Chap e 6
Teaching AI e hics o ansla ion
s uden s
Joss Moo kensa& Gökhan Doğ ub
aDublin Ci y Uni e si y/ADAPT Cen e, I eland bUni e si a Pompeu Fab a,
Spain
A key pa o eaching ansla ion in he age o Gene a i e AI (GenAI) is unde -
s anding when o use and when no o use GenAI. AI e hics is an applied sub ield
o E hics ha can help wi h decision-making, which may be use ul o s uden s
a e hey g adua e and as hey mo e h ough hei ca ee s. In his chap e , we
begin wi h some de ini ions o AI and e hics, in oduce some e hical issues, along
wi h some p og ess in add essing hose issues. We inally conside some ways o
in oduce discussion and e lec ion in he class oom o maximise he impac o his
eaching, ocusing in pa icula on case s udies and he use o mapping and images
o unde s and ou posi ion wi hin he complex in e ela ionships be ween indi id-
uals, o ganisa ions, and socie y ega ding e hical issues pe aining o ansla ion
and GenAI.
1 In oduc ion
The opic o a i icial in elligence (AI) e hics is e y high on many agendas a
p esen , as uni e si ies and schools g apple wi h how bes o inco po a e Gen-
e a i e AI (GenAI) in o hei cu icula in a cons uc i e manne . The in en ion
is o each AI li e acy, de ined by Long & Mage ko (2020: 598) as he abili y o
c i ically e alua e, communica e and collabo a e wi h, and use AI, equipping s u-
den s wi h c i ical knowledge and skills ha will s and o hem h oughou hei
ca ee s, balanced wi h unde s anding o e hical issues ega ding he use o AI.
AI e hics a e impo an . The e a e lo s o e hical issues ega ding GenAI, as we
will un h ough in his chap e . Ou g adua es will gain esponsibili ies as hey
Joss Moo kens & Gökhan Doğ u. 2026. Teaching AI e hics o ansla ion s uden s.
In JC Pene , Joss Moo kens & Masa u Yamada (eds.), Teaching ansla ion in he age o
gene a i e AI: New pa adigm, new lea ning?, 105–122. Be lin: Language Science P ess.
DOI: 10.5281/zenodo.17641074
Joss Moo kens & Gökhan Doğ u
con inue h ough hei ca ee s, so i ’s impo an ha hey be conscious o he
e hical epe cussions o decisions and ac ions ha hey migh ake. The cu en
hype abou GenAI migh mean p essu e o use ools e en i he ci cums ances a e
no app op ia e – wha a blogpos by he language se ice p o ision company
T anspe ec called he “push o implemen ”, when companies a e “p essu ed by
leade ship o implemen AI solu ions” (T anspe ec 2024). I ’s impo an ha
g adua es a e able o make well- ounded a gumen s as o why using GenAI migh
be app op ia e o inapp op ia e in any gi en si ua ion.
In he class oom, ques ions o e hics end o p o oke discussion and allow
o s uden inpu . This chimes wi h he call o c i ical and democ a ised educa-
ion om au ho s such as F ei e, o Gi oux and McLa en, inco po a ing “ o ms
o lea ning which se e o p epa e s uden s o esponsible oles as ans o -
ma i e in ellec uals, as communi y membe s, and as c i ically ac i e ci izens”
(Gi oux & McLa en 1997: 236). Cu en ly, his call eels mo e u gen han e e .
In T ansla ion S udies, Abdallah (2011) p esen ed a ela ed h ee-s ep ‘ideology
c i ique model o eaching’ ha in ol es disassembly o he eache s’ own be-
lie s, c i iquing and esis ing un ai ness, and inally os e ing hope and encou -
aging agency. The aim o eaching abou e hics in he ansla ion class oom is o
s imula e c i ical hinking so ha s uden s ecognise e hical p oblems and can
esis un ai ness. Howe e , conside a ion o e hics may p esen s uden s wi h
oppo uni ies a he han only imposing limi a ions, and i is impo an o s u-
den s o g adua e wi h a posi i e sense o hei own agency in amelio a ing e h-
ical issues. Equally, as educa o s and s uden s, we should be awa e o and ake
esponsibili y o ou choices and decisions. Wi h ha in mind, his chap e bo -
ows B yan’s (2022) ‘pedagogy o he implica ed’ o desc ibe and unde s and ou
posi ion wi hin he complex in e ela ed sys ems linking us as indi iduals o ou
su ounding poli ical, cul u al, social and economic sys ems.
