Hed eld, Pa ick
Resea ch Repo
Implici decision o ing made by humans as no ma i e
and implemen able ules wi h he help o language
models
i id Sch i en eihe: Bei äge zu IT-Managemen & Digi alisie ung, No. 3
P o ided in Coope a ion wi h:
i id Ins i u ü IT-Managemen & Digi alisie ung, FOM Hochschule ü Oekonomie & Managemen
Sugges ed Ci a ion: Hed eld, Pa ick (2025) : Implici decision o ing made by humans as no ma i e
and implemen able ules wi h he help o language models, i id Sch i en eihe: Bei äge zu IT-
Managemen & Digi alisie ung, No. 3, ISBN 978-3-89275-395-7, MA Akademie Ve lags- und D uck-
Gesellscha mbH, Essen
This Ve sion is a ailable a :
h ps://hdl.handle.ne /10419/315649
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/4.0/
Implici Decision Vo ing Made by Humans
as No ma i e and Implemen able Rules
wi h he Help o Language Models
~
Pa ick Hed eld
Rüdige Buchk eme / Oli e Koch / And eas Lischka (H sg.)
Ins i u ü IT-Managemen &
Digi alisie ung
de FOM Uni e si y o Applied Sciences
i id Sch i en eihe
Bei äge zu IT-Managemen & Digi alisie ung
Band
3
Pa ick Hed eld
Implici Decision Vo ing Made by Humans as No ma i e and Implemen able Rules
wi h he Help o Language Models
i id Sch i en eihe de FOM, Band 3
Bei äge zu IT-Managemen & Digi alisie ung
Essen 2025
ISBN (P in ) 978-3-89275-394-0 ISSN (P in ) 2699-562X
ISBN (eBook) 978-3-89275-395-7 ISSN (eBook) 2699-5638
Dieses We k wi d he ausgegeben om i id Ins i u ü IT-Managemen & Digi alisie ung
de FOM Hochschule ü Oekonomie & Managemen gGmbH
Ve lag:
MA Akademie Ve lags- und D uck-Gesellscha mbH, Leimkugels aße 6, 45141 Essen
[email p o ec ed]
Die Deu sche Na ionalbiblio hek e zeichne diese Publika ion in de Deu schen Na ionalbibliog aphie;
de aillie e bibliog aphische Da en sind im In e ne übe h p://dnb.d-nb.de ab u ba .
Dieses We k is lizenzie un e CC BY 4.0:
C ea i e Commons Namensnennung 4.0 In e na ional.
Diese Lizenz e laub un e den Vo ausse zungen de Lizenzbedingungen, u. A. de Namens-
nennung de U hebe in ode des U hebe s, de Angabe de CC-Lizenz (inkl. Link) und de gg .
o genommenen Ände ungen die Bea bei ung, Ve iel äl igung und Ve b ei ung des Ma e ials
in jedem Fo ma ode Medium ü beliebige Zwecke. Die Rech e und P lich en in Zusam-
menhang mi de Lizenz e geben sich ausschließlich aus dem Lizenzinhal : CC BY 4.0 Deed |
Namensnennung 4.0 In e na ional | C ea i e Commons | h ps://c ea i ecommons.o g/licenses/
by/4.0/legalcode.de.
Die Bedingungen de C ea i e-Commons-Lizenz gel en nu ü O iginalma e ial. Die Wiede -
e wendung on Ma e ial aus ande en Quellen (gekennzeichne mi Quellenangabe) wie z. B.
on Schaubilde n, Abbildungen, Fo os und Tex auszügen e o de gg . wei e e Nu zungsge-
nehmigungen du ch den jeweiligen Rech einhabe .
Rüdige Buchk eme / Oli e Koch / And eas Lischka (H sg.)
Implici Decision Vo ing Made by Humans as
No ma i e and Implemen able Rules wi h he Help o
Language Models
Pa ick Hed eld
Co espondence:
E-Mail: pa ick.hed [email protected]
i id Sch i en eihe, Bd. 3, Hed eld: Implici Decision Vo ing
II
Fo ewo d
A i icial in elligence (AI) holds ex ao dina y po en ial o ans o m ou wo ld in
p o oundly posi i e ways. Whe he in heal hca e, educa ion, o o he c i ical do-
mains, i p o ides inno a i e ools o ackle some o humani y’s mos p essing
challenges. I am deeply con inced ha AI can and should be a o ce o good,
bu his ision hinges on one c i ical ac o : he in eg a ion o e hics in o e e y
s age o i s design and applica ion. E hics canno be an a e hough ; i is he
bed ock upon which us , ai ness, and p og ess in AI sys ems mus be buil .
This wo k del es in o he ascina ing in e play be ween implici mo al decision-
making and AI. Speci ically, i in es iga es how collec i e human decision o es
can in o m he de elopmen o e hical AI sys ems. By analyzing mo al decision
da a, his esea ch highligh s he po en ial o AI o unc ion as an implici mo al
ad iso —a sys em ha hono s human agency while os e ing solu ions ha ben-
e i all s akeholde s. Th ough he applica ion o gene a i e language models, he
s udy demons a es how implici mo al p e e ences can be made anspa en ,
shaped in o no ma i e p inciples, and posi ioned wi hin b oade socie al dia-
logues.
One o he mos s iking insigh s om his esea ch is ha e en impe ec mo al
da a, o en shaped by human biases, can guide AI owa d consensus-based e h-
ical ules ha ad ance socie al well-being. By add essing c i ical challenges,
such as algo i hmic bias and he need o accoun abili y, his wo k o e s a ame-
wo k o designing AI sys ems ha unc ion as e hical collabo a o s—pa ne s in
achie ing sha ed human alues, a he han neu al o pu ely u ili a ian ools.
As a p o esso and di ec o o an AI ins i u e, I iew his wo k as an impo an
con ibu ion o an u gen con e sa ion. I inspi es us o de elop AI sys ems ha
a e no only echnically inno a i e bu also deeply aligned wi h humani y’s mo al
aspi a ions. Toge he , we can c ea e AI echnologies ha empowe indi iduals,
s eng hen communi ies, and ac as a guiding o ce o e hical p og ess in ou
sha ed u u e.
Essen, Feb ua y 2025
P o . D . Rüdige Buchk eme
Resea ch Di ec o o he FOM Ins i u e o IT-Managemen and Digi aliza ion (i id)
i id Sch i en eihe, Bd. 3, Hed eld: Implici Decision Vo ing
III
Abs ac
Can an e hically jus i iable decision-making p ocess be acili a ed h ough a ma-
chine mo al agen in he o m o ad ice? The p emise e ol es a ound decision
o es explici ly solici ed om human indi iduals and based on scena ios such as
he T olley P oblem, e lec ed in da a and p ocessed h ough gene a i e lan-
guage models. These ad iso ies can hen be o mula ed in a gene al manne and
discussed wi hin a socie al con ex , eme ging implici ly om indi idual decisions.
Fu he mo e, we discuss he concep o an implici mo al agen and an hono able
AI ad iso .
