Robots & AI Exposure and Wage Inequality: A Within Occupation Approach

Jaccoud, Flo encia
Wo king Pape
Robo s & AI Exposu e and Wage Inequali y: A Wi hin
Occupa ion App oach
UNU-MERIT Wo king Pape s, No. 2025-013
P o ided in Coope a ion wi h:
Maas ich Economic and Social Resea ch Ins i u e on Inno a ion and Technology (UNU-MERIT),
Uni ed Na ions Uni e si y (UNU)
Sugges ed Ci a ion: Jaccoud, Flo encia (2025) : Robo s & AI Exposu e and Wage Inequali y: A Wi hin
Occupa ion App oach, UNU-MERIT Wo king Pape s, No. 2025-013, Uni ed Na ions Uni e si y (UNU),
Maas ich Economic and Social Resea ch Ins i u e on Inno a ion and Technology (UNU-MERIT),
Maas ich ,
h ps://doi.o g/10.53330/EAJL3597
This Ve sion is a ailable a :
h ps://hdl.handle.ne /10419/326942
S anda d-Nu zungsbedingungen:
Die Dokumen e au EconS o dü en zu eigenen wissenscha lichen
Zwecken und zum P i a geb auch gespeiche und kopie we den.
Sie dü en die Dokumen e nich ü ö en liche ode komme zielle
Zwecke e iel äl igen, ö en lich auss ellen, ö en lich zugänglich
machen, e eiben ode ande wei ig nu zen.
So e n die Ve asse die Dokumen e un e Open-Con en -Lizenzen
(insbesonde e CC-Lizenzen) zu Ve ügung ges ell haben soll en,
gel en abweichend on diesen Nu zungsbedingungen die in de do
genann en Lizenz gewäh en Nu zungs ech e.
Te ms o use:
Documen s in EconS o may be sa ed and copied o you pe sonal
and schola ly pu poses.
You a e no o copy documen s o public o comme cial pu poses, o
exhibi he documen s publicly, o make hem publicly a ailable on he
in e ne , o o dis ibu e o o he wise use he documen s in public.
I he documen s ha e been made a ailable unde an Open Con en
Licence (especially C ea i e Commons Licences), you may exe cise
u he usage igh s as speci ied in he indica ed licence.
h ps://c ea i ecommons.o g/licenses/by-nc-sa/4.0/
#2025-013
Robo s & AI Exposu e and Wage Inequali y:
A Wi hin Occupa ion App oach
Flo encia Jaccoud
Published 22 Ap il 2025
DOI: h ps://www.doi.o g/10.53330/EAJL3597
Maas ich Economic and social Resea ch ins i u e on Inno a ion and Technology (UNU-MERIT)
email: in o@me i .unu.edu | websi e: h p://www.me i .unu.edu
Boschs aa 24, 6211 AX Maas ich , The Ne he lands
Tel: (31) (43) 388 44 00
UNU-MERIT Wo king Pape s
ISSN 1871-9872
Maas ich Economic and social Resea ch Ins i u e on Inno a ion and Technology
UNU-MERIT
UNU-MERIT Wo king Pape s in end o dissemina e p elimina y esul s o esea ch ca ied
ou a UNU-MERIT o s imula e discussion on he issues aised.
h ps://c ea i ecommons.o g/licenses/by-nc-sa/4.0/
Robo s & AI Exposu e and Wage Inequali y: A Wi hin
Occupa ion App oach∗
Flo encia Jaccoud†
Abs ac
This pape examines he linkages be ween occupa ional exposu e o ecen au-
oma ion echnologies and inequali y ac oss 19 Eu opean coun ies. Using da a om
he Eu opean Union S uc u e o Ea nings Su ey (EU-SES), a ixed-e ec s model is
employed o assess he associa ion be ween occupa ional exposu e o a i icial in elli-
gence (AI) and o indus ial obo s— wo dis inc o ms o au oma ion—and wi hin-
occupa ion wage inequali y. The analysis e eals ha occupa ions wi h highe expo-
su e o obo s end o ha e lowe wage inequali y, pa icula ly among wo ke s in he
lowe hal o he wage dis ibu ion. In con as , occupa ions mo e exposed o AI exhibi
g ea e wage dispe sion, especially a he op o he wage dis ibu ion. We a gue ha
his dispa i y a ises om di e ences in how each echnology complemen s indi idual
wo ke abili ies: obo - ela ed asks o en complemen ou ine physical ac i i ies, while
AI- ela ed asks end o ampli y he p oduc i i y o high-skilled, cogni i ely in ensi e
wo k.
Keywo ds: Inequali y; Robo s; A i icial In elligence; Occupa ions
JEL Codes: J31, O33, J24.
∗We a e g a e ul o Filippo Bon adini, Tommaso Cia li, Rinaldo E angelis a, Neil Fos e -McG ego ,
Önde Nomale , Fabien Pe i , Guido Pialli, Eka e ina P y ko a, Ma eo Tubiana, and Ba Ve spagen o
commen s on p e ious e sions o his pape . We a e ex emely hank ul o S ijn B oecke, Suga Cha u edi,
Tommaso Cia li, Alexand e Geo gie , Deyu Li, Fabien Pe i , Eka e ina P y ko a, and Jacopo S accioli o
kindly sha ing he da a on AI and/o obo s exposu e used in his pape . We a e also g a e ul o he
pa icipan s o he CORA 2024 Con e ence on Robo s and Au oma ion; Social Si ua ion Moni o Resea ch
Semina 2023 ‘The u u e o wo k: a i icial in elligence and i s labo ma ke and social impac s’; he 10 h
Ph.D. Wo kshop in Economics o Inno a ion, Complexi y and Knowledge, as well as he pa icipan s a he
UNU-MERIT 2023 Resea ch Week o he insigh ul commen s ha helped imp o e he pape . The au ho
acknowledge he suppo by he Eu opean Union’s Ho izon 2020 esea ch and inno a ion p og am unde
g an ag eemen No. 101004703 -PILLARS (Pa hways o inclusi e labo ma ke s).
†Came ino Uni e si y / UNU-MERIT. Email: [email p o ec ed]u.edu.
1 In oduc ion
O e he pas h ee decades, income inequali y wi hin coun ies has isen by an a e age o
10% ac oss OECD na ions (OECD,2015). S uc u al and ins i u ional changes, globaliza-
ion, and echnological ad ances a e among he key ac o s con ibu ing o his end. Gi en
he signi ican job pola iza ion esul ing om he widesp ead adop ion o compu e iza ion
in de eloped economies, changes in he occupa ional s uc u e ha e eme ged as a p ominen
d i e o wage inequali y (Au o e al.,2003;Goos and Manning,2007). As a esul , highly
skilled wo ke s in he Uni ed S a es and se e al Eu opean coun ies ha e seen a ela i e wage
inc ease compa ed o low-skilled wo ke s, widening he wage gap ac oss di e en occupa ions
(Acemoglu and Au o ,2011;Goos e al.,2014;Co es,2016).
Ye , ecen li e a u e emphasizes ha wi hin-occupa ion wage dispa i ies also play a sig-
ni ican ole in o e all inequali y. Fo ins ance, Kim and Sakamo o (2008) demons a e ha
in he U.S., o e hal o he ise in o al inequali y om 1983-1985 o 2000-2002 can be
a ibu ed o wi hin-occupa ion inequali y. This end has pe sis ed o e ime, as shown by
Mishel e al. (2013) o he pe iod om 1979 o 2007. In Eu ope, Ake man e al. (2013)
ound ha wi hin-occupa ion inequali y explained mo e han 70% o he ise in wage in-
equali y in Sweden om 2001 o 2007, a pa e n echoed ac oss o he Eu opean na ions
(Fe nández-Macías and A anz-Muñoz,2020; an de Velde,2020). Fu he mo e, an de
Velde (2020) inds ha wi hin-occupa ion wage inequali y ends o be highe in jobs wi h a
la ge p opo ion o non- ou ine asks.
This pape examines he associa ion be ween he exposu e o occupa ions o obo s and
AI and wi hin-occupa ion wage inequali y in 19 Eu opean coun ies. In his con ex , we
de ine exposu e as he ela ionship be ween echnology and he asks associa ed wi h an
occupa ion. Pu simply, when he e is a g ea e o e lap be ween he asks ha a pa icu-
la echnology can pe o m and hose in ol ed wi hin an occupa ion, he exposu e o ha
occupa ion will be highe .
While many s udies ha e ocused on au oma ion and i s ela ionship wi h ask con en ,
pa icula ly ou ine asks, eme ging au oma ion echnologies a ge dis inc ask ypes. Fo
example, AI has a g ea e ocus on cogni i e and complex asks and is hus mo e ele an o
high-wage occupa ions (Webb,2020;Fel en e al.,2021;Geo gie and Hyee,2021;Engbe g
e al.,2024). Robo s, in con as , ypically pe o m manual asks such as assembly and
welding (Squiccia ini and S accioli,2022;Mon obbio e al.,2022;P y ko a e al.,2024). This
dis inc ion in ask con en is c ucial o unde s anding wi hin-occupa ion wage dispa i ies
(Jung and Me cenie ,2014; an de Velde,2020). Occupa ions mo e exposed o AI in ol e
cogni i e asks, in which he e is mo e he e ogenei y in indi iduals’ abili ies and pe o mance,
esul ing in g ea e wage dispe sion. On he o he hand, jobs mo e exposed o obo s end
o be mo e s anda dized, limi ing wo ke s’ au onomy and a ia ion in ask pe o mance,
educing wage dispa i y wi hin hose occupa ions.1Based on his di e en ia ion in ask
con en , we hypo hesize a posi i e associa ion be ween AI exposu e and wi hin-occupa ion
wage dispe sion, and a nega i e associa ion o occupa ions mo e exposed o obo s. Thus we
a gue ha hese echnologies se e as p oxies o he unde lying ask s uc u e o occupa ions.
1I is wo h no ing ha , while obo s and AI can be analy ically dis inguished, he wo echnologies ha e
become inc easingly in e wined–pa icula ly since he 2010s–as AI de elopmen s a e p og essi ely in eg a ed
in o obo ic sys ems (Jaccoud e al.,2024).
1

We use da a om he Eu opean Union S uc u e o Ea nings Su ey (EU-SES) p o ided
by Eu os a o 19 Eu opean coun ies. To cap u e exposu e o hese echnologies, we use
he AI and obo occupa ional sco es de eloped by Webb (2020). By using pa en da a,
he au ho links he asks ha hese echnologies pe o m wi h he job ask desc ip ions in
he S anda d Occupa ional Classi ica ion 2010 (SOC-2010). We hen map he SOC-2010
classi ica ions o he In e na ional S anda d Classi ica ion o Occupa ions 2008 (ISCO-08)
in conco dance wi h ou da abase’s classi ica ion sys em. Ou empi ical s a egy elies on a
ixed-e ec model, wi h he dependen a iable being he loga i hm o he wage gap a he
2-digi occupa ional le el. The main independen a iables a e he measu es o obo and
AI exposu e de i ed om Webb (2020). Addi ionally, we conduc se e al obus ness checks
by inco po a ing al e na i e exposu e indices o obo s and AI, ensu ing he obus ness o
ou indings agains di e en speci ica ions.
Ou main indings suppo ou hypo hesis, indica ing ha occupa ions ha a e mo e
exposed o obo s a e hose in which he e is a lowe wage dispa i y, and his is mainly
d i en by he bo om hal o he wage dis ibu ion. Con e sely, he ones mo e exposed o
AI a e associa ed wi h la ge wi hin-occupa ion wage inequali y, pa icula ly in he uppe
hal o he wage dis ibu ion.
This pape con ibu es o he ex ensi e body o esea ch on au oma ion and inequali y.