This chap e will begin wi h some de ini ions o e ms pe aining o AI and
e hics, mo ing on o explaining some o he e hical issues. We ha e ound i
use ul o c ea e model use cases o in-class discussion and will sugges a couple
o hese. The ea e we look a possible ou es o wa d ega ding GenAI and
e hics and he ecen p og ession on hese. Finally, we look a e hical issues
o esea che s in ansla ion and in e p e ing ha migh be ele an o bo h
educa o s and s uden s.
2 De ini ions
The e ms ‘a i icial’ and ‘machine in elligence’ we e i s popula ised in he
1950s, usually ega ding he abili y o machines o ake on human asks, bu
106
6 Teaching AI e hics o ansla ion s uden s
a e a ely clea ly de ined. Mo e ecen de ini ions o AI a e s ill some imes e-
la ed o hinking o beha ing like humans o hinking o beha ing in a a ional
way. Russell (1999: 12), o example, de ines AI as “ he capaci y o gene a e max-
imally success ul beha iou gi en he a ailable in o ma ion and compu a ional
esou ces”. Up un il he 1980s, hese esou ces migh ha e in ol ed se s o ules
buil by human expe s – so-called ‘expe sys ems’. This was he common pa a-
digm o ule-based machine ansla ion (MT), be o e he ad en o s a is ical MT
(B own e al. 1988).
Bo h ule-based and s a is ical MT used human- eadable da a o ules in wha
is some imes called Symbolic AI, bu once echnologies such as MT, speech
ecogni ion, and ecommenda ion sys ems mo ed o neu al ne wo ks o machine
lea ning, we mo ed o Subsymbolic AI, in which da a and p ocesses a e ans-
o med in o numbe s and no longe comp ehensible by humans. Mi chell (2020:
12) de ines subsymbolic AI as a “s ack o equa ions – a hicke o o en ha d- o-
in e p e ope a ions on numbe s”.
In ecen imes, AI is o en aken o mean machine lea ning using ans o me -
based neu al ne wo ks, despi e he e m’s ela i ely long his o y. Machine lea n-
ing is de ined by Kellehe e al. (2015: 3) as “an au oma ed p ocess ha ex ac s
pa e ns om da a”, and can be supe ised, whe eby he sys em lea ns how o
ca y ou a epea ed ask based on aining da a, as wi h NMT, o sel -supe ised
(i.e. by he sys em i sel ), whe e pa e ns a e in e ed om aining da a wi hou
explici ex e nal ins uc ion, as is he case wi h la ge language models (LLMs) –
GenAI based on na u al language p ocessing.
Cu en ly, much o he ocus o e y la ge (some imes called ‘ on ie ’) LLMs
is scale: bigge language models (LMs) wi h mo e da a, as a key inding o LLMs
was he eme gence o unexpec ed abili ies a scale (Wei e al. 2022). This does no
mean ha smalle LMs canno be use ul (see Moo kens e al. 2025 o mo e on
local, small LMs), bu he endency owa ds la ge LMs b ings pa icula e hical
conce ns ha we will e u n o in he ollowing sec ion.
So, wha do we mean by e hics? E hics is a b anch o philosophy ha helps us
o decide i an ac ion is igh o w ong (see Moo kens 2024). No ma i e e hics
desc ibes heo ies o help us wi h ha decision, such as consequen ialism o de-
on ology, and applied e hics in ol es he applica ion o hose heo ies o pa icu-
la scena ios, o example in business e hics, da a e hics, o AI e hics. The ield o
AI e hics looks o balance he bene i s o AI wi h nega i e e ec s (which we will
come o in Sec ion 3), o en by es ablishing p inciples o p o ide s o adhe e o
a hei disc e ion o by es ablishing go e nance ules wi hin an o ganisa ion o
in na ional (o in e na ional) law.
107
Joss Moo kens & Gökhan Doğ u
The e’s a g owing sense ha echnology a ec s ou abili y o li e a good li e.
Technologies ha e a o dances, ac ions ha a e made a ailable o us as use s,
and hese can encou age posi i e o nega i e, heal hy o unheal hy beha iou s.
Machine lea ning is enabled by powe ul compu e s, buil using a e me als and
elec ically powe ed, and also by huge amoun s o p op ie a y and/o public da a.
The e ms ‘c awling’, ‘ha es ing’ o ‘sc aping’ a e o en used o he collec ion
o da a ia he in e ne , as i da a is na u ally-occu ing a he han he p oduc
o human e o . So he e a e lo s o e hical issues pe aining o AI and da a ha
ela e o ansla ion. We’ll discuss many o hese in Sec ion 3.
