we want
more data because we don't know what to settle on. Causal reasoning also helps greatly with problems with
fairness and explainability. the advantages of causal reasoning do include costs. Employers must first accept
the greater costs (based on the requirement for more data) and lower predictive power from algorithms where
we don't have causal models, and that they must work to develop consensus about causal assumptions before of
modeling. These challenges explain why the info science community is kind of skeptical about causally
reasoning AI systems. Randomization may be a second principle which will help with algorithmic-based
decisions. First, randomizing the inputs into an algorithm is cherish experimentation and might help to
ascertain causality. Second, if we random choose an HR results with the probability shown by an algorithm
which cannot predict outcomes that much accuracy acknowledges the inherently stochastic nature of HR
outcomes and unavoidable inaccuracy of algorithms. Employees may perceive such randomization—like
flipping a coin—to produce fairer outcomes under uncertainty. Formalizing processes is additionally necessary
to make reasonable algorithms. It ensures that the parties are attentive to the assumptions built into any
algorithms, the prices of building them, and also the likely challenges from employees who are adversely
littered with them. We hope that the concept and practical insights of this paper will move AI-management in
HR forward in both places, those of efficiency and appropriateness.

Reference:
1. Cowgill, Bo (2017) The Labor Market Effects of Hiring through Machine Learning Working Paperowgill, Bo.
2018. “Bias and Productivity in Humans and Algorithms. Theory and Evidence from Résumé Screening.”
Working paper
2. Acktar, Reese, Dave Winsborough , Uri Ort , Abigail Johnson , Tomas ChamorroPremuzic. 2018. Detercting
the Dark Side of Personality Using Social Media. Personality and Individual Differences, 132:90-97.
3. Barends, Eric and Denise M. Rousseau. 2018. Evidence-Based Management: How to Use Evidence to Make
Better Organizational Decisions. Kogan Page
4. Bloomberg, J. 2018. Don’t Trust Artificial Intelligence? Time to Open the AI Black Box. Forbes. Last accessed
at https://www.forbes.com/sites/jasonbloomberg/2018/09/16/dont-trust-artificialintelligence-time-to-open-the-
ai-black-box/#577a14153b4a on Nov 27, 2018.
5. Cappelli, Peter. 2017. “There’s No Such Thing as Big Data in HR.” Harvard Business Review. June
6. Lee, M. K., Kusbit, D., Metsky, E., & Dabbish, L. 2015. Working with machines: The impact of algorithmic,
data-driven management on human workers. Proceedings of the 33rd Annual ACM SIGCHI Conference: 1603-
1612. Begole, B., Kim, J., Inkpen, K & Wood, W (Eds.), New York, NY: ACM Press.
7. Lind, E. Allan and Kees Van den Bos. 2002. “When Fairness Works: Toward a General Theory of Uncertainty
Management.” Research in Organizational Behavior 24: 181-223.
8. LinkedIn. 2018. The Rise of HR Analytics
9. Meyer, David. 2018. Amazon Reportedly Killed an AI Recruitment System Because It Couldn't Stop the Tool
from Discriminating Against Women. Fortune. October 10th. http://fortune.com/2018/10/10/amazon-ai-
recruitment-bias-women-sexist/
10. Pearl, Judea. 2018. The Book of Why: The New Science of Cause and Effect. Basic Books.
11. Pfeffer, Jeffrey and Robert I. Sutton. 2006. Hard Facts, Dangerous Half-Truths and Total Nonsense: Profiting
from Evidence-Based Management. Harvard Business Review Press.
12. Rousseau, Denise (Editor). 2014. The Oxford Handbook of Evidence-Based Management. Oxford University
Press.
13. Spielkamp, Michael. 2017. Inspecting Algorithms for Bias. MIT Technology Review. June 12.
https://www.technologyreview.com/s/607955/inspecting-algorithms-forbias/
14. Tucker, Catherine. 2017. “Privacy, Algorithms, and Artificial Intelligence.” In The Economics of Artificial
Intelligence: An Agenda. Edited by Ajay K. Agrawal, Joshua Gans, and Avi Goldfarb. University of Chicago
Press. Forthcoming. http://www.nber.org/chapters/c14011.
15. Walsh, David. 2017. Employment Law for Human Resource Practice. Mason, Ohio: South-Western.
16. Bock, Laslo. 2015. Work Rules! Insights from Inside Google That Will Transform How You Live and Lead.
Hachette Book Group
17. Prasanna Tambe, , Peter Cappelli, and Valery Yakubovich, Artificial intelligence in human resources
management: Challenges and a path forward, https://journals.sagepub.com/, (2019)
18. Jade Taryn Graham, https://blog.sage.hr/five-ways-ai-artificial-intelligence-is-disrupting-human-resources-hr-
management/
19. By Scott W. O'Connor, Artificial Intelligence in Human Resource Management,
https://www.northeastern.edu/graduate/blog/artificial-intelligence-in-human-resource-management/