While s udies such as Au o e al. (2003); Goos and Manning (2007); Acemoglu and Au o
(2011); Goos e al. (2014); Kal enbe g and Fos e -McG ego (2020); an de Velde (2020)
ha e p ima ily examined ICT- ela ed au oma ion, ou s udy b oadens he scope by analyzing
he oles o bo h obo s and AI in wage inequali y. Ou esul s align wi h an de Velde
(2020), which inds a posi i e link be ween non- ou ine ask occupa ions and wage dispe sion
and simila ly sugges s a nega i e ela ionship be ween manual asks and wage a iabili y,
mi o ing he e ec s seen wi h obo exposu e. Al hough he e ha e been se e al a emp s o
s udy he ela ionship be ween obo s and inequali y, hey ha e mainly ocused on be ween-
occupa ion inequali y. By ocusing on he manu ac u ing sec o and examining he impac
o changes in he occupa ional s uc u e, B all and Schmid (2023) ound ha obo adop ion
p ima ily inc eases wage inequali y h ough a composi ional e ec in Ge many. Simila ly,
Ba h e al. (2020) inds ha in No way, while obo pene a ion gene ally aises wages, he
e ec is mos p onounced o manage s and STEM wo ke s.
Ou esul s should be in e p e ed wi h cau ion o wo main easons. Fi s , he measu es
o AI and obo exposu e used in his s udy do no e lec ac ual echnology adop ion, bu
a he he deg ee o which occupa ions a e associa ed wi h hese echnologies based on hei
ask con en . As such, we ea exposu e o AI and obo s as p oxies o occupa ional ask
s uc u e, no as di ec d i e s o inequali y. This means ha he obse ed associa ions
may be in luenced by o he media ing ac o s–such as p oduc i i y e ec s– ha can shape
wage ou comes. Second, due o he limi a ions o ou da a, we canno make any causal
claims ega ding he ela ionship be ween echnology exposu e and wage inequali y. Despi e
hese limi a ions, he pape con ibu es aluable empi ical insigh s by sys ema ically link-
ing occupa ional ask cha ac e is ics o pa e ns o wage dispe sion—an angle ha emains
unde explo ed in he exis ing li e a u e.
The es o he pape is o ganized as ollows. Sec ion 2 e iews ele an li e a u e on
AI and indus ial obo s. Sec ion 3desc ibes ou da a sou ces and key a iables. Sec ion 4
p o ides desc ip i e e idence o hese echnologies and hei ela ionship wi h labo ma ke .
2
Sec ion 5, p esen s he empi ical s a egy and esul s. Finally, Sec ion 6concludes.
2 Backg ound li e a u e
Task con en o ecen au oma ion echnologies
The seminal wo k by Au o e al. (2003) shi ed ocus o he ask con en o occupa ions,
highligh ing i s impo ance in unde s anding how echnology eshapes employmen .2F om
his pe spec i e, occupa ions in ol ing asks ha ollow explici ules can be easily codi ied
and au oma ed. Consequen ly, ou ine asks a e he mos suscep ible o au oma ion. This
insigh gained ac ion in bo h heo e ical and empi ical li e a u e, spa king he de elopmen
o a ious indica o s o measu e occupa ional exposu e o au oma ion (Au o e al.,2003;
Goos e al.,2009;Acemoglu and Au o ,2011;Au o ,2013;Goos e al.,2014;Nedelkoska
and Quin ini,2018;Ma colin e al.,2019;Fos e -McG ego e al.,2021;G ego y e al.,
2022). Ne e heless, di e en echnologies a e designed o dis inc ac i i ies, leading o
a ying associa ions wi h asks and occupa ions. Mo e ecen au oma ion echnologies, such
as indus ial obo s and AI, pe o m di e en ac i i ies and, he e o e, ela e di e en ly o
wo k o ce.3
In ecen yea s, ad ancemen s in AI ha e been os e ed by signi ican imp o emen s in
da a mining and compu a ional powe , which ha e di ec ly con ibu ed o he p og ess o
machine lea ning echnologies (B ynjol sson e al.,2021;Zolas e al.,2021;Bonney e al.,
2024). Supe ised lea ning sys ems o m he backbone o AI, whe e machines a e ained
o p edic ou comes using as da abases. Supe ised lea ning, he co e o AI, enables
machines o p edic ou comes using la ge da ase s. This ema kable p og ess has led o
he de elopmen o gene a i e AI based on ex ensi e na u al language models. Re lec ing
his shi , he O ganiza ion o Economic Co-ope a ion and De elopmen (OECD) ecen ly
upda ed i s de ini ion o AI, desc ibing i as “a machine-based sys em ha , o explici o
implici objec i es, in e s om inpu how o gene a e ou pu s such as p edic ions, con en ,
ecommenda ions, o decisions ha can in luence physical o i ual en i onmen s. Di e en
AI sys ems a y in hei le els o au onomy and adap i eness a e deploymen ” (OECD,
2023).
Due o i s ans o ma i e po en ial and c oss-indus y applica ions, AI is o en seen as a
gene al-pu pose echnology (GPT) (B ynjol sson e al.,2017;T aj enbe g,2018;Filippucci
2I is wo h no ing ha he discussion on echnology and i s implica ions o employmen and wo k o ce
composi ion is longs anding. Fo a de ailed su ey o he li e a u e, see Pi a and Vi a elli (2018) and Au o
(2022).
3I is impo an o no e ha ou dis inc ion be ween AI and obo s ocuses on he echnological cha ac-
e is ics o hese ools and hei associa ed occupa ional exposu e. Howe e , ecen esea ch has emphasized
ha digi al ans o ma ion a wo k also includes he inc easing use o digi al pla o ms, algo i hmic manage-
men , and moni o ing sys ems, pa icula ly a ec ing low-skilled and manual occupa ions (Fe nandez Macias
e al.,2023;Filippi e al.,2023;U zì B anca i e al.,2023). These de elopmen s o en eshape wo k o ganiza-
ion by in luencing ask alloca ion, decision-making, and con ol mechanisms–wi hou necessa ily in ol ing
physical obo s o ad anced AI sys ems. Fo example, wo ke s in logis ics and deli e y may be subjec o
algo i hmic con ol and pe o mance moni o ing ia digi al pla o ms, e lec ing a di e en bu signi ican
dimension o digi al exposu e. While no explici ly cap u ed in ou cu en amewo k, hese mechanisms
a e pa o he b oade landscape o echnological change.
3
e al.,2024).4I combines angible inpu s, such as compu ing powe and IT in as uc u e,
wi h in angibles like so wa e, da a, and skilled labo (e.g., la ge da ase s a e essen ial o
aining algo i hms, along wi h expe ise om p og amme s, da a scien is s, and IT p o es-
sionals) (Co ado e al.,2021;Filippucci e al.,2024). The combina ion o all hese inpu s
esul s in a echnology ha pe o ms cogni i e ac i i ies, ypically associa ed wi h high-
skilled wo ke s, which is an aspec ha dis inguishes his echnology om o he s (Acemoglu
e al.,2022;Cza ni zki e al.,2023).
Despi e me hodological di e ences, esea ch consis en ly shows ha occupa ions in ol -
ing p oblem-sol ing, easoning, and pe cep ion–i.e., non- ou ine cogni i e asks–a e mo e
exposed o AI han o he ypes o occupa ions (Webb,2020;Fel en e al.,2021;Geo gie
and Hyee,2021;Engbe g e al.,2024). Empi ical e idence indica es ha i ms highly exposed
o AI end o pos mo e AI- ela ed job openings while educing non-AI hi ing (Acemoglu
e al.,2022). Geo gie and Hyee (2021) u he emphasize his end, inding a posi i e
co ela ion be ween inc eases in AI- ela ed job pos ings and occupa ional exposu e o AI
ac oss 36 sec o s, unde sco ing he demand o AI-speci ic skills (F ank e al.,2019;Webb,
2020). Howe e , i emains challenging o de e mine ex-an e whe he AI exposu e leads o
he subs i u ion o augmen a ion o ce ain occupa ions, and mos indica o s o AI exposu e
do no ye e lec his dis inc ion (Gua ascio e al.,2023).
Al hough AI adop ion is s ill a an ea ly s age, p elimina y s udies sugges ha i has
no signi ican ly impac ed o e all employmen in he U.S., hough i may posi i ely a ec
employmen a egional o i m le els (Acemoglu e al.,2022;Gua ascio e al.,2023;Damioli
e al.,2024). Fu he , i is associa ed wi h highe wages o high-skilled wo ke s (E ns e al.,
2019;Fel en e al.,2019;Ozgul e al.,2024).
In con as , indus ial obo s pe o m a di e en se o asks. Mo e ecen ly, adi ional
mobile unc ions ha e been enhanced h ough imp o ed sensing capaci ies and in o ma ion-
sha ing wi h o he machines. Indus ial obo ics can execu e a ious ope a ions, including
objec manipula ion, pain ing, wi e welding, and assembly, among o he s, implying a s ong
connec ion o manual asks. Fo example, occupa ions mos exposed o his echnology
ypically in ol e asks such as mo ing objec s in ac o ies and welding. This is ansla ed
in o obo s pe o ming mo e physical and manual ac i i ies. Fo ins ance, Squiccia ini and
S accioli (2022) highligh he high exposu e o obo s in occupa ions such as ‘Hand and
pedal ehicle d i e s’ (ISCO-08 9331), and ‘Vehicle cleane s’ (ISCO-08 9122) among o he s.
Empi ical s udies show ha obo s ha e a skilled-biased echnological change (SBTC) na-
u e, subs i u ing mos ly low-skilled occupa ions (G ae z and Michaels,2018;Webb,2020;
Fe nández-Macías and A anz-Muñoz,2020;Mon obbio e al.,2022).
In line wi h he li e a u e p esen ed abo e, Figu es 2.1a o 2.1d co obo a e he di e en
associa ions o hese echnologies be ween di e en asks. We obse e a posi i e associa ion
be ween non- ou ine cogni i e in ensi e occupa ions and AI exposu e, while he opposi e is
obse ed o obo s. Con e sely, Figu e 2.1c indica es ha he e is no co ela ion be ween
non- ou ine manual asks and AI exposu e, while a posi i e associa ion is obse ed o he
case o obo exposu e.
4Some schola s ques ion he cha ac e iza ion o AI as a GPT, ins ead a guing ha i should be unde s ood
mo e as a sys em (Vannuccini and P y ko a,2024).
4
Figu e 2.1: Co ela ion be ween obo and AI exposu e and ask con en o occupa ions
(a) AI Non- ou ine Cogni i e (b) Robo Non- ou ine Cogni i e
(c) AI Non- ou ine Manual (d) Robo Non- ou ine Manual
No es: The AI and obo exposu e a e buil using he indices p o ided by Webb (2020) a he SOC-2010 and con e ed in o
ISCO-08. The ask con en measu es a e de i ed om O*NET da abase, ca ego ized a he SOC le el. Following Ha dy e al.
(2018), we selec ask- ela ed i ems ha cap u e ou ine and non- ou ine ac i i ies, dis inguishing be ween cogni i e and manual
dimensions. Task in ensi y sco es o each SOC occupa ion a e hen calcula ed by a e aging he impo ance sco es wi hin each
ca ego y. Finally, occupa ions a e mapped o ISCO-08 classi ica ions.
Figu e 2.2 p o ides u he insigh s in o he ela ionship be ween echnology exposu e
and wage dis ibu ion by depic ing he a e age exposu e o obo s and AI by wage deciles.5
The igu e e eals an in e se ela ionship be ween obo exposu e and wage deciles, as illus-
a ed by he downwa d-sloping cu e. Lowe -wage deciles show highe exposu e o obo s,
while highe -wage occupa ions display minimal exposu e. This pa e n aligns wi h he na u e
o indus ial obo s, which a e p ima ily u ilized o manual asks like welding, assembly,
and pain ing—ac i i ies gene ally associa ed wi h lowe -skilled oles. Consequen ly, his ob-
se a ion suppo s indings in he li e a u e indica ing ha low-skilled occupa ions a e mo e
suscep ible o au oma ion h ough obo ics (Chiacchio e al.,2018;G ae z and Michaels,
2018;Acemoglu and Res epo,2020;Dau h e al.,2021). Con e sely, he upwa d-sloping
end o AI exposu e sugges s ha a e age exposu e ises ac oss highe wage deciles. Oc-
cupa ions a he uppe end o he wage dis ibu ion demons a e g ea e a e age exposu e
o AI, which is consis en wi h he discussion abo e.
5Re e o Sec ion 3 o he de ails on he compu a ion o he indexes.