Discussion poin : How migh ansla ion echnologies o GenAI help us o li e a
be e li e? And migh hey also ha e nega i e e ec s? I so, wha migh hese be?
3 E hical issues pe aining o GenAI and ansla ion
As no ed p e iously, he a ailabili y o as amoun s o da a acili a es GenAI.
T ansla o s who wo k wi h compu e -aided ansla ion ools a e usually ex-
pec ed o include bilingual ansla ion memo ies wi h deli e y o ansla ion
jobs, ei he due o con ac s o p eceden . This d aws om he Be ne Con en-
ion, i s enac ed in 1886, in which ansla ions a e conside ed o be de i a i e
wo ks ha “shall be p o ec ed as o iginal wo ks wi hou p ejudice o he copy-
igh in he o iginal wo k” (Wo ld In ellec ual P ope y O ganisa ion 1979: a .
2). The e ha e been a gumen s ha ansla o s may ha e easonable claims o
copy igh (see Moo kens & Lewis 2020), bu his is no usually espec ed, and
in many cases, digi al pla o ms do no pe mi ansla o s o e en access hese
memo ies, no o con ol how hey a e epu posed. MT and LLMs a e ained
using monolingual da a om o ganisa ions’ eposi o ies and da a ‘ha es ed’
om he in e ne using c awle s. The legal basis o his di e s in di e en ju-
isdic ions and many copy igh owne s o ex c awled o ha es ed ha e aken
legal ac ion. Fo now, he e does no appea o be any majo es ic ion o using
c awled da a, e en i i ’s copy igh ed, pe haps because au ho i ies do no wan
o be seen o s i le inno a ion and ew con en owne s eel able o challenge
powe ul echnology companies. Fo ansla o s and w i e s whose da a is used,
his p esen s a p oblem in ha hei wo k may be pu owa ds pu poses ha
hey do no ag ee wi h, wi hou any u he compensa ion.
Wo ke s in ansla ion and o he p o essions may choose o use GenAI o MT
(see Ri as Ginel & Moo kens 2024), bu many ha e hese echnologies imposed
on hem wi hou any choice. Fı a (2024) and o he s demons a e he ela ionship
be ween he use o echnology o educe labou cos s, pa icula ly wi hin digi al
108
6 Teaching AI e hics o ansla ion s uden s
pla o ms, and educed access o wha he In e na ional Labo O ganisa ion e m
‘decen wo k’, ela ing o pay, ep esen a ion, ime, s abili y, equali y and man-
agemen (Fe e o e al. 2015). Ruokonen & S ahn (2024) summa ise nume ous
s udies ha show a ela ionship be ween inc eased echnologisa ion and educed
job sa is ac ion and mo i a ion, which h ea en he sus ainabili y o he ansla-
ion indus y. Beyond ansla ion, eams o (ghos -)wo ke s, li le-documen ed
and o en wi h poo pay and condi ions, wo k o mi iga e biased and o ensi e
ou pu om GenAI as pa o he p ocess o ein o cemen lea ning using human
eedback (RLHF) o o mode a e po en ially o ensi e aining da a (Rowe 2023),
wi h associa ed epe cussions on wo ke s’ men al heal h.
Some yea s a e NMT became popula , esea che s such as Vanmassenho e
e al. (2018) ound pa e ns o bias in NMT ou pu s. This ou pu ended o be less
lexically di e se han human ansla ions and con ained gende bias, wi h ce ain
wo ds ending o be associa ed wi h males o emales. The e is also likely o be
a cul u al o linguis ic bias owa ds English, he language o mos aining da a
(Bende 2011, Moo kens e al. 2025). The une en suppo o languages in he dig-
i al domain is likely o exace ba e exis ing digi al di ides. Resea che s wo king
wi h LLMs also ound biased ou pu ega ding ace, gende , and sexuali y, and a
endency o o e ep esen hegemonic iews. This makes sense, conside ing ha
a lo o aining da a comes om he in e ne . In o de o ebalance his and o
p e en emba assing cul u ally inapp op ia e o o ensi e ou pu , de elope s o
LLMs began o use RLHF (Ouyang e al. 2022). Howe e , his in e en ion also
o e s LLM de elope s he oppo uni y o supp ess ou pu due o geopoli ical con-
ce ns. Acco ding o he New Yo k Times, he ou pu o he Chinese GenAI ool
Deepseek “will la gely e lec he wo ld iew o he Chinese Communis Pa y”
(Mye s 2025: B1) and US-based GenAI G ok censo ed “un la e ing ac s abou
P esiden Donald T ump” and i s owne , Elon Musk on i s ini ial launch (Wig-
ge s 2025), one o a s ing o con o e sies ha appea s o ela e o adjus men s
o he RLHF gua d ails behind he scenes. Coope a ion be ween big ech compa-
nies and au oc a ic egimes has un il now mos ly in ol ed emo ing con en o
sea ch esul s (see Boyle 2025), bu RLHF o e s ano he oppo uni y o in e ene
and manipula e ou pu based on poli ical demands.
