The impact of artificial intelligence on skills at work in Denmark

• Published date: 01 March 2022
• Author: Jacob Rubæk Holm,Edward Lorenz
• Date: 01 March 2022
• DOI: http://doi.org/10.1111/ntwe.12215

New Technol Work Employ. 2021;00:1–23.

|

1

wileyonlinelibrary.com/journal/ntwe

INTRODUCTION

There is an ongoing debate on the effects of new and emerging technologies on jobs and skills

(Brynjolfsson et al., 2018; Felten et al., 2019; Frey & Osborne, 2017). Much of the literature has

focused on robots and artificial intelligence (AI) and has attempted to assess how the adoption

DOI: 10.1111/ntwe.12215

RESEARCH ARTICLE

The impact of artificial intelligence on skills at

work in Denmark

Jacob RubækHolm1

|

EdwardLorenz1,2

© 2021 Brian Towers (BRITOW) and John Wiley & Sons Ltd.

1IKE/DRUID, Aalborg University

Business School, Aalborg, Denmark

2College of Business and Economics,

University of Johannesburg,

Johannesburg, South Africa

Correspondence

Jacob Rubæk Holm, IKE/DRUID,

Aalborg University Business School,

Fibigerstræde 11, 9220 Aalborg Ø,

Denmark.

Email: jrh@business.aau.dk

Abstract

Based on a unique dataset on the use of artificial in-

telligence (AI) among employees in Denmark, we in-

vestigate within- job relationships between AI use and

skill requirements. We show that the effects of AI are

varied and depend on whether AI is used for providing

orders to humans or providing information for further

human handling and in which occupation it is used. AI

may enhance or augment skills through, for example,

the increased use of high- performance work practices,

or it may increase work pace constraints and reduce

employee autonomy. The results imply that the diffu-

sion of AI can increase inequalities in the labour market

by augmenting skills used in high- skill jobs, although

having relatively more adverse impacts on other jobs.

We use additive noise modelling to establish the likely

direction of causality in our results and find that the di-

rection of causality is from AI use to skill requirements.

KEYWORDS

additive noise modelling, artificial intelligence, autonomy,

constraints, high- performance work practices, job requirements

approach, learning, skills

79

New Technology, Work and Employment. 2022;37:79–101. wileyonlinelibrary.com/journal/ntwe

2

|

of these technologies will affect the employment and skills of different occupational groups.

Although the literature on robotics has drawn on the results of data collected mainly at the sec-

tor and national levels in the 1990s,1. the research on the effects of AI has been more speculative

in nature and has sought to predict future patterns of adoption based on technical assessments of

the susceptibility of different detailed occupations to substitution by AI. This article contributes

to this literature with an analysis of the impact of AI on the skills employees use in their daily

work based on what we believe to be a unique large- scale, employee- level survey with measures

of the use of AI. The survey results allow us to identify two basic types of AI used in daily work,

distinguishing between tasks where the employee receives orders or directions generated auto-

matically by a computer or computerised machinery and tasks where the employee makes use

of information compiled automatically by a computer or computerised machinery for further

decision- making or for advising clients or customers. We thus focus on AI defined as computer

systems performing tasks that normally require human intelligence and on how such systems are

used to automate and augment tasks in jobs. This focus may be contrasted both with ‘true’ AI that

is capable of replicating and outperforming human intelligence in all respects (Agrawal et al.,

2019: 3), and with AI defined more narrowly as a machine learning (ML) prediction technology.

Our way of measuring AI encompasses but is not limited to ML. We find that AI most often aug-

ments skills used in high- skill jobs, whereas other jobs more often become more constrained in

terms of work pace.

The article is structured in the following manner. In Section 2, we present an overview of the

recent economics literature on the adoption of AI and its impact on employment and skills. We

point to some of the limitations of this literature in terms of capturing how AI may transform em-

ployees’ skills in daily work activity as opposed to simply substituting for their skills. In Section

3, we describe the survey design and present descriptive statistics on the adoption of the two

forms of AI use we identify according to the sector and broad occupational category. In Section

4, our econometric analysis shows that the two different forms of AI use have different impacts

on employees’ skills at work and that these impacts may differ across high- skilled, middling and

low- skilled occupational categories. Using additive noise modelling (ANM), we provide indica-

tions that the direction of causality is from the use of AI at work to skill requirements. In Section

5, we conclude by pointing to the importance of our results for education and training systems

and argue that there is a clear need for further micro- level studies on AI and skills.

BACKGROUND TO THE DEBATE ON THE IMPACT OF AI ON

EMPLOYMENT AND SKILLS

Concerns about the impact of automation technologies on employment and skills are far from

new. Keynes (1930) is famously credited with popularising the term ‘technological unemploy-

ment’, referring to technology destroying jobs faster than we can discover new ones. In an article

focusing on the recent debate on the impact of new technology on employment, Autor (2015)

notes that during the 1950s and 1960s, concern in the United States led the Johnson administra-

tion to set up a commission on automation and employment. Braverman’s (1974) book on the la-

bour process gave rise to a large debate during the 1970s and 1980s on whether management has

an interest in pursuing a general strategy of deskilling to increase managerial control. Braverman

distinguished between organisational deskilling and technological deskilling and with respect to

the latter identified the use of numerical control (NC) machine tools as an example of how man-

agement may use automation technologies to increase their control by separating the tasks of

c

b

u

c

j

o

t

o

&

T

T

t

i

t

f

p

t

e

a

a

p

t

d

T

i

l

p

t

i

s

b

e

n

p

o

f

T

s

t

f

i

c

o

80