5
Table 3.2: Desc ip ion o a iables used in he inal da ase
Va iable name Va iable de ini ion
Log o he wage gap p90/p10 Log eal hou ly wage gap be ween he wage o he 90 h
pe cen ile and 10 h pe cen ile a he
occupa ion-coun y-yea cell
Log o he wage gap p90/p50 Log eal hou ly wage gap be ween he wage o he 90 h
pe cen ile and 50 h pe cen ile a he
occupa ion-coun y-yea cell
Log o he wage gap p50/p10 Log eal hou ly wage gap be ween he wage o he 50 h
pe cen ile and 10 h pe cen ile a he
occupa ion-coun y-yea cell
AI exposu e Webb Exposu e o AI o occupa ion obased on Webb (2020)
Robo exposu e Webb Exposu e o obo s o occupa ion obased on Webb (2020)
Robo exposu e Mon obbio e al Exposu e o obo s o occupa ion obased on Mon obbio
e al. (2022)
AI exposu e P y ko a e al Exposu e o AI o occupa ion obased on P y ko a e al.
(2024)
Robo exposu e P y ko a e al Exposu e o AI o occupa ion obased on P y ko a e al.
(2024)
AI exposu e Fel en e al Exposu e o AI o occupa ion obased on Fel en e al.
(2019)
AI exposu e OECD Exposu e o AI o occupa ion obased on Geo gie and
Hyee (2021)
AI exposu e Engbe g e al Exposu e o AI o occupa ion obased on Engbe g e al.
(2024)
N o o 4-digi occupa ions Numbe o 4-digi s occupa ions wi hin majo ISCO 2-digi
g oup
S d o ou ine cogni i e in ensi y S anda d de ia ion o he in ensi y o ou ine cogni i e
ac i i ies a 4-digi le el o occupa ion o
S d o ou ine manual in ensi y S anda d de ia ion o he in ensi y o ou ine manual
ac i i ies a 4-digi le el o occupa ion o
HHI o high-skilled occupa ions He indahl index o high-skilled employmen ( e ia y
educa ion o mo e) o occupa ion-coun y-yea cell
S d o AI exposu e S anda d de ia ion o he AI exposu e a ISCO 4-digi
le el o occupa ion o
S d o obo exposu e S anda d de ia ion o he obo exposu e a ISCO 4-digi
le el o occupa ion o
No es: This Table shows he lis o a iables used. The da a sou ce o cons uc ing he occupa ion-coun y-yea panel is
he EU-SES, while he in o ma ion on he echnology exposu e has been elabo a ed based on he li e a u e ci ed abo e.
12

Table 3.3: Desc ip i e s a is ics o dependen and explana o y a iables
mean sd p50 min max obs
Dependen a iables
Log o he wage gap p90/p10 0.98 0.35 0.93 0.26 3.50 2883
Log o he wage gap p90/p50 0.54 0.20 0.51 0.14 2.88 2883
Log o he wage gap p50/p10 0.44 0.20 0.41 0.01 1.31 2883
Independen a iables
AI exposu e Webb 0.41 0.15 0.40 0.06 0.91 2883
Robo exposu e Webb 0.56 0.45 0.36 0.11 1.61 2883
AI exposu e Fel en e al 0.67 0.03 0.66 0.61 0.72 2883
AI exposu e OECD 0.65 0.19 0.68 0.00 0.96 2067
AI exposu e P y ko a e al 0.00 0.02 0.00 0.00 0.12 2883
AI exposu e Engbe g e al 0.14 0.28 0.00 0.00 1.06 2883
Robo exposu e Mon obbio e al 0.31 0.10 0.28 0.13 0.56 2883
Robo exposu e P y ko a e al 0.24 0.93 0.01 0.00 7.40 2883
Con ol a iables
HHI index 0.33 0.21 0.24 0.10 1.00 2883
Sha e o manu ac u ing 0.23 0.25 0.15 0.00 1.00 2883
Blau index (gende ) 0.63 0.13 0.58 0.50 0.99 2883
Sha e o emale 0.45 0.25 0.45 0.00 0.96 2883
Sha e o unioniza ion 0.70 0.32 0.82 0.00 1.00 2883
N o o 4-digi occupa ions 11.53 6.76 11.00 2.00 26.00 2883
HHI o high-skilled occupa ions 0.59 0.15 0.55 0.33 1.00 2883
S d o AI exposu e 0.17 0.07 0.19 0.03 0.36 2883
S d o obo exposu e 0.29 0.23 0.22 0.04 0.94 2883
S d o ou ine cogni i e in ensi y 0.24 0.10 0.22 0.03 0.62 2883
S d o ou ine manual in ensi y 0.37 0.14 0.40 0.06 0.73 2883
No es: This Table shows he desc ip i e s a is ics o he dependen and independen a iables o he whole sample. Own elabo a ion
based on he i e wa es o he EU-SES.
13
4 Desc ip i e o e iew
The 2000s saw a signi ican inc ease in obo adop ion, leading o ex ensi e li e a u e explo -
ing i s e ec s on labo ma ke s a a ious le els o agg ega ion (G ae z and Michaels,2018;
Aghion e al.,2019;Acemoglu and Res epo,2020).14 Figu e 4.1 illus a es he e olu ion o
obo s ock pe 1,000 wo ke s-— e e ed o as obo densi y—-wi hin he wel e Eu opean
coun ies wi h he highes adop ion a es. Ge many s ands ou as a leade in obo pene-
a ion o e ime. Howe e , i is also clea ha some Eas e n Eu opean economies made
signi ican ad ancemen s in adop ion ollowing he inancial c isis. The Czech Republic and
Slo akia a e pa icula ly no ewo hy, wi h hei obo densi y expe iencing a e age annual
g ow h a es o 18.53% and 13.5%, espec i ely, om 1995 o 2017.15 As a esul , by 2017,
hese wo coun ies anked second and hi d in obo densi y, ollowing Ge many.16 Mean-
while, I aly, Sweden, Finland, Denma k, Belgium, Aus ia, and he Ne he lands ha e also
shown a no able inc ease in obo adop ion, albei a a slowe pace.
Figu e 4.1: Robo densi y 1995-2017.
No es: Figu e 4.1 shows he e olu ion o obo s s ock pe 1000 wo ke s in wel e Eu opean coun ies. Sou ce: Own elabo a ion
based on he In e na ional Fede a ion o Robo ics (IFR) and EU-KLEMS o employmen .
In con as o obo s, AI is mo e ecen and no widely di used ye (Co ado e al.,2021;
Vannuccini and P y ko a,2024). Consequen ly, sys ema ic da abases enabling ime se ies
14Fo a comp ehensi e e iew, see (Ju ka e al.,2022).
15Re e o Table A.1 o mo e de ails on he a e age g ow h a e o obo densi y by coun y.
16This end may be linked o he ise o Global Value Chains (GVCs), speci ically due o Ge many shi ing
p oduc ion o Eas e n Eu opean coun ies.
14
analysis a e cu en ly una ailable. Howe e , he e a e empi ical e o s o assess i s pe a-
si eness. One no able sou ce is he AI Obse a o y by he OECD, which o e s indica o s
o AI in in es men ac oss coun ies. Fo ins ance, Figu e A.1 in Appendix Billus a es
he inc easing de elopmen o AI o e ime, e lec ing a sha p ise in en u e capi al in-
es men s in his echnology ac oss se e al Eu opean coun ies. A clea upwa d end has
been e iden since 2016, in line wi h he indings o Co ado e al. (2021). Addi ionally,
Cal ino and Fon anelli (2023) ha e p o ided insigh ul analysis on AI adop ion a he i m
le el in 11 coun ies, using no el da a collec ed wi hin he amewo k o he OECD AI Di -
usion P ojec . No ewo hy indings indica e ha la ge and younge i ms ha e a highe
adop ion a e o AI, which is s ongly co ela ed wi h exis ing complemen a y capabili ies.
The au ho s emphasize ha o e 20% o la ge i ms in Eu opean coun ies use AI.17 These
indings align wi h ecen Eu os a da a on AI usage by en e p ises, indica ing ha 8% o
EU en e p ises employed a leas one AI echnology in 2021, wi h he igu e ising o 28%
o la ge en e p ises.18
In he emainde o his sec ion, we will zoom in o he ela ionship be ween obo s and AI
and wo k o ce composi ion. Al hough AI is no ye ully widesp ead, he abo e-men ioned
e idence sugges s ha i s adop ion is accele a ing quickly. The e o e, he scope o he p esen
analysis is limi ed o examining he po en ial channels h ough which a wide adop ion o
his echnology may a ec inequali y.
4.1 Inequali y and wage a iabili y
In his sec ion, we will p o ide empi ical e idence o he ele ance o wi hin-occupa ion wage
inequali y and i s ela ionship wi h exposu e o AI and obo s. Figu e 4.2 shows he Theil
index decomposi ion in o wi hin and be ween occupa ion inequali y o 19 Eu opean coun-
ies.19 This indica es ha wi hin-occupa ion inequali y accoun s o a signi ican po ion
o o e all inequali y in mos o he coun ies s udied. In 15 ou o 19 coun ies, he sha e
o wi hin-occupa ion inequali y exceeds 50%. The igu e a ies, anging om 38.35% o
79.60%, wi h I aly ha ing he lowes p opo ion and F ance ha ing he highes .
17The su eyed Eu opean coun ies include Belgium, Denma k, F ance, Ge many, I aly, Po ugal, and
Swi ze land, wi h su ey yea s spanning o e 2017–2020.
18Fo u he de ails e e o EUROSTAT - Use o AI in en e p ises.
19The Theil index se es as a me ic o quan i y he di e gence o a gi en income dis ibu ion om a
hypo he ical scena io whe e each indi idual ea ns an equal amoun . I can be decomposed in o inequali y
wi hin and be ween g oups. Re e o appendix A.3 o de ails on he compu a ion.
15
Figu e 4.2: Theil index decomposi ion: be ween and wi hin occupa ion inequali y.
No es: Figu e 4.2 shows he decomposi ion o he Theil index in o wi hin and be ween occupa ion inequali y a ISCO-08 2-digi
le el o he eal hou ly wage. I is he a e age ac oss he i e wa es o he EU-SES (2002, 2006, 2010, 2014 and 2018). Sou ce:
Own elabo a ion based on EU-SES.
The e idence p esen ed abo e is based on a ela i ely b oad 2-digi le el o occupa ional
classi ica ion. To p o ide a mo e de ailed analysis, Figu e A.2 in he appendix illus a es he
Theil decomposi ion o occupa ions a 3-digi le el o he ISCO-08 classi ica ion o coun ies
wi h a ailable da a. While he wi hin-occupa ion componen shows a sligh educ ion, i
s ill accoun s o hal o mo e o he o e all inequali y in mos coun ies. No able excep ions
include Cyp us and Luxembou g, whe e he wi hin-occupa ion componen explains only
32.6% and 41.2% o he inequali y, espec i ely.
The impo ance o wi hin-occupa ion wage dispa i y is u he con i med in Tables A.2
and A.3, whe e we conduc an ANOVA analysis o examine he a iance o he p90/p10 eal
hou ly wage gap as an indica o o inequali y. I sugges s ha a ound 67% o he a iance
o he p90/p10 wage gap is explained by wi hin-occupa ion a ia ion. I also de i es om
Table A.2 ha inequali y is highe in high-paying occupa ions (i.e., manage s, science and
enginee ing p o essionals) and shows a la ge dispe sion o he indica o .
Gi en ha occupa ions in ol e a combina ion o asks, which may ela e o speci ic ech-
nologies designed o ce ain ac i i ies, exposu e o hese echnologies may po en ially be
linked o wage inequali y. We aim o explo e his u he in he emainde o his sec ion.