T ansla ion and MT ha e long been closely ela ed o powe (see Tymoczko
2007 and Paullada 2020) wi h he pa on o ansla ion o MT esea ch ha ing
some sway in p oduc ion. The amoun o da a, money, and compu ing powe
equi ed o ain LLMs mean ha p oduc ion capaci y is in he hands o e y
ew weal hy o ganisa ions, wi h many p oduced by US big ech companies o
hei close ela ions. Use s a e u he elian on hese companies’ cloud se ices
o ain and ine- une, and la e deploy hei LLMs. Se ices may be changed o
109
Joss Moo kens & Gökhan Doğ u
wi hd awn wi h li le wa ning due o a change o ci cums ances (o geopoli ical
expediencies). A he ime o w i ing, he e is a push o LLM so e eign y, wi h
in e na ional o ganiza ions like he Eu opean Union, na ional go e nmen s and
some companies in es ing in supe compu e s and building hei own LLMs. O
cou se, no all LMs ha e o be huge, bu he ocus o US big ech p o ide s
con inues o be scale, c ea ing la ge and la ge models ha ake longe o ain,
wi h knock-on e ec s ega ding en i onmen al sus ainabili y.
The p e aining and deploymen o ounda ional LLMs is a esou ce-in ensi e
p ocess ha equi es signi ican capi al in es men o acqui ing he necessa y
ha dwa e, aining da a and human alen as well as ope a ing cos s, and each
la ge-scale p e- aining cycle has signi ican en i onmen al impac s due o high
elec ici y and wa e consump ion. Combined, hese ac o s allow only a ew
companies o ha e e y powe ul LLMs. Since many na ional go e nmen s, la ge
co po a ions and public o ganiza ions conside LLMs and AI in gene al as s a e-
gic enable s o hei u u e, hey a e inc easing hei in es men in his a ea as
well and se ing big a ge s o b oade AI use. Howe e , he impac o his AI
‘a ms ace’ on he en i onmen is likely o be ca as ophic on ou plane i e e y
en i y ies o de elop hei own p op ie a y LLM. In ecen yea s, AI sys ems
ha e in gene al become mo e e icien , bu he end o scale means quali y im-
p o emen is based on mo e da a and pa ame e s, equi ing mo e aining ime.
Luccioni e al. (2025) a gue ha inc eased e iciency in AI aining and in ime
sa ed by using AI will end o spu mo e use o AI, con inuing an upwa d end
no only in emissions, bu in wa e use o da a cen es, a e ea h me als used
o ICT, and mo e ha m ul elec onic was e.
These en i onmen al ha ms and he impac s di ec ly om AI a e di icul o
measu e exac ly. Emission le els may di e depending on whe he ene gy comes
om ossil uel o enewable sou ces (Sh e iono & Vanmassenho e 2022) o
depending on he ime o day and da a cen e loca ion (Dodge e al. 2022). Big
ech i ms a e looking o buy up enewables and nuclea powe sou ces as hey
come online, in o de o maximise hei ne -ze o c eden ials. Wa e oo p in s
a e also likely o di e depending on ime and egion (Li e al. 2025). P esen ly,
GenAI does no equi e a la ge p opo ion o esou ces, bu p ojec ions, such as
hose om he head o he UK Na ional G id p edic ing an AI-d i en six- old
inc ease in powe equi emen s o da a cen es in he nex decade (BBC 2024),
a e wo ying. Fo now, ene gy and wa e use a ibu able o GenAI a e e y small
in compa ison o he huge equi emen s o wa ching s eaming media o joining
a Zoom mee ing (My on e al. 2024).
110
6 Teaching AI e hics o ansla ion s uden s
4 Possible solu ions
Posi ioned wi hin he in e na ional AI ‘a ms ace’, wi h poli ically-mo i a ed
suppo o some de elope s leading o p e e ed companies ecei ing p e e en-
ial access o go e nmen s and cu ing-edge echnologies, he e a e also mo e-
men s o use ‘AI o Good’. One pla o m by ha name is led by he In e na ional
Telecommunica ion Union o he Uni ed Na ions (UN) o use AI o achie e UN
sus ainable de elopmen goals, and he e a e smalle ini ia i es such as he Dis-
ibu ed A i icial In elligence Resea ch Ins i u e ha looks o push back agains
he in luence o big ech on AI esea ch, de elopmen and deploymen . Many
people seek o use (Gen) AI o educe ha ms and inequali y. The e ha e been
p oposed uses o AI o ou e powe use o maximise he use o enewable ene gy
and o imp o e ene gy e iciency in he design, building, and use o comme cial
buildings (Ding e al. 2024).