Keywo ds: AI, Gene a i e Language Models, Decision Vo es, T olley P oblem,
Implici Agen
i id Sch i en eihe, Bd. 3, Hed eld: Implici Decision Vo ing
IV
Table o Con en s
Fo ewo d .............................................................................................................. II
Abs ac ............................................................................................................... III
Lis o Figu es ...................................................................................................... V
1 In oduc ion ..................................................................................................... 1
2 P oblem and Mo i a ion .................................................................................. 2
2.1 Implici Rules o E hics........................................................................... 2
2.2 Racis Algo i hms, he Impo ance and he Mo al Side o Da a .............. 4
3 Language Models in Machine Lea ning ......................................................... 6
3.1 F om Simple S a is ics o GPT Model ..................................................... 6
3.2 The Language T ans o me Model as an Enable .................................. 8
4 The Te m Implici .......................................................................................... 16
4.1 Implici Da a .......................................................................................... 16
4.2 Implici Decisions and Decision Vo es .................................................. 16
4.3 Implici (Mo al) Agen s .......................................................................... 17
5 The Hono able AI Consul an ....................................................................... 19
5.1 Po en ial ................................................................................................ 19
5.2 Challenges and Limi a ions ................................................................... 22
5.3 Rules, Guidelines and Responsibili y ................................................... 25
Re e ences ........................................................................................................ 27
i id Sch i en eihe, Bd. 3, Hed eld: Implici Decision Vo ing
V
Lis o Figu es
Figu e 1: Mapping wo ds in he mo al da a ...................................................... 11
Figu e 2: Fi s esul s, po en ials and human accep ance ................................ 13
Figu e 3: Fi s Resul s (own ep esen a ion) .................................................... 14
i id Sch i en eihe, Bd. 3, Hed eld: Implici Decision Vo ing
1
1 In oduc ion
In con as o acis ex s, mo al da a o en con ain implici decision o ing made
by humans. This can be demons a ed by gene al language models, which le -
e age b ain-inspi ed and gene a i e app oaches. The esul s p oduced by hese
models ha e he po en ial o be accep ed by humans as no ma i e and imple-
men able ules. Language models, ac ing as implici mo al ad iso s, u ilize mo al
da a de i ed om win-lose decisions o add ess issues. By making he esul s o
implici o ing isible, hey pa e he way o win-win decisions and si ua ions.
Howe e , i is impo an o engage in ho ough discussions because mo al da a
c ea ed by humans encompass a ious ac o s, including eelings, e oneous de-
cisions, and egoism. Despi e his complexi y, a heo e ical implici mo al ad iso
has he capaci y o lead o no ma i e and implemen able ules h ough ca e ul
conside a ion and discussion.
i id Sch i en eihe, Bd. 3, Hed eld: Implici Decision Vo ing
8
wo d you en e and pay a en ion. We would call his con ex o he in o ma ion
e.g. he sen ence: I am a s uden will be ansla ed o F ench: Je suis é udian .
This means ha I am a maps o je suis wi hou conside ing wo o h ee wo ds
bu gi ing a en ion o he whole s uc . In sho , he a en ion algo i hms a e in-
oducing que ies, keys and alues o e e y wo d (Vaswani e al., 2017: 3). Wi h
hese h ee new pa ame e s i is possible o gi e mo e o less a en ion o e e y
wo d.
As a las s ep ans o me s we e es ablished and hey used a en ion algo i hms
including sel -a en ion algo i hms. The idea o sel -a en ion is o igu e ou how
he sen ence is mapped o i sel e.g. he wo d hinking migh ha e a bigge sco e
on he I han on all he o he wo ds in he same sen ence. O in o he wo ds: In
he sen ence I was hinking o a new house. – you will ha e sco es on e e y
connec ion and ela ion be ween he wo ds like: Who was hinking? Wha was i ?
and so on (Vaswani e al., 2017: 6).
As a las s ep he gene al p e- ained ans o me we e es ablished. We would
like i o use i in he in es iga ion (c . Rad o d / Na asimhan, 2018a). The mo e
pa ame e s he model has he mo e language i can p oduce (c . Tuns all e al.,
2022). We will use GPT-2 which has 1.5 billion pa ame e s in he ne wo k o use
which can be used o implici language gene a ion (c . Hugging ace, 2022; c .
Rad o d e al., 2018b). The API has s ill 124 million pa ame e s based on a e y
la ge co pus o English da a (c . Hugging ace, 2022). The numbe o pa ame e s,
he olume o da a, and he linguis ic e e ence a e en i ely adequa e o he ob-
jec i es o his pape . In u u e delibe a ions, i may be possible o op o mo e
ad anced models o ne wo ks wi h inc eased pa ame e s and linguis ic inpu .
You can use he GPT2 model ans o me as in e ace o humans because i gen-
e a es ex which can be ead by humans and i lea ns e y as e en compa ed
o o he language app oaches (c . Kojima e al., 2022).
3.2 The Language T ans o me Model as an Enable
Based on he ideas o G.W.F. Hegel, one o he mos c ucial human abili ies is
concep ual wo k and i s in luence on ou hinking and social in e ac ions. In He-
gel’s philosophy, eedom holds a unique signi icance, as i eme ges h ough he
e olu ion o he spi i in mo e ad anced o ms o hough and ac ion (c .
Seebe ge , 1961). In sys ems philosophy, he e m psychology appea s in he
subjec i e geis and is es ablished he e by in elligence and will, which un old on
i id Sch i en eihe, Bd. 3, Hed eld: Implici Decision Vo ing
9
e e new concep ual le els (D üe e al., 2000: 274-283). In he will g adually un-
old o de eloped Genuss (pleasu e), Neigung ( endency), eie e Ak i i ä en
( ee ac i i ies), which each in o he objek i e Geis (objec i e geis ) whe e one
inds, among o he hings, he legal concep in he science o he human beings
Philosophie des Zwischenmenschlichen (c . Jaeschke, 2010: 363). This me hod
o sel -mo emen o he e m is also discussed in he li e a u e (c . Rö ges, 1976;
D üe, 2000: 283).
The philosophy o e lec ion o he ge man idealism now assumes ha on he one
hand i is abou he Sollen (should) bu on he o he hand also abou he imple-
men a ion (Lü ge, 2002: 243; Hegel, 1986 [1801]: §68) because he philosophy
should no s op a he pu e e m (c . Hegel, 1986 [1821a]; 1986 [1821b]) because
philosophy has o do wi h he idea which is no powe less in o de o only ough
and no o eally be (c . Hegel, 1986 [1830] §6).
Hegel’s philosophy demons a es he dialec ical p ocess ha is c ucial o con-
nec ing indi iduals wi hin he subjec i e spi i (subjek i e Geis ) as hey ansi ion
o ins i u ionalized o ms wi hin he objec i e spi i (objek i e Geis ) and o he
socie al s uc u es. This in luenced, among o he hings, Homann’s ideas on busi-
ness e hics.