Recalling om Figu e 2.2, occupa ions wi h highe exposu e o obo s a e p edominan ly
si ua ed a he lowe end o he wage dis ibu ion, whe eas hose mo e exposed o AI a e
concen a ed a he uppe end. Consequen ly, we can expec di e en pa e ns in wage
a iabili y, as highe -paid occupa ions exhibi g ea e dispe sion. Hence, Figu e 4.3 illus-
a es he co ela ion be ween he loga i hm o he p90/p10 wage gap and he occupa ional
exposu e sco es o obo s and AI. I de i es om his ha g ea e exposu e o AI posi i ely
16
co ela es wi h inequali y, while he opposi e occu s o obo s.20
Figu e 4.3: Co ela ion be ween wage inequali y wi hin he occupa ion and exposu e o AI
and obo s
(a) AI
(b) Robo
No es: Figu e 4.3 shows he co ela ion be ween he loga i hm o he eal hou ly wage gap o he p90/p10 and he AI sco e
(panel 4.3a) and he obo sco e (panel 4.3b). The obo and AI sco es a e a ISCO-08 2-digi le el. They a e buil using he
indexes p o ided by Webb (2020) a he SOC-2010 and con e ed in o ISCO-08. We mapped ISCO-88 o ISCO-08 o ha monize
occupa ions ac oss he di e en wa es. The de ails o he ha moniza ion can be consul ed in appendix A. Fo compa abili y
easons, we es ic he sample o he eu o coun ies. These a e Belgium, Cyp us, Es onia, Finland, F ance, I aly, Li huania,
Luxembu g, Ne he lands, Poland, Po ugal, Slo akia, and Spain. Sou ce: Own elabo a ion based on EU-SES.
The obse ed co ela ions be ween exposu e o AI o obo s and wage dispe sion can be
20This is also obse ed when analyzing he co ela ion wi h he s anda d de ia ion o he mean o he eal
hou ly wage in he occupa ion in Figu e A.6 in Appendix A.
17

be e unde s ood by conside ing he unde lying ask cha ac e is ics ha hese echnologies
p oxy. Speci ically, occupa ions wi h highe exposu e o obo s ypically in ol e manual and
s anda dized asks. These asks end o be less complex and allow o limi ed a ia ion in how
hey a e pe o med. As a esul , di e ences in indi idual pe o mance a e ela i ely mino ,
leading o mo e uni o m p oduc i i y and, consequen ly, mo e comp essed wage dis ibu ions
wi hin hese oles.
By con as , occupa ions wi h g ea e exposu e o AI a e gene ally cha ac e ized by cog-
ni i e and complex asks ha a e mo e he e ogeneous in na u e. These asks ely mo e
hea ily on indi idual judgmen , p oblem-sol ing, and o he capabili ies ha a y signi i-
can ly ac oss wo ke s. This a ia ion con ibu es o b oade di e ences in pe o mance and
ou pu , which in u n esul s in wide wage dispe sion wi hin hese occupa ions.
Thus, in line wi h ecen li e a u e, we ea AI and obo exposu e no as di ec causes
o wage inequali y, bu as indica o s o he unde lying ask s uc u e wi hin occupa ions–
s uc u es ha hemsel es a e closely associa ed wi h di e ences in wage dispe sion.21 In
he emainde o he pape , we u he examine hese associa ions and how hey e lec he
di e en ia ed ole o echnology ac oss occupa ional ask p o iles.
5 Empi ical s a egy and esul s
In ligh o he desc ip i es ou lined in 4.1, in his sec ion, we will examine he associa ion be-
ween exposu e o occupa ions o obo s and AI and wi hin-wage inequali y by implemen ing
a ixed-e ec s model. To do his, we apply he ollowing speci ica ion:
yoc =βKT
o+γo +¯
X+ϵoc (1)
Whe e yoc is he main dependen a iable, which is he loga i hm o he eal hou ly
wage gap in occupa ion oin coun y ca yea . We conside h ee ou comes o wi hin-
occupa ion inequali y yoc : he wage gap be ween he p90/p10, he p50/p10 and he p90/p50.
KT
o ep esen s he exposu e sco e o each occupa ion o each echnology T( obo and AI).
βis he main coefficien and ep esen s he change in wage inequali y o a uni change in he
obo o AI occupa ional sco e. Based on he e idence p esen ed in he p e ious sec ion, we
expec β o ha e a posi i e (nega i e) sign, meaning ha inc easing exposu e o AI ( obo s)
is associa ed wi h g ea e (lowe ) inequali y.
The a iable γo includes a se o con ol a iables ha a e occupa ion-speci ic.22 Fi s ,
we con ol o he deg ee o collec i e ba gaining co e age. By egula ing he wage ela ion-
ship, ade unions end o be associa ed wi h a educ ion in he dispe sion o wages wi hin
occupa ions (Ca d e al.,2004;Biewen and Seckle ,2019). Addi ionally, he gende compo-
si ion o he labo ma ke can in luence wage dis ibu ion. Ne e heless, empi ical e idence
has yielded mixed esul s ega ding he di ec ion o his e ec (Kollmeye ,2013). On one
hand, he e is li e a u e sugges ing ha an inc ease in he p opo ion o women in he la-
bo ma ke can educe wage inequali y. One o he main easons is he g ea e p opensi y
21This concep ual dis inc ion is c ucial: ou empi ical s a egy does no aim o demons a e a causal e ec
o AI o obo s on wage inequali y, bu a he o explo e how hei exposu e is associa ed wi h di e ing
pa e ns o wi hin-occupa ion inequali y, media ed by he na u e o ask con en .
22Re e o appendix B.1 o he de ails on he cons uc ion o con ol a iables.
18
o low-educa ed women en e ing he labo o ce (Esping-Ande sen,2007;Ha kness,2010).
Con e sely, some schola s a gue ha highe emale pa icipa ion in he labo ma ke can ex-
ace ba e wage inequali y, pa icula ly i emale employmen subs i u es o high-skilled male
employmen (Acemoglu e al.,2004) o due o di e ing labo cha ac e is ics, such as lowe
ade union affilia ion (Ca d,2001). The e o e, despi e he unclea di ec ion o he e ec ,
we con ol o he sha e o emale employmen in each occupa ion o ne ou i s impac on
wage inequali y.23
Fu he , he e is e idence ha wage dispe sion is mo e p onounced in ce ain indus ies.
In pa icula , he e is a s and o li e a u e ha has poin ed o a posi i e implica ion o
se ici ica ion on wage inequali y (Blum,2008;Boddin and K oege ,2022), mainly ela ed
o hei ela i ely g ea e sha e o high-skilled wo ke s. Thus, o accoun o he indus ial
composi ion, we con ol o he sha e o manu ac u ing employmen .
The ac ha he occupa ional analysis is conduc ed a he ISCO-08 2-digi le el p esen s
ce ain challenges. As no ed in Sec ion 3, his choice s ems om he lack o consis en in o -
ma ion on occupa ions a he 3-digi le el o mos coun ies. Two impo an conside a ions
should be highligh ed. Fi s , he numbe o occupa ions wi hin each majo subg oup a he
2-digi le el a ies signi ican ly, especially o p o essional ca ego ies. Na u ally, his implies
ha a iabili y in he ou come a iable can pa ly be a ibu ed o g ea e dispe sion wi hin
occupa ional g oups. To add ess his, we con ol o he numbe o 4-digi occupa ions wi hin
each majo 2-digi ca ego y.
A ela ed issue is he subs an ial a ia ion in echnology exposu e wi hin each majo ca -
ego y (Fos e -McG ego e al.,2021).24 Consequen ly, we con ol o he s anda d de ia ion
in exposu e o obo s and AI a he 4-digi le el o each occupa ion o. In he same ein,
we also include he s anda d de ia ion o ou ine in ensi y o each occupa ion o o u he
con ol o occupa ional cha ac e is ics.
Finally, ¯
X ep esen s he se o ixed e ec s. We con ol o coun y and yea - ixed
e ec s. The i s one accoun s o aspec s ela ed o ins i u ional dimensions which a e ime-
in a ian ; while he second cap u es changes ela ed o business cycles–i.e. he inancial
c isis. We use i e wa es o he EU-SES: 2002, 2006, 2010, 2014 and 2018 and 19 coun ies
in ou analysis.25
Resul s Table 5.1 shows he associa ion o exposu e o obo s and AI and inequali y
wi hin occupa ions o he loga i hm o he p90/10, p90/50 and p50/10 wage gap. When
examining AI exposu e, he posi i e coefficien indica es ha occupa ions mo e exposed o
his echnology end o ha e, on a e age, highe wage inequali y. Fo he p90/p10 wage
gap, a 0.1 inc ease in he AI exposu e sco e is associa ed wi h an inc ease in inequali y o
23As an addi ional measu e, we also con ol o he Blau index o accoun o labo ma ke di e si y ins ead
o he emale sha e. This indica o cap u es po en ial non-linea i ies ha may exis in he ela ionship
be ween inequali y and emale pa icipa ion.
24As occupa ions a e agg ega ed, a ia ion in echnology exposu e is educed. This end is e iden in
Tables C.5 o C.7, whe e we show co ela ions be ween he al e na i e echnology indica o s used in his
pape . These ables e eal ha as occupa ions a e agg ega ed, he co ela ion be ween di e en measu es
inc eases, indica ing no able he e ogenei y a mo e g anula le els.
25The coun ies included a e Belgium, Bulga ia, Cyp us, Czech Republic, Es onia, Finland, F ance, Hun-
ga y, I aly, Li huania, Luxembu g, Ne he lands, No way, Poland, Po ugal, Romania, Slo akia, Spain, and
Sweden. Table B.2 p o ides u he de ails on he composi ion o he sample.
19
app oxima ely 2.8%. This associa ion is p ima ily d i en by he uppe hal o he wage
dis ibu ion. Speci ically, a 0.1 inc ease in exposu e co esponds o a ise in wage dispe sion
o a ound 1.7% o he p90/p50 gap and 1.1% o he p50/p10 gap.
Unlike AI, occupa ions ha a e mo e exposed o obo s ha e a lowe wage inequali y.
Fo he p90/p10 wage gap, a 0.1 inc ease in he obo sco e is ela ed o a 3.7% dec ease
in wi hin-occupa ion inequali y. This e ec is pe asi e ac oss he wo po ions o he wage
dis ibu ion as i is highly signi ican and nega i e. Ne e heless, i is sligh ly highe o he
lowe pa o he wage dis ibu ion, whe e a 0.1 change in he obo sco e exposu e is ela ed
o a 1.9% dec ease in wage inequali y, compa ed o a 1.8% o he uppe hal .
These indings indica e ha occupa ions mos exposed o AI ( obo s) a e hose whe e
he e is mo e (less) wage inequali y. The con as ing signs obse ed in he coefficien s o
obo s and AI a e likely linked o he di e en ypes o asks cha ac e izing occupa ions
associa ed wi h each echnology. Speci ically, AI exposu e ends o be highe in occupa-
ions in ol ing non- ou ine cogni i e asks, while obo exposu e is mo e p e alen in hose
in ol ing non- ou ine manual asks.
Impo an ly, we in e p e exposu e o AI and obo s no as di ec causes o wage in-
equali y, bu a he as p oxies o he unde lying ask s uc u e wi hin occupa ions. Tha
is, hese echnologies end o be deployed in occupa ional se ings wi h dis inc ask cha ac-
e is ics, which hemsel es a e associa ed wi h di e en deg ees o he e ogenei y in wo ke
pe o mance and compensa ion.
Following Jung and Me cenie (2014), his pa e n can be explained by di e ences in he
dispe sion o (unobse ed) wo ke abili ies ac oss occupa ions. Wo ke s in non- ou ine cog-
ni i e jobs–o en hose mo e exposed o AI– ypically possess highe skill le els and exhibi
a b oade dis ibu ion o capabili ies. Since hese asks ely hea ily on indi idual judgmen
and p oblem-sol ing (e.g., g aphical design), pe o mance ou comes a e mo e a iable, e-
sul ing in g ea e wage dispe sion. In con as , occupa ions mo e exposed o obo s in ol e
s anda dized manual asks (e.g., mo ing objec s), whe e wo ke au onomy and disc e ion a e
limi ed. In hese con ex s, indi idual abili y has a smalle impac on ask execu ion, leading
o mo e uni o m p oduc i i y and na owe wage dis ibu ions.26
26I is wo h no ing ha an addi ional channel ha can in luence wage inequali y is di e ences wi hin and
be ween i ms (Ake man e al.,2013;Ca d e al.,2018;Ca d,2022). Fo ins ance, Song e al. (2018) ind
ha o he U.S. wo- hi ds o wage inequali y is explained by wage dispe sion be ween i ms, whe eas he
es is d i en by wage a ia ion wi hin i ms. In Eu ope, i is ound ha ecen inc ease in inequali y a
he agg ega e le el is posi i ely associa ed wi h g ow h in la ge i ms (Muelle e al.,2017). Fu he , when
also inco po a ing he echnical change dimension, he e a e also la ge dispa i ies obse ed be ween i ms
ha inno a e and hose ha do no inno a e (Ci illo e al.,2017). Au oma ion adop ion can u he spu
wage di e en ials be ween i ms ha in es in hese echnologies and he ones ha do no by expanding he
demand o wo ke s wi h speci ic skills ha ha e highe wages (Bessen e al.,2022).