Van Wynsbe ghe (2021) eels ha he e a e wo mo i a ions behind ‘AI o
sus ainabili y’ and ‘sus ainabili y o AI’ ha ough o be combined. The o me
seeks o do good, ye migh en ail nega i e en i onmen al impac s, whe eas he
la e acknowledges ha o AI o be sus ainable, he e needs o be lowe en i-
onmen al cos s o AI aining, uning and in e ence. She de ines sus ainable AI
as a necessa y mo emen o “ os e change in he en i e li ecycle o AI p oduc s
(i.e. idea gene a ion, aining, e- uning, implemen a ion, go e nance) owa ds
g ea e ecological in eg i y and social jus ice” (Van Wynsbe ghe 2021: 217).
Na ional and in e na ional legisla ion, mos no ably he EU AI Ac , seek o
limi ha m ul uses o AI. The AI Ac de ines a ypology o ie ed uses o AI based
on isk, wi h ‘unaccep able isk’ uses o bidden and high- isk uses, such as e-
c ui men decision-making and job alloca ion, subjec o special egula ion. This
ende s algo i hmic job alloca ion in he EU illegal, al hough i ’s likely ha he
ecommenda ion o an au oma ed p ojec managemen ool will s ill be ollowed.
In a posi ion pape by Moo kens e al. (2024) we bo owed he idea o a
iple bo om line om Business E hics (Elking on 1997) o p opose ha ans-
la ion echnologies and LLMs be e alua ed no jus ocusing on pe o mance,
bu a he gi ing equal weigh o people, plane and pe o mance. This is nec-
essa ily a heu is ic a he han an exac me ic, bu ollows on om c i icisms
o a ocus pu ely on pe o mance pushing AI de elopmen in he w ong di ec-
ion by Schwa z e al. (2020) and o he s. Fo people, we migh conside how
he LLM impac s anno a o s, ansla o s, pla o m wo ke s, and hose who ha e
been dispossessed o hei da a, balancing hese agains bene i s o people. Fo
he plane , we migh look a ene gy cos s/CO2, e icien models, and ICT cos
111
Joss Moo kens & Gökhan Doğ u
and disposal. Finally, o pe o mance, we should use ask-app op ia e and com-
p ehensi e s anda ds.
A p e ious sugges ion o ansla ion da a om Moo kens & Lewis (2019) was
a communi y-owned and managed digi al commons, ollowing he ideas o Os-
om (2011), wi h ie ed access a ailable o a cos . We a gued ha his would
help o “sus ain he occupa ion o ansla ion and o minimise he po en ial isks
and ha ms o ansla o s and he public” (Moo kens & Lewis 2019: 17). This idea
seems o be simila o he in ended implemen a ion o a Eu opean Language Da a
Space as a decen alised ma ke place o ex , ideo and audio da a in di e en
languages (Rehm e al. 2024).
The ad ice om some esea che s, such as Rudin (2019), is o en i ely a oid
black-box, subsymbolic sys ems o high- isks uses. This is because opaci y is
la gely baked in o subsymbolic AI sys ems, as desc ibed p e iously, al hough
Rudin and o he s belie e ha , in many cases, compa able esul s may be achie ed
wi h mo e anspa en , hyb id sys ems.
Fo e y la ge, closed sou ce sys ems, we do no know wha aining da a has
been used o wha he RLHF gua d ails a e. Howe e , no all LLMs a e closed
sou ce – o e en ha la ge. We men ioned he al e na i e op ions o small-scale
LMs in Sec ion 2. These can be use ul o na owly-de ined asks, wi h bene-
i s o low cos , low en i onmen al impac , and cus omisabili y. Fo echnically
con iden s uden s, guidelines o building a cus om small LM a e p o ided by
Moo kens e al. (2025).