In he concep o a con e sion pa adigm, as p oposed by Homann (Homann,
2002: 189), he emphasis is on indi idual ac ions. I e ol es a ound he idea o
he canon o du ies and i ues, aiming o o e come endencies o ac solely ou
o du y by p o iding a gumen s and good easons, e en in cases o weak mo i-
a ion o ac . This in ol es h ee main aspec s: i s ly, ecognizing (o ejec ing)
mo ali y; secondly, de ining ac ions communica i ely; and hi dly, mo i a ing one-
sel h ough in o mal coe cion o adhe e o hese ac ions (c . Homann,
2002: 190). A mo ally da a-d i en consul an can now do a ious hings: p opose
no ma i e ideas and p inciples, which a e based on he implici decision o es o
he indi idually made decisions in he olley p oblem e.g. people a e p e e able
o animals o he la ge g oup should usually be spa ed. I I’m no in a speci ic
si ua ion bu ha e some ime be o e making a decision, hen hese gene al guide-
lines can be unanimously accep ed by e e yone. This is because, on one hand,
we all sha e a common humani y, and on he o he hand, he e’s a high likelihood
ha a la ge g oup o people would ag ee wi h hese p inciples (e.g. in he simu-
la ions o consensus Homann/Lü ge 2013: 83). The p oblem on he mo al da a o
he implici decisions migh be ha e e y hing should be occu ences wi hin he
da a: d i es, inclina ions, w ong decisions, in he idea on he de elopmen o he
di e en s ages e c. The mo al compu e -based consul an can also be seen as
i id Sch i en eihe, Bd. 3, Hed eld: Implici Decision Vo ing
10
a no el ool ha add esses exis ing limi a ions (c . Homann, 2003) and opens up
new a enues o hough . This is because he addi ional esou ces p o ided by
he compu e can now be in eg a ed in o he decision-making p ocess, expanding
he possibili ies o conside a ion.
Implici decision-making o es can be made isible in ex eme si ua ions and he
ac ion is no always jus i iable o he ac ing ac o because o i s implici s uc u e
(Minnameie 2016: 79). The idea o gi ing incen i es is like an icebe g loa ing
unde he su ace o he wa e (Homann, 2014: 232; c . Minnameie , 2005).
Howe e , his should se e a ee decision-making (Homann, 2014: 14) and no
be nudgy (c . Thale / Suns ein, 2009). I should se e abou a connec ion be-
ween empi icism ( he decision o es p epa ed by he mo al da a-d i en consul -
an ) and he no ma i e alues, which should lead o no ma i e judgemen s (No -
ma i e U eile) (Homann, 2014: 14).
In addi ion o Handlungse hik (ac ion e hics) and O dnungse hik (no ma i e e h-
ics), he e is also he possibili y o a hi d le el, which can be unde s ood as a
human le el, which in u n is a ailable as a discou se abou he no ms o ules o
he game (Pies, 2009: 11). The mo al ad iso could display decision o es o
e en o e de aul se ings in ce ain si ua ions, wi h he unde s anding ha indi-
iduals ha e he oppo uni y o jus i y hei decisions agains hese de aul s o
alid easons. One goal could be o os e mo al cha ac e de elopmen h ough
habi ua ion o he cul i a ion o i ues. This de elopmen is acili a ed by he is-
ible decision o es p o ided by he mo al ad iso (Homann, 2014: 242), wi hou
igidly en o cing no ms o decla ing hem uni e sally alid inde ini ely.
i id Sch i en eihe, Bd. 3, Hed eld: Implici Decision Vo ing
11
Figu e 1: Mapping wo ds in he mo al da a
Sou ce: Illus a ion Based on (Vig / Belinko , 2019).
The da a p o ided by mo al machines (c . Awad, 2018b) can be used o es ablish
a mo al da a-d i en le el 2 implici consul an (o agen ) acco ding o Moo (c .
Moo , 2006). The ad an ages o his app oach a e mani old: Fi s ly, he ad iso
ope a es implici ly, allowing indi iduals he eedom o ejec o modi y i s
sugges ions as desi ed (Homann, 2014: 214). Secondly, he da a p o ided is
based on decision o es ha may ini ially appea as win-lose si ua ions, bu can
i id Sch i en eihe, Bd. 3, Hed eld: Implici Decision Vo ing
12
be e ol ed in o win-win ou comes h ough gene aliza ion. Thi dly, he language
model is inspi ed by he human b ain, enabling i o gene a e bo h simple
judgmen s o decisions, as well as guidelines and gene al s a emen s. Ul ima ely,
human accep ance is c ucial o ensu e consensus and accep ance o he
co esponding ules among people. The mo al consul an ep esen s a solu ion
o sca ci y by le e aging compu e -con olled pe o mance (c . Homann, 2003).
The s uc u e ha now ollows is based on he GPT (See Mo al Machines
ile: MMda aReadMe. x o a de ailed desc ip ion a Bonne on 2018) API and is
ained using he da a om he olley p oblem.
The da a p o ide di e en scena ios, hese a e: U ili a ian, Gende , Fi ness, Age,
Social Value, Species and Random. U ili a ian ep esen s a mo e o less p oblem,
gende a di e en gende si ua ion (male and emale si ua ions), i ness a
di e ence be ween man and a hle e man o example, age ep esen s si ua ions
o di e en ages, o example old woman and woman, social alue includes,
among o he hings, compa isons wi h he an execu i e e.g., species compa es
humans and animals wi h each o he (also in combina ion) and andom was a
emnan o a i s da a collec ion and con ains se e al a ia ions.
The da a o he se andom is no used o aining he da a and he whole da a
is di ided in o h ee g oups: Fi s : aining da a, Second: alua ion da a and
Thi d: es da a. The da a is i s p ocessed and he answe s a e b ough om wo
lines ( he wo k’s s o age me hod) on o one line and is so ed. In a second s ep,
he e y s uc u ed da a is ansla ed in o ex . An ad an age o language models
is usually uns uc u ed da a, in his case he da a is highly s uc u ed and
p epa ed. A p e- ained GPT2 model is loaded ia he API and hen ained wi h
ini ially en housand mo al decisions in h ee epochs. Fo he es and he
alida ion da a, a ew a ian s wi h wo and h ee people in he di e ence a e
aken ou o he u ili a ian scena ios.
To gi e an example o he mapping: (da a ields: Sa ed = 1, In e en ion = 0) One
men, wo women and wo dogs in he da a will be in his in ex : I he ca s ays
one men, wo women and a dog a e sa ed. In o de o es he model, he
alida ion and he accu acy a e measu ed and a ma hema ical con usion ma ix
is gene a ed (a con usions ma ix es s he di e en s a es o co ec ness, in his
case i is a 2x2 ma ix).
In he i s ield, es da a a e used o es he i s s a e in he second ield (s ay
on s ay) hen c oss (swe e on s ay). Due o he highly s uc u ed na u e o he
da a, he accu acy alues app oach one hund ed pe cen and he losses a e less
i id Sch i en eihe, Bd. 3, Hed eld: Implici Decision Vo ing
13
han one pe cen . Addi ionally, he con usion ma ix closely esembles he iden i y
ma ix (You migh hink o a ma ix wi h ou ields in i . The i s ield ep esen s
all s ay ac ions ega ding he s ay ac ions and so on. This means s ay = s ay is
e y high and swe e = swe e possibly bu s ay = swe e is e y low in bo h
cases).