20
Table 5.1: Dependen a iable: Log o eal hou ly wage gap.
(1) (2) (3)
Wage gap p90/p10 Wage gap p90/p50 Wage gap p50/p10
Robo exposu e Webb -0.363∗∗∗ -0.178∗∗∗ -0.185∗∗∗
(0.023) (0.012) (0.015)
AI exposu e Webb 0.273∗∗∗ 0.166∗∗∗ 0.107∗∗∗
(0.053) (0.036) (0.028)
Cons an 1.075∗∗∗ 0.635∗∗∗ 0.439∗∗∗
(0.069) (0.042) (0.038)
Obse a ions 2883 2883 2883
R20.548 0.485 0.453
Coun y FE Yes Yes Yes
Yea FE Yes Yes Yes
Manu ac u ing sha e Yes Yes Yes
Sex Yes Yes Yes
Union Yes Yes Yes
N o Ocupp Yes Yes Yes
HHI High skill Yes Yes Yes
No es:∗∗∗ p < 0.01;∗∗ p < 0.05;∗p < 0.1. Robus s anda d e o s be ween pa en heses. Table 5.1 displays he esul s o
eg essing he log o he eal hou ly wage gap on he AI and obo exposu e sco e a he occupa ion-coun y-yea le el. The
AI and obo occupa ional sco es a e a ISCO-08 2-digi le el. They a e buil using he index p o ided by Webb (2020)
a he SOC-2010 and con e ed in o ISCO-08. Fo yea s be o e 2010, we mapped ISCO-88 o ISCO-08. The de ails o he
ha moniza ion can be consul ed in appendix A. Con ol a iables o each occupa ion-coun y-yea cell include: he sha e
o employmen in manu ac u ing, he sha e o emale employmen , he sha e o unionized wo ke s, and he HHI index o
high-skill employmen . Addi ional con ol a iables a he occupa ional le el include: he numbe o 4-digi occupa ions
wi hin each ISCO-08 2-digi occupa ion, he s anda d de ia ion o obo and AI exposu e a he 4-digi le el wi hin ISCO-08
2-digi occupa ions, he s anda d de ia ion o ou ine ask con en a he 4-digi le el wi hin ISCO-08 2-digi occupa ions.
Coun y and yea ixed e ec s a e also included. Fo u he de ails on he cons uc ion o he con ol a iables e e o
appendix B.1. Own elabo a ion based on he i e wa es o he EU-SES.
21
F om ou analysis, we can en ision wo scena ios. Fi s , i AI’s de elopmen leads o
g ea e s anda diza ion o asks ac oss a ying skill le els, i could, in ac , educe wage
inequali y—-incipien empi ical e idence sugges s his as a possible di ec ion (e.g., Engbe g
e al. (2024) and Geo gie (2024)). Howe e , because AI can bo h subs i u e o and com-
plemen high-skilled wo ke s, i also has he po en ial o wo sen inequali y by inc easing he
demand o specialized skills. Acco ding o he model p oposed by Acemoglu and Res epo
(2022), echnology no only au oma es exis ing asks bu also c ea es new ones ha comple-
men i . The e o e, i is plausible ha AI will gene a e new asks ha a e associa ed wi h
high-skilled wo ke s.
In esponse o he inequali y isks posed by AI, policymake s may conside adap i e and
imely measu es. T aining p og ams o equip wo ke s wi h skills ha align wi h AI-d i en
changes will be c ucial o a esilien wo k o ce. Addi ionally, edis ibu i e policies will play
a c i ical ole in na owing he wage gap be ween high- and low-paying occupa ions. These
app oaches will be key o ensu ing ha AI’s economic bene i s a e b oadly sha ed and ha
i s adop ion os e s inclusi e g ow h ac oss he wo k o ce.
Ou pape is subjec o ce ain limi a ions. Pe haps he mos signi ican limi a ion is
ha we canno es ablish causali y due o he cha ac e is ics o ou da a. Due o limi a ions
in he a ailable da a, we p ima ily ocus on occupa ions a he 2-digi le el, e en hough a
mo e de ailed analysis would ha e been bene icial. While we a emp ed o p o ide es ima es
using he ISCO-08 3-digi le el o some coun ies, his app oach was easible o only a
limi ed numbe o hem. Las ly, echnology con inually e ol es, engaging in new ac i i ies
o e ime. Consequen ly, exposu e o echnology may no emain cons an , as assumed by
he index employed in his pape . We pa ially add essed his issue by explo ing a dynamic
indica o de eloped by P y ko a e al. (2024) and Engbe g e al. (2024). Howe e , hese
indica o s a e only a ailable om 2012 onwa d, while ou da ase begins in 2002. The e o e,
i is impo an o in e p e his ca e ully.
Finally, his pape pa es he way o u he esea ch. As mo e da a on AI becomes
a ailable p og essi ely, i would be ele an o s udy he implica ions o AI adop ion on
inequali y. Gi en ha his echnology has he po en ial o bo h subs i u e and complemen
high-skilled wo ke s, i would be in e es ing o in es iga e u he whe he i s adop ion in lu-
ences o e all employmen composi ion o ins ead exace ba es wage inequali y, pa icula ly
among wo ke s mo e exposed o i .
28

Re e ences
Acemoglu, D. and Au o , D. (2011). Skills, Tasks and Technologies: Implica ions o Em-
ploymen and Ea nings. In Ca d, D. and Ashen el e , O., edi o s, Handbook o Labo
Economics, olume 4, chap e 12, pages 1043–1171. Else ie .
Acemoglu, D., Au o , D. H., Hazell, J., and Res epo, P. (2022). A i icial In elligence and
Jobs: E idence om Online Vacancies. Jou nal o Labo Economics, 40(S1):S293–S340.
Acemoglu, D., Au o , D. H., and Lyle, D. (2004). Women, wa , and wages: The e ec o
emale labo supply on he wage s uc u e a midcen u y. Jou nal o poli ical Economy,
112(3):497–551.
Acemoglu, D. and Res epo, P. (2020). Robo s and jobs: E idence om US Labo Ma ke s.
Jou nal o Poli ical Economy, 128(6):2188–2244.
Acemoglu, D. and Res epo, P. (2022). Tasks, au oma ion, and he ise in US wage inequali y.
Econome ica, 90(5):1973–2016.
Aghion, P., An onin, C., and Bunel, S. (2019). A i icial In elligence, G ow h and Employ-
men : The Role o Policy. Economie e S a is ique / Economics and S a is ics, 510-511-
51:149–164.
Ake man, A., Helpman, E., I skhoki, O., Muendle , M.-A., and Redding, S. (2013). Sou ces
o Wage Inequali y. Ame ican Economic Re iew, 103(3):214–219.
Au o , D. H. (2013). The “ ask app oach” o labo ma ke s: An o e iew. Jou nal o
Labou Ma ke Resea ch, 46(3):185–199.
Au o , D. H. (2022). The Labo Ma ke Impac s o Technological Change: F om Unb idled
En husiasm o Quali ied Op imism o Vas Unce ain y. NBER Wo king Pape Se ies
30074, Na ional Bu eau o Economic Resea ch (NBER), Camb idge, MA.
Au o , D. H., Le y, F., and Mu nane, R. J. (2003). The Skill Con en o Recen Technological
Change: An Empi ical Explo a ion. The Qua e ly Jou nal o Economics, 118(4):1279–
1333.
Ba h, E., Røed, M., Schøne, P., and Umblijs, J. (2020). How Robo s Change Wi hin-Fi m
Wage Inequali y. IZA DP No. 13605, In i u e o Labo Economics (IZA).
Bessen, J., Denk, E., and Meng, C. (2022). The emainde e ec : How au oma ion comple-
men s labo quali y. Bos on Uni e si y School o Law Resea ch Pape Se ies, (22-3).
Bessen, J. E. (2016). How compu e au oma ion a ec s occupa ions: Technology, jobs, and
skills. Law & Economics Wo king Pape No. 15-49, Bos on Uni e si y School o Law.
Biewen, M. and Seckle , M. (2019). Unions, in e na ionaliza ion, asks, i ms, and wo ke
cha ac e is ics: A de ailed decomposi ion analysis o ising wage inequali y in Ge many.
The Jou nal o Economic Inequali y, 17:461–498.
29
Blum, B. S. (2008). T ade, echnology, and he ise o he se ice sec o : The e ec s on US
wage inequali y. Jou nal o In e na ional Economics, 74(2):441–458.
Boddin, D. and K oege , T. (2022). Se i iza ion, Inequali y, and Wages. Labou Economics,
77:102011.
Bonney, K., B eaux, C., Buffing on, C., Dinle soz, E., Fos e , L. S., Goldschlag, N., Hal i-
wange , J. C., K o , Z., and Sa age, K. (2024). T acking i m use o AI in eal ime: A
snapsho om he Business T ends and Ou look Su ey. NBER Wo king Pape 32319,
Na ional Bu eau o Economic Resea ch (NBER).
B all, F. and Schmid, R. (2023). Au oma ion, obo s and wage inequali y in ge many: A
decomposi ion analysis. LABOUR, 37(1):33–95.
B ynjol sson, E., Jin, W., and McElhe an, K. (2021). The powe o p edic ion: p edic-
i e analy ics, wo kplace complemen s, and business pe o mance. Business Economics,
56:217–239.
B ynjol sson, E., Rock, D., and Sy e son, C. (2017). A i icial In elligence and he Mode n
P oduc i i y Pa adox: A Clash o Expec a ions and S a is ics. NBER Wo king Pape
24001, Na ional Bu eau o Economic Resea ch (NBER).
Cal ino, F. and Fon anelli, L. (2023). A po ai o AI adop e s ac oss coun ies: Fi m
cha ac e is ics, asse s’ complemen a i ies and p oduc i i y. OECD Science, Technology
and Indus y Wo king Pape s 2023/02, O ganiza ion o Economic Coope a ion and De-
elopmen (OECD).
Ca d, D. (2001). The E ec o Unions on Wage Inequali y in he U.S. Labo Ma ke .
Indus ial and Labo Rela ions Re iew, 54(2):296–315.
Ca d, D. (2022). Who se you wage? Ame ican Economic Re iew, 112(4):1075–1090.
Ca d, D., Ca doso, A. R., Heining, J., and Kline, P. (2018). Fi ms and labo ma ke
inequali y: E idence and some heo y. Jou nal o Labo Economics, 36(S1):S13–S70.
Ca d, D., Lemieux, T., and Riddell, W. C. (2004). Unions and wage inequali y. Jou nal o
Labo esea ch, 25:519–559.
Chiacchio, F., Pe opoulos, G., and Pichle , D. (2018). The impac o indus ial obo s on
EU employmen and wages: A local labou ma ke app oach. B uegel Wo king Pape
2018/02, B uegel.
Ci illo, V., Sos e o, M., and Tamagni, F. (2017). Inno a ion and wi hin- i m wage in-
equali ies: Empi ical e idence om majo Eu opean coun ies. Indus y and Inno a ion,
24(5):468–491.
Co ado, C., Haskel, J., and Jona-Lasinio, C. (2021). A i icial in elligence and p oduc i i y:
An in angible asse s app oach. Ox o d Re iew o Economic Policy, 37(3):435–458.
30
Co es, G. M. (2016). Whe e ha e he middle-wage wo ke s gone? A s udy o pola iza ion
using panel da a. Jou nal o Labo Economics, 34(1):63–105.
Cza ni zki, D., Fe nández, G. P., and Ramme , C. (2023). A i icial in elligence and i m-
le el p oduc i i y. Jou nal o Economic Beha io & O ganiza ion, 211:188–205.