In academic esea ch e hics, he c edo has mo ed on om ‘do no ha m’ o
he need o ac ually bene i esea ch pa icipan s. Rela edly, bes p ac ice o en-
gaged esea ch in ol es a pa icipa o y app oach, wo king coope a i ely, pa -
icula ly wi h ma ginalised g oups, as co-c ea o s o knowledge a he han im-
posing na a i es o pu ing wo ds in o hei mou hs. Bi hane e al. (2022) p o-
pose his app oach o building AI sys ems, o e ing examples o pa icipa o y
app oaches ha aim o lessen exis ing imbalances o powe . In his way, de elop-
e s “acknowledge ha he communi ies and publics beyond echnical designe s
ha e knowledge, expe ise and in e es s ha a e essen ial o he de elopmen o
AI ha aims o s eng hen jus ice and p ospe i y” (Bi hane e al. 2022: 7).
5 T ansla ion and AI e hics in he class oom
5.1 Class oom discussion ac i i ies
As s uden use s o ansla ion echnology – and mos likely use s o ela ed
AI ools and cloud se ices mo e b oadly – s uden s in a ansla ion class oom
112
6 Teaching AI e hics o ansla ion s uden s
a e al eady pa o he in e connec ed web o e hical issues om he p e ious
sec ion. They may no ha e gi en hem much hough , as so much o he hype
abou GenAI ocuses on i s posi i e po en ial a he han e hical issues. GenAI
is also en angled wi h wha B and & Wissen (2021) call he ‘impe ial mode o
li ing’, h ough which public and p i a e o ganisa ional s a egies and indi idual
li es yles and p ac ices in he Global No h ely on he unlimi ed app op ia ion
o esou ces, a disp opo iona e claim o global and local ecosys ems, and cheap
labou , ideally om dis an loca ions. B yan (2022: 330) concep ualises indi idual
ela ions wi h he clima e c isis as a “ o m o ‘di icul knowledge’, pa icula ly as
i ela es o lea ne s’ sel -implica ion in he condi ions ha a e being add essed”.
We can easonably b oaden his o AI e hics o ainee ansla o s and educa o s.
T ansmission-based lec u es alone seem inapp op ia e o his opic, as di icul
knowledge may aise sensi i i ies; nobody likes o be hec o ed o o eel ha
hei pe sonal e hics a e in ques ion.
In his sec ion, we in oduce wo me hods o s imula ing discussion and e lec-
ion abou ansla ion and AI e hics in he class oom. The i s uses scena ios o
case s udies, placing e hical dilemmas in o amilia con ex s o discussion. The
second d aws om B yan’s (2022) ‘pedagogy o he implica ed’, which seeks o
p omp c i ical e lec ion abou ou own posi ioning as ‘implica ed subjec s’ and
o os e agency o change. B yan uses he no ion o he implica ed subjec om
Ro hbe g (2019) o look beyond dicho omies o indi idual e sus ins i u ional e-
sponsibili y o injus ices o a discussion o how we a e enmeshed wi h sys ems
in many ways ac oss his o ical (diach onic) and con empo a y social-s uc u e
(synch onic) lines.
5.2 Case s udies
Acco ding o Benbunan-Fich (1998), a combina ion o lec u es and discussion
a e complemen a y ways o in oducing e hical issues in he class oom. To be-
gin wi h, lec u es abou “e hical concep s can lay he heo e ical ounda ions”
so ha s uden s can “p ac ice e hical analyses” he ea e using case s udies
(Benbunan-Fich 1998: 20). Case s udies ha e p o ed o be a use ul ool, pa ic-
ula ly in business schools, o many yea s. Acco ding o Ba nes e al. (1994),
case s udies ex end lea ning beyond each class, s imula ing deepe insigh s ha
link ac oss classes and modules. Led by ins uc o s wi h app op ia e case s ud-
ies, s uden s will engage and can de elop and a icula e c i ical insigh s. These
come h ough ou pa icula ac o s: si ua ional analysis, ac i e s uden in ol e-
men , a non- adi ional ins uc o ole and he need o ela e analysis and ac ion.
Si ua ional analysis means ha e hical issues a e applied in si u a he han in
113
Joss Moo kens & Gökhan Doğ u
Long, Du i & B ian Mage ko. 2020. Wha is AI li e acy? Compe encies and design
conside a ions. In P oceedings o he 2020 CHI Con e ence on Human Fac o s in
Compu ing Sys ems (CHI ’20), 1–16. Honolulu, USA: Associa ion o Compu -
ing Machine y. DOI: 10.1145/3313831.3376727. h ps://doi.o g/10.1145/3313831.
3376727.
Luccioni, Alexand a Sasha, Emma S ubell & Ka e C aw o d. 2025. F om e i-
ciency gains o ebound e ec s: The p oblem o Je ons’ pa adox in AI’s pola ized
en i onmen al deba e. a Xi :2501.16548 [cs]. DOI: 10.48550/a Xi .2501.16548.
h p://a xi .o g/abs/2501.16548.