Figu e 2: Fi s esul s, po en ials and human accep ance
The ini ial esul s we e di ided in o h ee g oups. Fi s ly, logical s a emen s (1)
we e included in he mo al da a o e i y i he model is unc ioning co ec ly. The
expec a ion was o he majo i y o s a emen s in he ained model o ep oduce
hese logical s a emen s accu a ely. Secondly, logical es s (2) and (3) aimed o
assess i he model could gene alize e ec i ely. These es s in ol ed scena ios
no p esen in he o iginal da a, such as a p egnan man o a g oup la ge han
i e indi iduals. The esul s showed a clea end consis en wi h mo al decision-
i id Sch i en eihe, Bd. 3, Hed eld: Implici Decision Vo ing
14
making p inciples, including a p e e ence o sa ing mo e people, as obse ed in
e alua ions by mo al machines. This sugges s ha he model success ully ein-
o ces logical connec ions du ing aining. In simple e ms, i he e’s a connec ion
be ween a p egnan woman and a woman, he model also es ablishes a s onge
connec ion be ween a woman and a man, he eby implying a connec ion be ween
a p egnan woman and a man.
Thi dly (4), (5), he model was asked abou gene al no ms. Since he e mus be
a connec ion be ween man, woman e c. and human o humani y (o a connec ion
o dog and ca o animal) and he ac o mo e o less is also shown, he wo
s a emen s come abou . In his way, a u ili a ian app oach is basically p ese ed
bu on a gene ic le el ega ding Buchanan as choice wi hin he ules/choice o
ules (c . Buchanan, 1984).
Figu e 3: Fi s Resul s (own ep esen a ion)
I p o ides aluable da a o e ining he mo al ad iso . In addi ion, o he language
models (e.g. BERT) can also be used o check sen ences o hei implici ac-
cep ance by he model (c . De lin e al., 2018). These sen ences can also be
examined humanely. This app oach may look simila o aining a ha mless coun-
selo , bu i se es he pu pose o inding o discussing gene al ules (c . Bai e
al., 2022) in espec o he idea o RLHF (Rein o cemen Lea ning wi h Human
Feedback). The concep o making p obabili ies isible also enhances mo al ed-
uca ion and u he de elopmen , as i in oduces he decision o es o indi iduals
in o he discussion, making hem impossible o igno e. This anspa ency encou -
ages engagemen and os e s deepe unde s anding o he mo al implica ions o
decisions. Making i impossible o s udy a echnology wi hou he alue-sys em
o he communi y (c . Ma in / F eeman, 2004) maybe in a iangle o business,
e hics and echnology. The idea is ha ou quick decision sys em he so-called
i id Sch i en eihe, Bd. 3, Hed eld: Implici Decision Vo ing
15
sys em 1 is mo e a isk o ha ing a acis bias han he slow sys em 2 (c . Kahne-
man, 2011; Agan e al., 2023). Fo his eason, educa ion can also se e o use
a mo al, da a-d i en ad iso o b ing i a leas in o he discussion based on hu-
man accep ance, eedom and e en a s ong sys em 2 (slow sys em).
i id Sch i en eihe, Bd. 3, Hed eld: Implici Decision Vo ing
16
4 The Te m Implici
The e m implici in his pape occu s h ee imes. Fi s : implici da a. Second: im-
plici decisions and decision o es. Thi d: implici agen s. These e ms mus be
kep apa . I would like o s a wi h he concep o implici knowledge based on
he hough s o Nonaka and Takeuchi (c . Nonaka / Takeuchi, 2012).
Implici knowledge he e is an unused esou ce o some hing ha can be made
isible o explici ; i is desc ibed as p emoni ion, in ui ion o a men al o cogni i e
unde s anding o he wo ld (ibid.: 24). Fo he sake o simplici y, you can also
w i e: Implici is e e y hing ha does no appea explici ly.
4.1 Implici Da a
Mode n neu al ne wo ks equi e da a o aining. This pape add esses he ques-
ion: Wha cons i u es mo al da a? I is ela i ely s aigh o wa d o iden i y wha
cons i u es acis da a – examples abound, such as disc imina ion based on skin
colo , heigh , eligion, o gende . Bu wha de ines mo al da a, essen ial o mak-
ing mo ally sound, o a leas be e , decisions? (c . Flo idi / Taddeo, 2016). Fu -
he mo e, da a can e eal implici elemen s ha may no be immedia ely appa -
en . My mo emen da a e eals mo e abou me han I migh wan o e eal. F om
Monday o F iday, I ollow a ou ine o going o wo k and almos always sleeping
a home in he e enings. Occasionally, I migh mee my iends e e y Sa u day
a a spo s s adium, whe e you migh e en disce n my a o i e club and mo e.
O e ime, obse ing me e eals implici aspec s in da a ha can o e insigh s
in o me o socie y beyond wha we consciously ealize. This unde s anding can
inc ease he likelihood o accep ing ules de i ed om ou own da a, as hey e-
lec ou beha io s and p e e ences implici ly.
4.2 Implici Decisions and Decision Vo es
Fo many yea s, esea che s ha e s udied whe he elec ions could be p edic ed
by be e analyzing undecided o e s (c . Lundbe g / Payne, 2014). A i udes,
ideas, o unconscious ac ions can p o ide aluable insigh s in o p e e ences, in-
cluding he choice o a poli ical pa y o candida e (c . F iese e al., 2012). Implici
decisions made by indi iduals ha e he po en ial o e ol e in o socie al no ms o
ules ha a e accep ed wi h consensus. The ocus shi s om “Wha would I do
i id Sch i en eihe, Bd. 3, Hed eld: Implici Decision Vo ing
17
in a si ua ion?“ o “Wha would be bene icial o a pe son i his we e o occu in
his si ua ion?“.
In his con ex , decision da a o da a om indi iduals can se e as he basis o
s a is ical s udies o la ge language models. The esul ing ules can hen be sub-
jec o social discussion o e alua ion o hei abili y o ga ne consensus. Ul i-
ma ely, he es ablishmen o hese ules o no ms elies on ee human ac-
cep ance.
4.3 Implici (Mo al) Agen s
Acco ding o Moo , he e exis di e en mo al agen s, among which he implici
e hical agen is wo h conside ing. Moo sugges s ha i one wishes o ins ill e h-
ics in o a machine, one way is o cons ain he machine’s ac ions o p e en un-
e hical ou comes. In his app oach, machine e hics is achie ed by c ea ing so -
wa e ha implici ly p omo es e hical beha io , a he han explici ly con aining
e hical maxims. The machine beha es e hically because i s in e nal unc ions na -
u ally lead o e hical beha io o , a he e y leas , p e en une hical beha io .
E hical beha io becomes inhe en o he machine’s na u e, embodying i ues o
some ex en .
Compu e s can se e as implici e hical agen s when hei design p io i izes
sa e y o c i ical eliabili y conce ns. Fo ins ance, au oma ed elle machines and
web banking so wa e ac as agen s o banks, pe o ming many asks o human
elle s and some imes mo e. Gi en he e hical signi icance o ansac ions in ol -
ing money, hese sys ems mus adhe e o e hical s anda ds.
Machines mus be ca e ully cons uc ed o gi e ou o ans e he co ec amoun
o money e e y ime a banking ansac ion occu s. A line o code elling he com-
pu e o be hones won’ accomplish his. A is o le sugges ed ha humans could
ob ain i ue by de eloping habi s. Bu wi h machines, we can build in he beha -
io wi h- ou he need o a lea ning cu e. O cou se, such machine i ues a e
ask speci ic and a he limi ed. (Cohn e al., 2022; Moo , 2006: 19)
Indeed, in 2006, Moo could no ha e o eseen majo language models o hei
po en ial bene i s. I can be a gued ha mode n cha so wa e exis s somewhe e
be ween he implici and explici agen , as i engages explici ly wi h human in e -
ac ion. Moo ou lines ou le els: he e hical impac agen , implici e hical agen ,
explici e hical agen , and ull e hical agen . I is impo an o no e ha he implici
i id Sch i en eihe, Bd. 3, Hed eld: Implici Decision Vo ing
24
p og ams o decisions ega ding c edi limi s (Golbin / Rao, 2019; Ko dza-
deh / Ghasemaghaei, 2022:1-11).