Damioli, G., Van Roy, V., Vé esy, D., and Vi a elli, M. (2024). D i e s o employmen
dynamics o ai inno a o s. Technological Fo ecas ing and Social Change, 201:123249.
Dau h, W., Findeisen, S., Suedekum, J., and Woessne , N. (2021). The adjus men o labo
ma ke s o obo s. Jou nal o he Eu opean Economic Associa ion, 19(6):3104–3153.
Engbe g, E., Gö g, H., Lode alk, M., Ja ed, F., Längk is , M., Mon ei o, N. P., No dås,
H. K., Sch oede , S., and Tang, A. (2024). AI Unboxed and Jobs: A No el Measu e and
Fi m-Le el E idence om Th ee Coun ies. IZA Discussion Pape s 16717, Ins i u e o
Labo Economics (IZA).
E ns , E., Me ola, R., and Samaan, D. (2019). Economics o A i icial In elligence: Impli-
ca ions o he Fu u e o Wo k. IZA Jou nal o Labo Policy, 9(1).
Esping-Ande sen, G. (2007). Sociological Explana ions o Changing Income Dis ibu ions.
Ame ican Beha io al Scien is , 50(5):639–658.
Fana, M., Ci illo, V., Gua ascio, D., and Tubiana, M. (2020). A Compa a i e Na ional
Tasks Da abase. JRC Technical Repo 2020/13, Join Resea ch Cen e (JRC), Eu opean
Commission.
Fel en, E., Raj, M., and Seamans, R. (2021). Occupa ional, indus y, and geog aphic expo-
su e o a i icial in elligence: A no el da ase and i s po en ial uses. S a egic Managemen
Jou nal, 42(12):2195–2217.
Fel en, E. W., Raj, M., and Seamans, R. (2019). The occupa ional impac o a i icial
in elligence: Labo , skills, and pola iza ion. NYU S e n School o Business. A ailable a
h ps://pape s.ss n.com/sol3/pape s.c m?abs ac _id=3368605.
Fe nández-Macías, E. and A anz-Muñoz, J. M. (2020). Occupa ions and he ecen ends
in wage inequali y in Eu ope. Eu opean Jou nal o Indus ial Rela ions, 26(3):331–346.
Fe nandez Macias, E., U zi B anca i, M. C., W igh , S., and Pesole, A. (2023). The Pla -
o misa ion o Wo k. Technical Repo JRC133016, Publica ions Office o he Eu opean
Union, Luxembou g.
Filippi, E., Bannò, M., and T en o, S. (2023). Au oma ion echnologies and hei impac on
employmen : A e iew, syn hesis and u u e esea ch agenda. Technological Fo ecas ing
and Social Change, 191:122448.
Filippucci, F., Gal, P., Jona-Lasinio, C., Leand o, A., and Nicole i, G. (2024). The impac
o a i icial in elligence on p oduc i i y, dis ibu ion and g ow h: Key mechanisms, ini ial
e idence and policy challenges. OECD A i icial In elligence Pape s 15, O ganiza ion o
Economic Coope a ion and De elopmen (OECD).
31
Fo in, N., Lemieux, T., and Fi po, S. (2011). Occupa ional asks and changes in he wage
s uc u e. IZA Discussion Pape s 5542, In i u e o Labo Economics (IZA).
Fos e -McG ego , N., Nomale , Ö., and Ve spagen, B. (2021). Job Au oma ion Risk, Eco-
nomic S uc u e and T ade: a Eu opean Pe spec i e. Resea ch Policy, 50(7):104269.
F ank, M. R., Au o , D. H., Bessen, J. E., B ynjol sson, E., Ceb ian, M., Deming, D. J.,
Feldman, M., G oh, M., Lobo, J., Mo o, E., Wang, D., Youn, H., and Rahwan, I. (2019).
Towa d unde s anding he impac o a i icial in elligence on labo . P oceedings o he
Na ional Academy o Sciences o he Uni ed S a es o Ame ica, 116(14):6531–6539.
F ey, C. B. and Osbo ne, M. A. (2017). The u u e o employmen : How suscep ible a e jobs
o compu e isa ion? Technological Fo ecas ing and Social Change, 114:254–280.
Geo gie , A. (2024). A i icial in elligence and wage inequali y. OECD a i icial in elligence
pape s 13, O ganiza ion o Economic Coope a ion and De elopmen (OECD).
Geo gie , A. and Hyee, R. (2021). A i icial in elligence and employmen : New c oss-coun y
e idence. OECD Social, Employmen and Mig a ion Wo king Pape s 265, O ganiza ion
o Economic Coope a ion and De elopmen (OECD).
Goos, M. and Manning, A. (2007). Lousy and Lo ely Jobs: The Rising Pola iza ion o Wo k
in B i ain. The Re iew o Economics and S a is ics, 89(1):118–133.
Goos, M., Manning, A., and Salomons, A. (2009). Job pola iza ion in Eu ope. Ame ican
Economic Re iew, 99(2):58–63.
Goos, M., Manning, A., and Salomons, A. (2014). Explaining job pola iza ion: Rou ine-
biased echnological change and o sho ing. Ame ican Economic Re iew, 104(8):2509–26.
G ae z, G. and Michaels, G. (2018). Robo s a wo k. The Re iew o Economics and S a is ics,
100(5):753–768.
G ego y, T., Salomons, A., and Zie ahn, U. (2022). Racing wi h o Agains he Machine?
E idence on he Role o T ade in Eu ope. Jou nal o he Eu opean Economic Associa ion,
20(2):869–906.
Gua ascio, D., Reljic, J., and S öllinge , R. (2023). A i icial in elligence and employmen :
A look in o he c ys al ball. GLO Discussion Pape No. 1333, Global Labo O ganiza ion
(GLO).
Hall, P. A. and Soskice, D. (2001). An In oduc ion o Va ie es o Capi alism. In Va ie ies
o capi alism: The ins i u ional ounda ion o compa a i e ad an age, pages 1–70. Ox o d
Uni e si y P ess, New Yo k.
Ha dy, W., Keis e , R., and Lewandowski, P. (2018). Educa ional upg ading, s uc u al
change and he ask composi ion o jobs in eu ope. Economics o T ansi ion, 26(2):201–
231.
32
Ha kness, S. (2010). The con ibu ion o women’s employmen and ea nings o household
income inequali y: A c oss-coun y analysis. LIS Wo king Pape Se ies 531, Luxembou g
Income S udy (LIS).
Heß, P., Janssen, S., and Lebe , U. (2023). The e ec o au oma ion echnology on wo ke s’
aining pa icipa ion. Economics o Educa ion Re iew, 96:102438.
Jaccoud, F., Pe i , F., Cia li, T., and Sa ona, M. (2024). Au oma ion and Employmen
O e he Technology Li e Cycle: E idence om Eu opean Regions. CESi o Wo king
Pape 10987, CESi o.
Jung, J. and Me cenie , J. (2014). Rou iniza ion-biased Technical Change and Globaliza ion:
unde s anding Labo Ma ke Pola iza ion. Economic Inqui y, 52(4):1446–1465.
Ju ka , A., Klump, R., and Schneide , F. (2022). T acking he ise o obo s: The IFR
da abase. Jah büche ü Na ionalökonomie und S a is ik, 242(5-6):669–689.
Kal enbe g, M. and Fos e -McG ego , N. (2020). The impac o au oma ion on inequali y
ac oss Eu ope. Wo king Pape Se ies 31, UNU-MERIT.
Kim, C. and Sakamo o, A. (2008). The Rise o In a-Occupa ional Wage Inequali y in he
Uni ed S a es, 1983 o 2002. Ame ican Sociological Re iew, 73(1):129–157.
Kollmeye , C. (2013). Family S uc u e, Female Employmen , and Na ional Income In-
equali y: A C oss-Na ional S udy o 16 Wes e n Coun ies. Eu opean sociological e iew,
29(4):816–827.
Ma colin, L., Mi oudo , S., and Squiccia ini, M. (2019). To be ( ou ine) o no o be ( ou ine),
ha is he ques ion: A c oss-coun y ask-based answe . Indus ial and Co po a e Change,
28(3):477–501.
Mishel, L., Shie holz, H., and Schmi , J. (2013). Don’ Blame he Robo s: Assessing he
Job Pola iza ion Explana ion o G owing Wage Inequali y. EPI–CEPR Wo king Pape 19,
Economic Policy Ins i u e (EPI).
Mon obbio, F., S accioli, J., Vi gilli o, M. E., and Vi a elli, M. (2022). Robo s and he o igin
o hei labou -sa ing impac . Technological Fo ecas ing and Social Change, 174:121122.
Mon obbio, F., S accioli, J., Vi gilli o, M. E., and Vi a elli, M. (2024). Labou -sa ing
au oma ion: A di ec measu e o occupa ional exposu e. The Wo ld Economy, 47(1):332–
361.
Muelle , H. M., Ouime , P. P., and Simin zi, E. (2017). Wage inequali y and i m g ow h.
Ame ican Economic Re iew, 107(5):379–383.
Nedelkoska, L. and Quin ini, G. (2018). Au oma ion, skills use and aining. OECD Social,
Employmen and Mig a ion Wo king Pape s 202, O ganiza ion o Economic Coope a ion
and De elopmen (OECD).
33

OECD (2015). The Inno a ion Impe a i e: Con ibu ing o P oduc i i y, G ow h and Well-
Being. OECD Publishing.
OECD (2023). Recommenda ion o he council on a i icial in elligence. Accessed: 2024-10-
18.
Ozgul, P., F egin, M.-C., S ops, M., Janssen, S., and Le els, M. (2024). High-skilled Human
Wo ke s in Non-Rou ine Jobs a e Suscep ible o AI Au oma ion bu Wage Bene i s Di e
be ween Occupa ions. Technical epo .
Pi a, M. and Vi a elli, M. (2018). Inno a ion, jobs, skills and asks: A mul i ace ed ela-
ionship. GIORNALE DI DIRITTO DEL LAVORO E DI RELAZIONI INDUSTRIALI,
2018:599–619.
P y ko a, E., Pe i , F., Li, D., Cha u edi, S., and Cia li, T. (2024). The Employmen
Impac o Eme ging Digi al Technologies. Wo king Pape .
Song, J., P ice, D. J., Gu enen, F., Bloom, N., and on Wach e , T. (2018). Fi ming Up
Inequali y. The Qua e ly Jou nal o Economics, 134(1):1–50.
Squiccia ini, M. and S accioli, J. (2022). Labou -sa ing echnologies and employmen le els:
A e Robo s Really Making Wo ke s Redundan ? OECD Science, Technology and Indus y
Policy Pape s 124, O ganiza ion o Economic Coope a ion and De elopmen (OECD).
T aj enbe g, M. (2018). AI as he Nex GPT: A Poli ical-Economy Pe spec i e. NBER
Wo king Pape 24245, Na ional Bu eau o Economic Resea ch (NBER).
U zì B anca i, M.-C., Gonzalez, V. I., and Fe nández-Macías, E. (2023). The ise o digi al
moni o ing and algo i hmic managemen o wo k: Empi ical insigh s. Technical epo ,
Publica ions Office o he Eu opean Union. JRC Wo king Pape , JRC133407.
an de Velde, L. (2020). Wi hin Occupa ion Wage Dispe sion and he Task Con en o
Jobs. Ox o d Bulle in o Economics and S a is ics, 82(5):1161–1197.
Vannuccini, S. and P y ko a, E. (2024). A i icial In elligence’s new clo hes? A sys em
echnology pe spec i e. Jou nal o In o ma ion Technology, 39(2):317–338.
Webb, M. (2020). The impac o a i icial in elligence on he labo ma ke . Wo king Pape .
Weksle , G. and Las a, F. (2022). Guidowe/occupa ionc oss: ini ial elease. So wa e,
A ailable a h ps://doi.o g/10.5281/zenodo.7025097.
Zolas, N. J., K o , Z., B ynjol sson, E., McElhe an, K. S., Beede, D., Buffing on, C., Gold-
schlag, N., Fos e , L., and Dinle soz, E. (2021). Ad anced Technologies Adop ion and Use
by U.S. Fi ms: E idence om he Annual Business Su ey. NBER Wo king Pape Se ies
28290, Na ional Bu eau o Economic Resea ch (NBER).