Mi chell, Melanie. 2020. A i icial in elligence: A guide o hinking humans. Lon-
don, UK: Pelican.
Moo kens, Joss. 2022. E hics and machine ansla ion. In Do o hy Kenny (ed.),
Machine ansla ion o e e yone, 121–140. Be lin, Ge many: Language Science
P ess.
Moo kens, Joss. 2024. E hics. In ENTI (Encyclopedia o ansla ion and in e p e -
ing). G anada, Spain: AIETI. DOI: 10.5281/ZENODO.10940899. h ps://zenodo.
o g/doi/10.5281/zenodo.10940899.
Moo kens, Joss, Sheila Cas ilho, Fede ico Gaspa i, An onio To al & Maja Popo ić.
2024. P oposal o a iple bo om line o ansla ion au oma ion and sus ain-
abili y: An edi o ial posi ion pape . Jou nal o Specialised T ansla ion (41). 2–
25. DOI: 10.26034/cm.jos ans.2024.4706.
Moo kens, Joss & Da e Lewis. 2019. Resea ch ques ions and a p oposal o he
u u e go e nance o ansla ion da a. Jou nal o Specialised T ansla ion (32).
2–25.
Moo kens, Joss & Da e Lewis. 2020. Copy igh and he e-use o ansla ion
as da a. In Minako O’Hagan (ed.), The Rou ledge handbook o ansla ion and
echnology, 469–481. London, UK: Rou ledge. DOI: 10.4324/9781315311258.
Moo kens, Joss, Andy Way & Séamus Lank o d. 2025. Au oma ing ansla ion.
London, UK: Rou ledge.
Mye s, S e en Lee. 2025. P opaganda inds home on DeepSeek. New Yo k Times.
B1. h ps://www.ny imes.com/2025/01/31/ echnology/deepseek- chinese-
p opaganda.h ml.
My on, Da id, Dag Lundén & Jens Malmodin. 2024. Ne wo k ene gy use no
di ec ly p opo ional o da a olume: The powe model app oach o mo e e-
liable ne wo k ene gy consump ion calcula ions. Jou nal o Indus ial Ecology
28(4). 966–980. DOI: 10.1111/jiec.13512. h ps://onlinelib a y.wiley.com/doi/10.
1111/jiec.13512.
120
6 Teaching AI e hics o ansla ion s uden s
Na a o, Jessica L. & Jona han R. H. Tudge. 2023. Technologizing b on enb en-
ne : Neo-ecological heo y. Cu en psychology 42(3). 19338–19354. DOI: 10.
1007/s12144-022-02738-3.
Os om, Elino . 2011. Backg ound on he ins i u ional analysis and de elopmen
amewo k. Policy S udies Jou nal 39(1). 7–27. DOI: 10.1111/j.1541-0072.2010.
00394.x.
Ouyang, Long, Je Wu, Xu Jiang, Diogo Almeida, Ca oll L. Wainw igh , Pamela
Mishkin, Chong Zhang, Sandhini Aga wal, Ka a ina Slama, Alex Ray, John
Schulman, Jacob Hil on, F ase Kel on, Luke Mille , Maddie Simens, Amanda
Askell, Pe e Welinde , Paul Ch is iano, Jan Leike & Ryan Lowe. 2022. T aining
language models o ollow ins uc ions wi h human eedback. h ps://a xi .o g/
abs/2203.02155.
Paullada, Amandalynne. 2020. How does machine ansla ion shi powe ? In
Resis ance AI wo kshop. Vancou e , Canada.
Rehm, Geo g, S elios Pipe idis, Khalid Chouk i, And ejs Vasiļje s, Ka in Ma hei-
necke, Vic o ia A anz, Ai a s Bē ziņš, Mil os Deligiannis, Dimi is Gala-
nis, Ma ia Giagkou, Ka e ina Gki zou, Dimi is Gkoumas, Annika G ü zne -
Zahn, A hanasia Kolo ou, Penny Lab opoulou, Andis Lagzdiņš, Elena Lei ne ,
Valé ie Mapelli, Hélène Mazo, Simon Os e mann, S e ania Racioppa, Mickaël
Rigaul & Leon Voukou is. 2024. Common Eu opean language da a space. In
Nicole a Calzola i, Min-Yen Kan, Vé onique Hos e, Alessand o Lenci, Sak i-
ani Sak i & Nianwen Xue (eds.), P oceedings o LREC-COLING 2024, 3579–3586.
Tu in, I aly: ELRA & ICCL. h ps://aclan hology.o g/2024.l ec-main.317/.