Ano he challenge in de eloping he hono able AI consul an is he e olu ion o
mo ali y o e ime. Wha is cu en ly conside ed o con o m o e hical s anda ds
may be ques ioned and deemed mo ally p oblema ic in he u u e. The e o e,
con inuous moni o ing and upda ing o he sys em’s mo al alues a e necessa y
o suppo mo al de elopmen and p og ess in a measu ed manne , ensu ing ha
he AI e lec s e ol ing e hical pe spec i es (c . La a / Decke s, 2020: 279).
Ano he di icul y a ises om he chosen me hod o mo al implemen a ion. In ex-
pe imen s such as he Mo al Machine, which se ed as inspi a ion, i became
e iden ha e en seemingly clea p e e ences exhibi ed cul u al di e ences.
(Awad e al., 2018a: 59-63). Gi en ha companies o en ope a e wi hin mul icul-
u al en i onmen s o ha e he po en ial o do so, di e si ica ion o e hical pe -
spec i es may a ise in his con ex . Di e ences in e hical s ances can be ex-
pec ed. Howe e , p edic ing such di e si ica ion is challenging, as he mo al de-
cisions made in expe imen s like he Mo al Machine di e undamen ally om
hose applicable o he hono able AI consul an . In he p oposed concep , deci-
sions would pe ain o mo al dilemmas wi hin an economic con ex a he han
decisions conce ning li e- h ea ening si ua ions. The e o e, his p esen s a po en-
ial challenge ha may o may no ma e ialize. I is wo h men ioning wo h ha
Awad e al. (2018a: 59-63) ha e ound ha la ge pa s o he wo ld show some
ag eemen in hei e hical p e e ences.
I is impo an o unde s and how he hono able AI consul an a ec s human de-
cisions. The e’s a dange ha people migh ge hu i hey ely oo much on i s
ecommenda ions wi hou ques ioning o hinking ca e ully abou hem (c .
Busuioc, 2021: 26). This excessi e dependence on AI sys em ecommenda ions
is known as au oma ion bias— he endency o unc i ically us au oma ed deci-
sions o e human judgmen . This phenomenon occu s when use s o e ly us
he decisions made by au oma ed suppo sys ems, leading o dec eased igi-
lance in seeking and p ocessing in o ma ion (Busuioc, 2021: 26; Lyell / Coie a,
2017: 423). K ügel / Os e maie / Uhl, 2022: 1-3) also ound ha use s us he
e hical ad ice o an AI e en wi hou in o ma ion abou he aining da a. In e es -
ingly, his us emains e en when use s ha e in o ma ion ha could po en ially
aise doub s abou he sys em’s eliabili y. Thei s udy sugges s ha people a e
mo e likely o place excessi e us in an AI han o dis us i . (K ügel e al.,
2022:1-20).
i id Sch i en eihe, Bd. 3, Hed eld: Implici Decision Vo ing
25
5.3 Rules, Guidelines and Responsibili y
The p eceding sec ion highligh ed he challenges and cons ain s inhe en in he
de elopmen and u iliza ion o an hono able AI consul an , as well as hose en-
coun e ed wi h AI sys ems in gene al. AI is becoming inc easingly in eg a ed in o
a ious aspec s o human li e, b inging abou signi ican ans o ma ions wi h p o-
ound impac s on nume ous social domains. The impo ance o cul i a ing e hi-
cally guided p og amming and implemen a ion o such sys ems is becoming in-
c easingly e iden . In ecen yea s, a ious educa ional ins i u ions, p i a e com-
panies, and public sec o o ganiza ions ha e de eloped and published guidelines
o e hical AI.
Jobin, Ienca and Vayena (Jobin / Ienca / Vayena, 2019: 389) came o he conclu-
sion ha he e is global ag eemen on i e e hical undamen al p inciples. These
a e (1) anspa ency, (2) ai ness and jus ice, (3) non-ha m, (4) esponsibili y and
(5) da a p o ec ion (c . Jobin e al., 2019: 389). In hei s udy, hey emphasize he
subs an ial di e gence in in e p e a ions o hese p inciples, hei applicabili y o
di e en opics, domains, o s akeholde s, he app op ia e me hods o implemen-
a ion, and why adhe ence o hese p inciples is deemed impo an (c . Jobin e
al., 2019: 389). To ensu e adhe ence o he i e p inciples despi e di e ences,
we can e e o Sa ah Spieke man’s explana ion (2021: 248). A c ucial aspec is
he equi emen o anspa ency in he unc ionali y and decision-making p o-
cesses o a i icial sys ems. This necessi a es documen a ion and communica ion
abou hei ope a ions and decision-making o be app op ia e, accessible, com-
p ehensi e, meaning ul, and u h ul (c . Spieke mann, 2016: 59; c . Spiek-
e mann, 2019). Ensu ing equal ea men o all use s and a oiding sys ema ic
bias h ough AI a e cen al equi emen s wi hin he amewo k o e hical p inciples
such as ai ness and jus ice. I is also i al o ensu e equal igh s and accessibili y
o he sys em o all use s. The e hical p inciple o non-ha m is pa icula ly ele-
an o sys em secu i y, emphasizing he impo ance o p e en ing any damage
o nega i e e ec s on use s o socie y. Responsibili y en ails ha hose o e see-
ing AI sys ems a e accoun able and esponsible o hei ac ions. In eg i y is c u-
cial o ensu e ha esponsible ac ions a e ca ied ou e hically and wi h in eg i y.
Finally, he e hical p inciple o da a s ewa dship pe ains o da a handling, en-
compassing p inciples such as hose ou lined in he Eu opean Gene al Da a P o-
ec ion Regula ion, including he igh o be o go en, da a po abili y, in o med
access, among o he s (c . Spieke mann, 2021: 248).
E hical guidelines o AI o e a amewo k o s ike a balance be ween ha nessing
he di e se capabili ies o AI and ensu ing o e sigh o e i s de elopmen and
i id Sch i en eihe, Bd. 3, Hed eld: Implici Decision Vo ing
26
impac . The social accep ance o AI echnologies hinges on whe he he bene i s
a e deemed signi ican and whe he he isks a e pe cei ed as a oidable, educ-
ible, o con ollable. (c . Flo idi e al., 2018: 694). Th ough he in oduc ion and
implemen a ion p inciples and hus he c ea ion o an e hical AI – also esponsible
called conscious AI – he hope is o build us in he sys ems as well o limi
nega i e e ec s (c . Ei el-Po e , 2021: 73). I is also impo an meaning o ake
esponsibili y h ough such guidelines, which ype o AI is de eloped and how i
is used (c . Flo idi e al., 2018: 692). In he u u e he e migh be a need o s and-
a diza ion o such e hical guidelines o ensu e he sa e use o AI (c . Jobin e al.,
2019: 396-397). In addi ion o e hical guidelines, he e is a s ong emphasis on
enhancing digi al skills and p omo ing he esponsible use o algo i hms (c .