34
Appendices
A Appendix A
A.1 Compa ibiliza ion be ween ISCO-08 and ISCO-88
Since he e is a b eak in he ISCO classi ica ion om 2010 onwa ds, we apply a c osswalk be-
ween ISCO-88 and ISCO-08. This is done by implemen ing he unc ion isco88_ o_isco08_ndigi
in he R package ‘occupa ionc oss’ (Weksle and Las a,2022). D awing upon he ISCO
co espondence ables p o ided by ILO, his unc ion akes he indi idual obse a ion a
ISCO-88 2-digi (o 3-digi ) le el and d aws an assignmen o an ISCO-08 code among all
he possible c osses38. We ha e used his unc ion o con e ISCO-88 in o ISCO-08 o he
2002 and 2006 wa es o he SES.
A.2 Desc ip i es o AI and obo pene a ion
Figu e A.1: Sum o en u e capi al in es men s in AI by coun y.
No es: Figu e A.1 shows he e olu ion o he sum o in es men s in AI in he op i een Eu opean coun ies (in U.S. dolla s).
Sou ce: OECD.AI (2022), isualisa ions powe ed by JSI using da a om P eqin, accessed on 28/11/2022, www.oecd.ai
38Fo mo e de ails on he package e e o R package ‘occupa ionc oss’.
35
Table A.1: Annual g ow h a e o obo densi y. 1995-2017.
Aus ia 5.98
(3.26)
Belgium 3.73
(5.64)
Czech Republic 18.53
(9.77)
Denma k 10.27
(4.92)
Finland 4.43
(6.13)
F ance 3.85
(4.32)
Ge many 5.71
(3.94)
Hunga y 18.09
(21.67)
I aly 4.23
(4.22)
Ne he lands 8.76
(5.03)
Slo akia 13.50
(21.10)
Spain 7.48
(7.07)
Sweden 4.20
(2.97)
To al 8.37
(10.86)
No es: Own elabo a ion based on In e na ional Fede a ion o Robo ics (IFR)
and EU-KLEMS. Robo densi y is he a io be ween obo s ock and employmen
in housand.
36
A.3 Theil decomposi ion a wo and h ee-digi le els
The Theil index is as ollows:
T=1
N
N
X
i=1
yi
¯yln yi
¯y
Whe e yiis he wage o indi idual i,Nis he o al numbe o employees and ¯yis he
a e age wage o he o al numbe o employees N.
I can be u he decomposed in o be ween and wi hin-occupa ion inequali y in he ol-
lowing way:
T=
O
X
o=1
No
N
¯yo
¯y
∗To
| {z }
Wi hin-Occupa ion
+
O
X
o=1
No
N
¯yo
¯yln ¯yo
¯y
| {z }
Be ween-Occupa ion
whe e Nois he numbe o employees in occupa ion o,Ois he o al numbe o occu-
pa ions, ¯yois he mean wage o occupa ion o, and Tois he Theil index o occupa ion o.
In his way, he wi hin-g oup componen is he weigh ed a e age o he Theil index ac oss
occupa ions. On he o he hand, he be ween-g oup componen ep esen s he inequali y
s emming om di e ences be ween occupa ions. This is calcula ed based on he dispa i ies
in a e age wage o e occupa ions in ela ion o he a e age o he en i e wo king popula ion.
In he con ex o his pape , y ep esen s he eal hou ly wage.
37
Figu e A.6: Co ela ion be ween s anda d de ia ion o he eal hou ly wage by occupa ion
and exposu e o AI and obo s
(a) AI
(b) Robo
No es: Figu e A.6 shows he co ela ion be ween he s anda d de ia ion o he eal hou ly wage gap and he AI sco e (panel
A.6a) and he obo sco e (panel A.6b) by occupa ion. The obo and AI sco es a e a ISCO-08 2-digi le el. They a e buil
using he indexes p o ided by Webb (2020) a he SOC-2010 and con e ed in o ISCO-08. We mapped ISCO-88 o ISCO-08
o ha monize occupa ions ac oss he di e en wa es. The de ails o he ha moniza ion can be consul ed in appendix A. Fo
compa abili y easons, we es ic he sample o eu o coun ies. These a e Belgium, Cyp us, Es onia, Finland, F ance, I aly,
Li huania, Luxembu g, Ne he lands, Poland, Po ugal, Slo akia, and Spain. Sou ce: Own elabo a ion based on EU-SES.
44

Figu e A.7: Dis ibu ion o he p90/p10 wage gap be o e and a e log ans o ma ion
No es: Figu e A.7 shows he dis ibu ion o he eal hou ly wage gap be ween he p90/p10 a he occupa ion-coun y-yea cell
be o e and a e i is log ans o med. Own elabo a ion based on EU-SES.
B Appendix B
B.1 Cons uc ion o con ol a iables o wi hin occupa ion inequali y
Sha e o emale by occupa ion We compu e he sha e o emale employmen on o al
employmen o each occupa ion-coun y-yea cell a ISCO-08 2-digi s.
Blau index We es ima e he Blau index o each occupa ion-coun y-yea cell by applying
he ollowing equa ion
Blau_indexoc = 1 −X
Lo c
Loc 2
(2)
Whe e Lo c is he le el o employmen in occupa ion o o ca ego y (i.e. male and
emale) in coun y ca ime and Loc is he o al employmen in occupa ion o, coun y ca
yea . The index a ies be ween 0 and 1, whe e alues close o 0 (1) indica e lowe (highe )
di e si y.
Sha e o unioniza ion To accoun o collec i e ag eemen co e age, we c ea ed a dummy
a iable aking alue equal o 1 i he wo ke has any so o ag eemen and 0 i she has
45
no ag eemen a all, as Table B.1 depic s. While his classi ica ion is somewha nuanced,
gi en ha he ype o collec i e ag eemen can signi ican ly impac wo ke s’ ba gaining
powe , i is impo an o no e ha no all coun ies p o ide da a o e e y ca ego y, posing
challenges o compa ison. Fo consis ency, we op ed o his app oach. Subsequen ly, we
calcula ed he p opo ion o wo ke s co e ed by any o m o collec i e ag eemen wi hin each
occupa ion-coun y-yea cell.
Table B.1: Classi ica ion o accoun o collec i e ag eemen
Dissemina ion in SES Union dummy
A Na ional le el o in e -
con ede al ag eemen
1
B Indus y ag eemen 1
C Ag eemen o indi id-
ual indus ies in indi-
idual egions
1
D En e p ise o single
employe ag eemen
1
E Ag eemen applying
only o wo ke s in he
local uni
1
F Any o he ype o
ag eemen
1
N No collec i e ag ee-
men exis s
0
No es: Own elabo a ion based on EU-SES
classi ica ion.
Sha e o employmen in manu ac u ing We calcula e he sha e o pe sons employed
in manu ac u ing on o al employmen o each occupa ion-coun y-yea cell a ISCO-08
2-digi s.
He indahl-Hi schman Index o Occupa ions (a indus y le el) is calcula ed by
agg ega ing he squa e oo o he sha e o employmen in he di e en indus ies in each
occupa ion. The indica o is as ollows:
HHIoc =
I
X
i=1 Loic
Loc 2
(3)
Whe e Loic is he le el o employmen in occupa ion oin indus y iin coun y ca ime
and Loc is he o e all employmen in occupa ion o, coun y ca yea . In his sense,
a highe alue o he index indica es ha employmen is concen a ed in ewe indus ies
and lowe le els imply ha employmen in ha occupa ion is mo e e enly sp ead ac oss
indus ies.
46
He indahl-Hi schman Index o Occupa ions ( o high-skill employmen ) is cal-
cula ed by agg ega ing he squa es o he sha es o high-skilled wo ke s—de ined as hose
wi h e ia y o highe educa ion—in each occupa ion. The indica o is exp essed as ollows:
HHIoc =XLohc
Loc 2
(4)
Whe e Lohc is he le el o high-skilled employmen in occupa ion oin in coun y ca
ime and Loc is he o e all employmen in occupa ion o, coun y ca yea . A highe
alue o he index o a speci ic occupa ion indica es a g ea e concen a ion o high-skilled
employees wi hin ha occupa ion.
Numbe o occupa ions wi hin majo subg oups We agg ega e he numbe o 4-digi s
occupa ions wi hin each ISCO08 2-digi occupa ions.
S anda d de ia ion o echnology exposu e We calcula e he s anda d de ia ion by
majo 2-digi ISCO08 g oups o he AI and obo exposu e sco es, along wi h he ou ine
in ensi y indexes—bo h manual and cogni i e—de ined o each occupa ion a he 4-digi
le el.
Task in ensi y measu es We e ie ed ask da a om he O*NET da abase, which is a
he SOC le el. In pa icula , we eso o he Abili ies, Skills, Wo k Ac i i ies, and Wo k
Con ex da a iles om O*NET. Following Ha dy e al. (2018), we selec ask- ela ed i ems
ha cap u e ou ine and non- ou ine ac i i ies, dis inguishing be ween cogni i e and manual
dimensions. Task in ensi y sco es o each SOC occupa ion is hen calcula ed by a e aging
he impo ance sco es wi hin each o he ou ca ego ies. Finally, occupa ions a e mapped
o ISCO-08 classi ica ions.
B.2 Desc ip i e s a is ics
C Appendix C. Robus ness checks
C.1 Reg essions wi h al e na i e con ol a iables
47
Table B.2: Composi ion o he sample (unbalanced)
Coun y 2002 2006 2010 2014 2018 To al
BE 24 24 20 21 17 106
BG 37 37 37 37 37 185
CY 31 33 30 33 33 160
CZ 37 37 37 37 37 185
EE 37 37 34 35 34 177
ES 37 37 37 35 37 183
FI 37 37 35 35 34 178
FR 37 37 36 36 36 182
HU 25 25 24 25 23 122
IT 36 0 0 36 36 108
LT 36 36 33 34 32 171
LU 34 34 0 32 0 100
NL 37 37 37 37 37 185
NO 0 37 37 37 0 111
PL 0 37 37 37 36 147
PT 36 37 37 37 37 184
RO 37 37 36 35 37 182
SE 0 0 0 0 37 37
SK 36 36 36 36 36 180
To al 554 595 543 615 576 2,883
No es: This Table shows he composi ion o he sample used in he analysis. Ou uni o analysis a e
coun y-occupa ion-yea , he e o e each cell o he able e e s o he numbe o occupa ions included.
Table B.3: Composi ion o he sample (balanced)
Coun y 2002 2006 2010 2014 2018 To al
BG 37 37 37 37 37 185
CY 29 29 29 29 29 145
EE 33 33 33 33 33 165
ES 35 35 35 35 35 175
FI 32 32 32 32 32 160
FR 36 36 36 36 36 180
HU 22 22 22 22 22 110
NL 37 37 37 37 37 185
PT 36 36 36 36 36 180
RO 35 35 35 35 35 175
SK 36 36 36 36 36 180
To al 368 368 368 368 368 1,840
No es: This Table shows he composi ion o he sample used in he analysis. Ou uni o analysis a e
coun y-occupa ion-yea , he e o e each cell o he able e e s o he numbe o occupa ions included.
48
Table C.1: Dependen a iable: Log o eal hou ly wage gap. Balanced panel.