Ri as Ginel, Ma ía Isabel & Joss Moo kens. 2024. A yea o Cha GPT: T ansla-
o s’ a i udes and deg ee o adop ion. T adumà ica ecnologies de la aducció
(22). 258–275. DOI: 10.5565/ e / aduma ica.369. h ps:// e is es.uab.ca /
aduma ica/a icle/ iew/n22- i as-ginel-moo kens.
Ro hbe g, Michael. 2019. The implica ed subjec : Beyond ic ims and pe pe a o s
(Cul u al Memo y in he P esen ). S an o d, USA: S an o d Uni e si y P ess.
DOI: 10.1515/9781503609600.
Rowe, Niamh. 2023. ‘I ’s des oyed me comple ely’: Kenyan mode a o s dec y
oll o aining o AI models. The Gua dian. h ps://www. hegua dian.com/
echnology/2023/aug/02/ai-cha bo - aining-human- oll-con en -mode a o -
me a-openai.
Rudin, Cyn hia. 2019. S op explaining black box machine lea ning models o
high s akes decisions and use in e p e able models ins ead. Na u e Machine
In elligence (1). DOI: 10.1038/s42256-019-0048-x. h ps://www.na u e.com/
a icles/s42256-019-0048-x.
121
Joss Moo kens & Gökhan Doğ u
Ruokonen, Minna & Elin S ahn. 2024. Wha do we know abou ansla o s’ job
sa is ac ion? T ansla ion Spaces 13(1). 7–31. DOI: 10.1075/ s.23019. uo.
Russell, S ua . 1999. Ra ionali y and in elligence. In Founda ions o a ional
agency. Michael Woold idge & Anand Rao (eds.). Do d ech , Ne he lands:
Sp inge . 11–33. DOI: 10.1007/978-94-015-9204-8_2.
Schwa z, Roy, Jesse Dodge, Noah A. Smi h & O en E zioni. 2020. G een AI. Com-
munica ions o ACM 63(12). 54–63. DOI: 10.1145/3381831.
Sh e iono , Dimi a & E a Vanmassenho e. 2022. The ecological oo p in o neu-
al machine ansla ion sys ems. h ps://a xi .o g/abs/2202.02170.
T anspe ec . 2024. T ansla ion AI: hype s. impac . A T anspe ec Wo ld. h ps:
//www. anspe ec .com/blog/ ansla ion-ai-hype- s-impac .
Tymoczko, Ma ia. 2007. Enla ging ansla ion, empowe ing ansla o s. Manch-
es e , UK: S Je ome. DOI: 10.1057/9780230593442.
Van Wynsbe ghe, Aimee. 2021. Sus ainable AI: AI o sus ainabili y and he sus-
ainabili y o AI. AI and E hics 1(3). 213–218. DOI: 10.1007/s43681-021-00043-6.
h ps://link.sp inge .com/10.1007/s43681-021-00043-6.
Vanmassenho e, E a, Ch is ian Ha dmeie & Andy Way. 2018. Ge ing gende
igh in neu al machine ansla ion. In P oceedings o he 2018 Con e ence on
Empi ical Me hods in Na u al Language P ocessing, 3003–3008. B ussels, Bel-
gium: Associa ion o Compu a ional Linguis ics.
Wei, Jason, Yi Tay, Rishi Bommasani, Colin Ra el, Ba e Zoph, Sebas ian
Bo geaud, Dani Yoga ama, Maa en Bosma, Denny Zhou, Donald Me zle , Ed
H. Chi, Ta suno i Hashimo o, O iol Vinyals, Pe cy Liang, Je Dean & William
Fedus. 2022. Eme gen abili ies o la ge language models.DOI: 10.48550/a Xi .
2206.07682.
Wigge s, Kyle. 2025. G ok 3 appea s o ha e b ie ly censo ed un la e ing men-
ions o T ump and Musk. TechC unch. h ps:// echc unch.com/2025/02/23/
g ok-3-appea s- o-ha e-b ie ly-censo ed-un la e ing-men ions-o - ump-
and-musk/.
Wo ld In ellec ual P ope y O ganisa ion. 1979. Be ne con en ion o he p o ec-
ion o li e a y and a is ic wo ks (as amended on Sep embe 28, 1979). h ps:
//www.wipo.in / ea ies/en/ip/be ne/.
Zhou, Meng. 2022. Educa ing ansla ion e hics: A neu ocogni i e e hical
decision-making app oach. The In e p e e and T ansla o T aine 16(4). 391–
408. DOI: 10.1080/1750399X.2022.2030534.
122