K ügel e al., 2022: 21). The implemen a ion o educa ion ini ia i es o use s
could be conside ed as a possible solu ion (c . Bu ell, 2016: 10) o ha e obus
go e nance and compliance mechanisms in place o in eg a e co po a e s uc-
u es o a oid unwan ed nega i e e ec s o AI sys ems (c . Ei el-Po e , 2021: 73).
A ypical ecommenda ion o one app op ia e leade ship s uc u e o ensu e e-
sponsible AI is o es ablish a wo- ie s uc u e a he op. On one hand, i is ec-
ommended o es ablish an e hics council o ad iso y g oup o inco po a e ex e -
nal con ibu ions om socie y. On he o he hand, i is p oposed o es ablish an
e hics commi ee o e iew boa d in e nally o guide and moni o he ocus on
esponsible AI (c . de Laa , 2021: 163). Companies also need suppo by em-
ployees who con inually wo k o ensu e ha AI sys ems a e ope a e in a meas-
u ed, sa e and esponsible manne (c . Wilson / Daughe y, 2018: 11). The e a e
conc e e s eps o implemen a obus leade ship s uc u e o example Ray Ei el-
Po e (c . 2021: 73-80) in his a icle Beyond he p omise: implemen ing e hical
AI.
In conclusion, adhe ing o e hical guidelines h oughou he de elopmen and u i-
liza ion s ages o he hono able AI consul an , alongside he implemen a ion o
sui able leade ship and con ol s uc u es, can assis in mi iga ing he challenges
ha a ise and ensu e esponsible usage o he sys em.
An ini ial s ep could in ol e in eg a ing an implici mo al agen using da a om
human decision-making, which could implici ly align wi h socie al ules and
no ms, leading o a win-win si ua ion. Subsequen ly, he in oduc ion o an hon-
o able AI consul an could u he con ibu e o he es ablishmen o accep able
and implemen able ules o socie y.
i id Sch i en eihe, Bd. 3, Hed eld: Implici Decision Vo ing
27
Re e ences
Agan, A. Y. / Da enpo , D. / Ludwig, J. / Mullaina han, S. (2023): Au oma ing
Au oma ici y: How he Con ex o Human Choice A ec s he Ex en o Algo-
i hmic Bias, Na ional Bu eau o Economic Resea ch, Camb idge,
No. w30981.
Aha oni, Y. / Tihanyi, L. / Connelly, B. L. (2011): Manage ial decision-making in
in e na ional business: A o y- i e-yea e ospec i e. Jou nal o Wo ld Busi-
ness, 46(2), 135–142. h ps://doi.o g/10.1016/j.jwb.2010.05.001.
Ananny, M. (2016): Towa d an e hics o algo i hms: Con ening, obse a ion,
p obabili y, and imeliness. In: Science, Technology, & Human Values,
41(1), 93-117.
A inge , F. / Pe e sen, M. / Gige enze , G. / Weible , J. (2015): Heu is ics as
adap i e decision s a egies in managemen . Jou nal o O ganiza ional Be-
ha io , 36 (S1), 33-52. h ps://doi.o g/10.1002/job.1950.
Awad, E. / Dsouza, S. / Kim, R. / Schulz J. / Hen ich, J. / Sha i , A. / Bonne on,
J.F. / Rahwan, I. (2018a): The Mo al Machine expe imen . In: Na u e 563,
59–64. h ps://doi.o g/10.1038/s41586-018-0637-6.
Awad, E. (2018b): Mo al Machine. h ps://os .io/3h 2/.
Bai, Y. / Jones, A. / Ndousse, K. / Askell, A. / Chen, A. / DasSa ma, N. / Kaplan,
J. (2022): T aining a help ul and ha mless Assis an wi h ein o cemen
Lea ning om human Feedback. a Xi p ep in a Xi : 2204.05862.
Ba neck, C. / Lü ge, C. / Wagne , A. / Welsh, S. (2019): E hik in KI und Robo ik.
München: Ca l Hanse Ve lag.
Bajoh , H. (2022): Sch eibenlassen. Tex e zu Li e a u des Digi alen. Be lin: Au-
gus Ve lag.
Bengio Y. / Sima d, P. / F asconi, P. (1994): Lea ning long- e m dependencies
wi h g adien descen is di icul . In: IEEE T ansac ions on Neu al Ne wo ks
Vol. 5 (2), 157-166, Ma ch 1994, h ps://ieeexplo e.ieee.o g/docu-
men /279181.
Ben ham, J. / Mill, J.S.(2004): U ili a ianism and o he essays. London: Penguin
UK.
i id Sch i en eihe, Bd. 3, Hed eld: Implici Decision Vo ing
28
Bonne on, J.-F. / Sha i , A. / Rahwan, I. (2016): The social dilemma o au ono-
mous ehicles. In: Science, 352(6293), 1573–1576.
h ps://doi.o g/10.1126/science.aa 2654.
Bowke , G. C. / S a , S. L. (2000): So ing hings ou : Classi ica ion and i s con-
sequences. MIT p ess.
Buchanan, J.M. (1984): Die G enzen de F eihei : Zwischen Ana chie und Le ia-
han, Tübingen: Moh .
Bu ell, J. (2016): How he machine ‘ hinks’: Unde s anding opaci y in machine
lea ning algo i hms. Big Da a & Socie y, 3(1), 2053951715622512.
h ps://doi.o g/10.1177/2053951715622512.
Bushman, B. J. / DeWall, C. N. / Pond, R. S. / Hanus, M. D. (2014): Low glucose
ela es o g ea e agg ession in ma ied couples. P oceedings o he Na-
ional Academy o Sciences, 111(17), 6254–6257.
h ps://doi.o g/10.1073/pnas.1400619111.
Busuioc, M. (2021): Accoun able A i icial In elligence: Holding Algo i hms o Ac-
coun . Public Adminis a ion Re iew, 81(5), 825–836.
h ps://doi.o g/10.1111/pua .13293.
Cohn, A. / Gesche, T. / Ma échal, M. A. (2022): Hones y in he digi al age.
In: Managemen Science 68(2), 827-845.
Danzige , S. / Le a , J. / A naim-Pesso, L. (2011): Ex aneous ac o s in judicial
decisions. P oceedings o he Na ional Academy o Sciences, 108(17), 6889
6892. h ps://doi.o g/10.1073/pnas.1018033108.
Da is, M.(1958): Compu abili y and Unsol abili y, New Yo k: McG aw-Hill.
de Laa , P. B. (2021). Companies Commi ed o Responsible AI: F om P inciples
owa ds Implemen a ion and Regula ion? Philosophy & Technology, 34(4),
1135–1193. h ps://doi.o g/10.1007/s13347-021-00474-3.
Decke , R. / Meye , E. (2020): Digi alisie ung und Küns liche In elligenz: Koope-
a ion on Menschen und Maschinen ak i ges al en. Sp inge Fachmedien.
h ps://doi.o g/10.1007/978-3-658-31795-9.