(1) (2) (3)
Wage gap p90/p10 Wage gap p90/p50 Wage gap p50/p10
Robo exposu e Webb -0.341∗∗∗ -0.170∗∗∗ -0.171∗∗∗
(0.031) (0.016) (0.020)
AI exposu e Webb 0.305∗∗∗ 0.178∗∗∗ 0.128∗∗∗
(0.076) (0.053) (0.037)
Cons an 1.018∗∗∗ 0.590∗∗∗ 0.428∗∗∗
(0.082) (0.050) (0.047)
Obse a ions 1840 1840 1840
R20.516 0.500 0.410
Coun y FE Yes Yes Yes
Yea FE Yes Yes Yes
Manu ac u ing sha e Yes Yes Yes
Sex Yes Yes Yes
Union Yes Yes Yes
N o Ocupp Yes Yes Yes
HHI High skill Yes Yes Yes
No es:∗∗∗ p < 0.01;∗∗ p < 0.05;∗p < 0.1. Robus s anda d e o s be ween pa en heses. Table 5.1 displays he esul s o
eg essing he log o he eal hou ly wage gap on he AI and obo exposu e sco e a he occupa ion-coun y-yea le el. The
AI and obo occupa ional sco es a e a ISCO-08 2-digi le el. They a e buil using he index p o ided by Webb (2020) a
he SOC-2010 and con e ed in o ISCO-08. Fo yea s p io o 2010, we mapped ISCO-88 o ISCO-08. The de ails o he
ha moniza ion can be consul ed in appendix A. Con ol a iables o each occupa ion-coun y-yea cell include: he sha e
o employmen in manu ac u ing, he sha e o emale employmen , he sha e o unionized wo ke s, and he HHI index o
high-skill employmen . Addi ional con ol a iables a he occupa ional le el include: he numbe o 4-digi occupa ions
wi hin each ISCO-08 2-digi occupa ion, he s anda d de ia ion o obo and AI exposu e a he 4-digi le el wi hin ISCO-08
2-digi occupa ions, he s anda d de ia ion o ou ine ask con en a he 4-digi le el wi hin ISCO-08 2-digi occupa ions.
Coun y and yea ixed e ec s a e also included. Fo u he de ails on he cons uc ion o he con ol a iables e e o
appendix B.1. Own elabo a ion based on he i e wa es o he EU-SES.
49

Table C.2: Dependen a iable: Log o eal hou ly wage gap. Al e na i e con ol o gende
(1) (2) (3)
Wage gap p90/p10 Wage gap p90/p50 Wage gap p50/p10
Robo exposu e Webb -0.293∗∗∗ -0.137∗∗∗ -0.155∗∗∗
(0.024) (0.013) (0.016)
AI exposu e Webb 0.563∗∗∗ 0.284∗∗∗ 0.279∗∗∗
(0.045) (0.030) (0.025)
Cons an 0.907∗∗∗ 0.612∗∗∗ 0.295∗∗∗
(0.063) (0.038) (0.035)
Obse a ions 2883 2883 2883
R20.533 0.490 0.424
Coun y FE Yes Yes Yes
Yea FE Yes Yes Yes
Manu ac u ing sha e Yes Yes Yes
Blau Index (gende ) Yes Yes Yes
Union Yes Yes Yes
N o Ocupp Yes Yes Yes
HHI High skill Yes Yes Yes
No es:∗∗∗ p < 0.01;∗∗ p < 0.05;∗p < 0.1. Robus s anda d e o s be ween pa en heses. Table 5.1 displays he esul s o
eg essing he log o he eal hou ly wage gap on he AI and obo exposu e sco e a he occupa ion-coun y-yea le el. The
AI and obo occupa ional sco es a e a ISCO-08 2-digi le el. They a e buil using he index p o ided by Webb (2020)
a he SOC-2010 and con e ed in o ISCO-08. Fo yea s p io o 2010, we mapped ISCO-88 o ISCO-08. The de ails o
he ha moniza ion can be consul ed in appendix A. Con ol a iables o each occupa ion-coun y-yea cell include: he
sha e o employmen in manu ac u ing, he Blau index, he sha e o unionized wo ke s, and he HHI index o high-skill
employmen . Addi ional con ol a iables a he occupa ional le el include: he numbe o 4-digi occupa ions wi hin each
ISCO-08 2-digi occupa ion, he s anda d de ia ion o obo and AI exposu e a he 4-digi le el wi hin ISCO-08 2-digi
occupa ions, he s anda d de ia ion o ou ine ask con en a he 4-digi le el wi hin ISCO-08 2-digi occupa ions. Coun y
and yea ixed e ec s a e also included. Fo u he de ails on he cons uc ion o he con ol a iables e e o appendix
B.1. Own elabo a ion based on he i e wa es o he EU-SES.
50
Table C.3: Dependen a iable: Log o eal hou ly wage gap.
(1) (2) (3)
Wage gap p90/p10 Wage gap p90/p50 Wage gap p50/p10
Robo exposu e Webb -0.332∗∗∗ -0.163∗∗∗ -0.168∗∗∗
(0.023) (0.013) (0.015)
AI exposu e Webb 0.273∗∗∗ 0.166∗∗∗ 0.108∗∗∗
(0.053) (0.037) (0.028)
Cons an 1.013∗∗∗ 0.606∗∗∗ 0.407∗∗∗
(0.066) (0.042) (0.037)
Obse a ions 2883 2883 2883
R20.569 0.500 0.472
Coun y FE Yes Yes Yes
Yea FE Yes Yes Yes
HHI index Yes Yes Yes
Sex Yes Yes Yes
Union Yes Yes Yes
N o Ocupp Yes Yes Yes
HHI High skill Yes Yes Yes
No es:∗∗∗ p < 0.01;∗∗ p < 0.05;∗p < 0.1. Robus s anda d e o s be ween pa en heses. Table 5.1 displays he esul s o
eg essing he log o he eal hou ly wage gap on he AI and obo exposu e sco e a he occupa ion-coun y-yea le el. The
AI and obo occupa ional sco es a e a ISCO-08 2-digi le el. They a e buil using he index p o ided by Webb (2020)
a he SOC-2010 and con e ed in o ISCO-08. Fo yea s p io o 2010, we mapped ISCO-88 o ISCO-08. The de ails o
he ha moniza ion can be consul ed in appendix A. Con ol a iables o each occupa ion-coun y-yea cell include: he
HHI index o indus y concen a ion, he sha e o emale employmen , he sha e o unionized wo ke s, and he HHI index
o high-skill employmen . Addi ional con ol a iables a he occupa ional le el include: he numbe o 4-digi occupa ions
wi hin each ISCO-08 2-digi occupa ion, he s anda d de ia ion o obo and AI exposu e a he 4-digi le el wi hin ISCO-08
2-digi occupa ions, he s anda d de ia ion o ou ine ask con en a he 4-digi le el wi hin ISCO-08 2-digi occupa ions.
Coun y and yea ixed e ec s a e also included. Fo u he de ails on he cons uc ion o he con ol a iables e e o
appendix B.1. Own elabo a ion based on he i e wa es o he EU-SES.
51
Table C.4: Dependen a iable: Log o eal hou ly wage gap. Con olling o labo ins i u-
ions
(1) (2) (3)
Wage gap p90/p10 Wage gap p90/p50 Wage gap p50/p10
Robo exposu e Webb -0.343∗∗∗ -0.179∗∗∗ -0.164∗∗∗
(0.025) (0.014) (0.016)
AI exposu e Webb 0.190∗∗∗ 0.129∗∗∗ 0.061∗∗
(0.057) (0.036) (0.031)
Cons an 0.956∗∗∗ 0.595∗∗∗ 0.361∗∗∗
(0.114) (0.069) (0.068)
Obse a ions 2109 2109 2109
R20.494 0.431 0.410
Coun y FE Yes Yes Yes
Yea FE Yes Yes Yes
Manu ac u ing sha e Yes Yes Yes
Sex Yes Yes Yes
Union Yes Yes Yes
N o Ocupp Yes Yes Yes
HHI High skill Yes Yes Yes
No es:∗∗∗ p < 0.01;∗∗ p < 0.05;∗p < 0.1. Robus s anda d e o s be ween pa en heses. Table 5.1 displays he esul s o
eg essing he log o he eal hou ly wage gap on he AI and obo exposu e sco e a he occupa ion-coun y-yea le el. The
AI and obo occupa ional sco es a e a ISCO-08 2-digi le el. They a e buil using he index p o ided by Webb (2020) a
he SOC-2010 and con e ed in o ISCO-08. Fo yea s p io o 2010, we mapped ISCO-88 o ISCO-08. The de ails o he
ha moniza ion can be consul ed in appendix A. Con ol a iables o each occupa ion-coun y-yea cell include: he sha e
o employmen in manu ac u ing, he sha e o emale employmen , he sha e o unionized wo ke s, and he HHI index o
high-skill employmen . Addi ional con ol a iables a he occupa ional le el include: he numbe o 4-digi occupa ions
wi hin each ISCO-08 2-digi occupa ion, he s anda d de ia ion o obo and AI exposu e a he 4-digi le el wi hin ISCO-08
2-digi occupa ions, he s anda d de ia ion o ou ine ask con en a he 4-digi le el wi hin ISCO-08 2-digi occupa ions.
Coun y and yea ixed e ec s a e also included. I also includes he log o employmen p o ec ion legisla ion index om
he OECD (indi idual and collec i e dismissals). Fo u he de ails on he cons uc ion o he con ol a iables e e o
appendix B.1. Own elabo a ion based on he i e wa es o he EU-SES.
C.2 Robus ness checks wi h al e na i e AI and obo exposu e measu es
Table C.5: Spea man co ela ions be ween AI and obo measu es a ISCO-08 4-digi le el
AI Webb Rob Webb AI Fel en e al AI P y ko a e al Engbe g e al Rob P y ko a e al Rob Mon obbio e al
AI Webb 1.00
Rob Webb 0.27∗∗∗ 1.00
AI Fel en e al 0.09 -0.68∗∗∗ 1.00
AI P y ko a e al 0.11∗0.07 0.10 1.00
Engbe g e al 0.02 -0.72∗∗∗ 0.93∗∗∗ 0.10∗1.00
Rob P y ko a e al 0.28∗∗∗ 0.34∗∗∗ -0.08 0.41∗∗∗ -0.18∗∗∗ 1.00
Rob Mon obbio e al 0.30∗∗∗ 0.51∗∗∗ -0.35∗∗∗ 0.19∗∗∗ -0.34∗∗∗ 0.48∗∗∗ 1.00
∗p < 0.05,∗∗ p < 0.01,∗∗∗ p < 0.001
No es: This able shows he Spea man co ela ion be ween he di e en al e na i es o he AI and obo sco es a ISCO-08 4-digi le el.
Co ela ions
52
Table C.6: Spea man co ela ions be ween AI and obo measu es a ISCO-08 3-digi le el
AI Webb Rob Webb AI Fel en e al AI P y ko a e al Engbe g e al Rob P y ko a e al Rob Mon obbio e al
AI Webb 1.00
Rob Webb 0.29∗∗ 1.00
AI Fel en e al -0.04 -0.80∗∗∗ 1.00
AI P y ko a e al 0.21∗0.06 0.11 1.00
Engbe g e al -0.12 -0.84∗∗∗ 0.94∗∗∗ 0.14 1.00
Rob P y ko a e al 0.43∗∗∗ 0.40∗∗∗ -0.17 0.49∗∗∗ -0.26∗∗ 1.00
Rob Mon obbio e al 0.40∗∗∗ 0.56∗∗∗ -0.39∗∗∗ 0.33∗∗∗ -0.41∗∗∗ 0.65∗∗∗ 1.00
∗p < 0.05,∗∗ p < 0.01,∗∗∗ p < 0.001
No es: This able shows he Spea man co ela ion be ween he di e en al e na i es o he AI and obo sco es a ISCO-08 3-digi le el.
Table C.7: Spea man co ela ions be ween AI and obo measu es a ISCO-08 2-digi le el
AI Webb Rob Webb AI Fel en e al AI OECD AI P y ko a e al Engbe g e al Rob P y ko a e al Rob Mon obbio e al
AI Webb 1.00
Rob Webb 0.28 1.00
AI Fel en e al -0.00 -0.85∗∗∗ 1.00
AI OECD 0.07 -0.87∗∗∗ 0.94∗∗∗ 1.00
AI P y ko a e al 0.22 -0.14 0.27 0.31 1.00
Engbe g e al -0.08 -0.87∗∗∗ 0.96∗∗∗ 0.93∗∗∗ 0.35∗1.00
Rob P y ko a e al 0.60∗∗∗ 0.37∗-0.11 -0.11 0.35∗-0.17 1.00
Rob Mon obbio e al 0.53∗∗∗ 0.62∗∗∗ -0.46∗∗ -0.50∗∗ 0.19 -0.44∗∗ 0.68∗∗∗ 1.00
∗p < 0.05,∗∗ p < 0.01,∗∗∗ p < 0.001
No es: This able shows he Spea man co ela ion be ween he di e en al e na i es o he AI and obo sco es a he ISCO-08 2-digi le el.
53