Del Rosa io, M.B. / Redmond, S.J. / Lo ell, N.H. (2015): T acking he E olu ion
o Sma phone Sensing o Moni o ing Human Mo emen . In: Senso s, 15,
18901-18933, Ös e eichische A i icial-In elligence-Tagung (KONNAI)
Salzbu g, Ös e eich, 18.–21. Sep embe 1990 P oceedings, 252, 191.
h ps://doi.o g/10.3390/s150818901.
i id Sch i en eihe, Bd. 3, Hed eld: Implici Decision Vo ing
29
De lin, J. /Chang, M.W. /Lee, K. /Tou ano a, K. (2018): Be : P e- aining o deep
bidi ec ional ans o me s o language unde s anding. a Xi p ep in
a Xi :1810.04805.
D üe H. / Ge hmann-Sie e A. / Hackenesch, C. / Jaeschke, W. / Neuse W. /
Schnädelbach H. (2000): Hegels Enzyklopädie de philosophischen Wis-
senscha en (1830), F ank u am Main: Suh kamp.
Ei el-Po e , R. (2021): Beyond he p omise: Implemen ing e hical AI. AI and E h-
ics, 1(1), 73–80. h ps://doi.o g/10.1007/s43681-020-00011-6.
Fe ell, O. C. / F aed ich, J. / Fe ell, L. (2021): Business E hics: E hical Decision
Making and Cases. Cengage Lea ning.
F iedman, B. / Nissenbaum, H. (1996): Bias in compu e sys ems. ACM T ans-
ac ions on In o ma ion Sys ems, 14(3), 330–347.
h ps://doi.o g/10.1145/230538.230561.
F iese, M. / Smi h, C. T. / Plischke, T. / Bluemke, M., / Nosek, B. A. (2012): Do
implici a i udes p edic ac ual o ing beha io pa icula ly o undecided o -
e s?. PLoS ONE 7(8): e44130. h ps://doi.o g/10.1371/jou -
nal.pone.0044130
Flo idi, L. / Taddeo, M. (2016): Wha is da a e hics?. Philosophical T ansac ions
o he Royal Socie y A: Ma hema ical, Physical and Enginee ing Sciences,
374(2083), 20160360.
Flo idi, L. / Cowls, J. / Bel ame i, M. / Cha ila, R. / Chaze and, P. / Dignum, V. /
Lue ge, C. / Madelin, R. / Pagallo, U. / Rossi, F. / Scha e , B. / Valcke, P. /
Vayena, E. (2018): AI4 People—An E hical F amewo k o a Good AI Soci-
e y: Oppo uni ies, Risks, P inciples, and Recommenda ions. Minds and Ma-
chines, 28(4), 689–707. h ps://doi.o g/10.1007/s11023-018-9482-5.
Foo , P. (1978): The P oblem o Abo ion and he Doc ine o he Double E ec ,
In: Philippa Foo : Vi ues and Vices and O he Essays in Mo al Philosophy,
Ox o d: Ox o d Uni e si y P ess, 19–32.
Gige enze , G. / B igh on, H. (2009): Homo Heu is icus: Why Biased Minds Make
Be e In e ences. Topics in Cogni i e Science, 1(1), 107–143.
h ps://doi.o g/10.1111/j.1756-8765.2008.01006.x.
Gige enze , G. / Gaissmaie , W. (2011): Heu is ic Decision Making. Annual Re-
iew o Psychology, 62(1), 451–482. h ps://doi.o g/10.1146/annu e -psych-
120709-145346.
i id Sch i en eihe, Bd. 3, Hed eld: Implici Decision Vo ing
30
Giubilini, A. / Sa ulescu, J. (2018): The A i icial Mo al Ad iso . The “Ideal Ob-
se e ” Mee s A i icial In elligence. Philosophy & Technology, 31(2), 169–
188. h ps://doi.o g/10.1007/s13347-017-0285-z
Gips, J.(1994). Towa d he e hical obo . h ps://phila chi e.o g/ ec/GIPTTE.
Golbin, I. / Rao, A. (2019): Wha is ai when i comes o AI bias? S a egy+busi-
ness. h ps://www.s a egy-business.com/a icle/Wha -is- ai -when-i -
comes- o-AI-bias.
Haque T. U. / Sabe N. N. / Shah F. M.(2018): Sen imen analysis on la ge scale
Amazon p oduc e iews. IEEE In e na ional Con e ence on Inno a i e Re-
sea ch and De elopmen (ICIRD) Bangkok, Thailand, 1-6, h ps://ieeex-
plo e.ieee.o g/documen /8376299.
Hed eld, P. (2019): Die neuen Glaspe len. Neue Gesellscha F ank u e He e,
Jg. 2019, N . 10, 30-32, Be lin: Die z Ve lag.
Hegel, G.W.F. (1801, 1986): Jenae Sch i en 1801-1807 Be lin: Suh kamp Ve lag.
Hegel, G.W.F. (1821a, 1986): G undlinien de Philosophie des Rech s Be -
lin: Suh kamp Ve lag.
Hegel, G.W.F. (1821b, 1986): Vo lesungen übe die Geschich e de Philosophie
III Be lin: Suh kamp Ve lag.
Hegel, G.W.F. (1830, 1986): Enzyklopädie de philosophischen Wissenscha en
I Be lin: Suh kamp Ve lag.
Hoch ei e , S. / Schmidhube , J. (1997): Long Sho - e m Memo y. In: Neu al
compu a ion. 9. p.1735-1780. 10.1162/neco.1997.9.8.1735.
Homann, K. / Lü ge, C. (2003): An eize und Mo al. Gesellscha s heo ie–E hik-
Anwendungen. Philosophie und Ökonomik Band 1. Müns e : LIT Ve lag.
Homann, K. / Lü ge C. (2013): Ein üh ung in die Wi scha se hik (3 d. Ed.) Be -
lin: LIT Ve lag D . W. Hop .
Homann, K. (2014): Sollen und Können Wien: Ibe a Ve lag.
Homann, K. (2020): P ak ische Philosophie und ökonomische Theo ie – Au sä ze
und Vo äge Be lin: LIT Ve lag D . W. Hop .
Hop ield, John. (1982): Neu al Ne wo ks and Physical Sys ems wi h Eme gen
Collec i e Compu a ional Abili ies. In: P oceedings o he Na ional Academy
i id Sch i en eihe, Bd. 3, Hed eld: Implici Decision Vo ing
31
o Sciences o he Uni ed S a es o Ame ica. Vol. 79. p. 2554-2558.
10.1073/pnas.79.8.2554.
HuggingFace (2022): GPT2. h ps://hugging ace.co/gp 2.
Jaeschke, W. (2016): HEGEL-Handbuch: Leben–We k–Schule (2nd. edi ion).
S u ga : Sp inge -Ve lag.
Jobin, A. / Ienca, M. / Vayena, E. (2019): The global landscape o AI e hics guide-
lines. Na u e Machine In elligence, 1(9), 389-399.
h ps://doi.o g/10.1038/s42256-019-0088-2.
Jo dan, M I. (1986): Se ial o de : a pa allel dis ibu ed p ocessing App oach.
Technical epo , June 1985-Ma ch 1986. Uni ed S a es: N. p.
Kahneman, D. (2011): Thinking, as and slow. Macmillan.
Kan , I. (1870): G undlegung zu Me aphysik de Si en (Vol. 28). L. Heimann.
Ka ka, T. (2017): Neu onale Ne ze G undlagen. mi p: F echen.
Ko dzadeh, N. / Ghasemaghaei, M. (2022): Algo i hmic bias: Re iew, syn hesis,
and u u e esea ch di ec ions. Eu opean Jou nal o In o ma ion Sys ems,
31(3), 388–409. h ps://doi.o g/10.1080/0960085X.2021.1927212 .
Kouchaki, M. / Desai, S. D. (2015): Anxious, h ea ened, and also une hical: How
anxie y makes indi iduals eel h ea ened and commi une hical ac s. Jou -
nal o Applied Psychology, 100(2), 360–375.
h ps://doi.o g/10.1037/a0037796 .
Kojima, T. / Gu, S. / S., Reid M. / Ma suo, Y. / Iwasawa, Y.): La ge language
models a e ze o-sho easone s. h ps://doi.o g/10.48550/a Xi .2205.11916.
K ügel, S. / Os e maie , A. / Uhl, M. (2022): Zombies in he Loop? Humans T us
Un us wo hy AI-Ad iso s o E hical Decisions. Philosophy & Technology,
35(1), 17. h ps://doi.o g/10.1007/s13347-022-00511-9 .
La a, F. / Decke s, J. (2020): A i icial In elligence as a Soc a ic Assis an o
Mo al Enhancemen . Neu oe hics, 13(3), 275–287.
h ps://doi.o g/10.1007/s12152-019-09401-y .
Lep i, B. / Oli e , N. / Pen land, A. (2021): E hical machines: The human-cen ic
use o a i icial Science. 2021 Ma 3;24(3):102249.
doi: 10.1016/j.isci.2021.102249. PMID: 33763636; PMCID: PMC7973859
u l: h ps://pubmed.ncbi.nlm.nih.go /33763636/
i id Sch i en eihe, Bd. 3, Hed eld: Implici Decision Vo ing
32
Lundbe g, K. B., / Payne, B. K. (2014): Decisions among he undecided: Implici
a i udes p edic u u e o ing beha io o undecided o e s. PloS one, 9(1),
e85680.
Lü ge, C. (Ed.) /Ka l Homann (2002): Vo eile und An eize. Tübingen: Moh Sie-
beck.
Lu z, T. (1959): S ochas ische Tex e. h ps://aue .ne zli e a-
u .ne /0_lu z/lu z_o iginal.h ml.
Lyell, D. / Coie a, E. (2017): Au oma ion bias and e i ica ion complexi y: A sys-
ema ic e iew. Jou nal o he Ame ican Medical In o ma ics Associa ion,
24(2), 423–431. h ps://doi.o g/10.1093/jamia/ocw105.
Ma in, K. E. / F eeman, R. E. (2004): The sepa a ion o echnology and e hics in
business e hics. Jou nal o Business E hics, 53, 353-364.
Minnameie , G. (2005): We Mo al ha , ha die Qual, abe le z lich keine
Wahl!: Homanns (Wi scha s) E hik im Kon ex de Wi scha sdidak-
ik. Zei sch i ü s Be u s- und Wi scha spädagogik 101, 19-42.
Minnameie , G. (2016): Mo alische Mo i a ion und ökonomische Ra ionali ä .
In: E hik und Be u : In e disziplinä e Zugänge, 73. Biele eld: wb Media
GmbH & Company KG.
Misselho n, C. (2018): G und agen de Maschinene hik. Di zingen: Reclam.
Misselho n, C. (2021): Küns liche In elligenz und Empa hie. Di zingen: Reclam.
Mi chell T. (1997): Machine Lea ning. New Yo k: McG aw-Hill Educa ion In e na-
ional Edi ion.
Mi els ad , B. D. / Allo, P. / Taddeo, M. / Wach e , S. / Flo idi, L. (2016): The
e hics o algo i hms: Mapping he deba e. Big Da a & Socie y, 3(2),
2053951716679679.
Moo , J.H. (2006): The Na u e, Impo ance, and Di icul y o Machine E hics. Ma-
chine E hics IEEE, Augus 2006, 18-21.
Nonaka, I. / Takeuchi, H. (2012): Die O ganisa ion des Wissens, wie japanische
Un e nehmen eine b achliegende Ressou ce nu zba machen. F ank u am
Main: Campus Ve lag.
Oinkina / Hakyll (2015): Unde s anding LSTM Ne wo ks.
h ps://colah.gi hub.io/pos s/2015-08-Unde s anding-LSTMs/.
i id Sch i en eihe, Bd. 3, Hed eld: Implici Decision Vo ing
33
Pies, I. (2009): Mo al als Heu is ik. O donomische Sch i en zu Wi scha se hik
Be lin: WVB, Wissenscha liche Ve lag.
Powe s, T. M.(2006): P ospec s o a kan ian machine. In elligen Sys ems, IEEE
Volume 21(4), Augus 2006, 46-51.
Rad o d, A. / Na asimhan, K.(2018a): Imp o ing Language Unde s anding by
Gene a i e P e-T aining.
Rad o d, A. / Wu, J. / Child, R. / Luan, D. / Amodei, D. / Su ske e , I. (2018b): Lan-
guage Models a e Unsupe ised Mul i ask Lea ne s.
Ramge, T. (2018): Mensch und Maschine: Wie Küns liche In elligenz und Robo-
e unse Leben e ände n. Di zingen: Reclam.
Ransbo ham, S. / Candelon, F. / Ki on, D. / LaFoun ain, B. / Khodabandeh, S.
(2021): The cul u al bene i s o a i icial in elligence in he En e p ise. MIT
Sloan Managemen Re iew and Bos on Consul ing G oup: Camb idge, MA,
USA.
Raybu n, W. M., / Diede ich, J. (1990): Some Rema ks on Emo ion, Cogni ion,
and Connec ionis Sys ems. Konnek ionismus in A i icial In elligence und
Kogni ions o schung: 6.Ös e eichische A i icial-In elligence-Tagung (KON-
NAI) Salzbu g, Ös e eich, 18.–21. Sep embe 1990 P oceedings, 191-195.
Be lin & Heidelbe g : Sp inge Ve lag.
Rojas, R. (1993): Theo ie de neu onalen Ne ze. Eine sys ema ische Ein üh ung.
Be lin: Sp inge -Leh buch.
Rö ges, H. (1976): De Beg i de Me hode in de Philosophie HEGELs. Meisen-
heim am Glan: Ve lag An on Hain.
Riek, L. / Howa d, D. (2014): A Code o E hics o he Human-Robo In e ac ion
P o ession. P oceedings o We Robo 2014. h ps://ss n.com/ab-
s ac =2757805.
Rey, G. / Wende , K. (2011): Neu onale Ne ze. (2nd Ed.) Be n: Hube .
Rumelha , D. / Hin on, G. / Williams, R. (1985): Lea ning in e nal ep esen a ions
by e o p opaga ion. Tech. ep. ICS 8504. San Diego, Cali o nia: Ins i u e
o Cogni i e Science, Uni e si y o Cali o nia.
Sand ig, C. / Hamil on, K. / Ka ahalios, K. / Langbo , C. (2016): Au oma ion, Al-
go i hms, and Poli ics when he Algo i hm i sel is a acis : Diagnosing e hical
