In e na ional Jou nal on Cloud Compu ing: Se ices and A chi ec u e (IJCCSA) Vol. 15, No. 1, Feb ua y 2025
DOI: 10.5121/ijccsa.2025.15101 1
HUMAN-AI COLLABORATION: BALANCING
AGENTIC AI AND AUTONOMY IN HYBRID SYSTEMS
Gau a Samdani 1, Ganesh iswana han 2 and Abi ami Dasu Jegadeesh 3
1 Depa men o Da a Science, Uni e si y o No h Ca olina, Cha lo e
2 In o ma ion Technology, 3530 Alis e A e conco d 28027
3 Depa men o In o ma ion Technology, Anna uni e si y, Chennai
ABSTRACT
In his a icle, he au ho explo es he ension be ween he human ac o and a i icial in elligence as a
symbiosis o wo e ec i e app oaches o sol ing mul i ace ed, ealis ic asks. Conside ing he p emises o
human-AI coope a ion, i iden i ies how combined s uc u es can imp o e hese p ocesses as decision
making, scalabili y and lexibili y in sphe es including heal hca e, au o anspo indus y as well as
educa ion.
The discussion combines heo ies and case s udies o explain how hyb id sys ems may e ain anspa en ,
ai , and e hical p ocedu es while achie ing ope a ional pe o mance. Bene icial samples include one
ocusing on de eloping possible issues wi h he implemen a ion o , o ins ance, human supe ision o AI
and he g ow h o AI decision making sel -go e nance, he p oblem o AI biases, and o he s pe aining o
d awbacks o o e -essen ializa ion o AI.
KEYWORDS
Human-AI collabo a ion, Hyb id sys ems, Agen ic AI and au onomy,E hical conside a ions,Decision-
making amewo ks
1. INTRODUCTION
Human-AI collabo a ion is a opic o inc easing in e es as socie y becomes mo e awa e o i s
po en ial. This phenomenon is d i en by inc eased in e es in he de elopmen o AI om bo h
indus y and academia, u he enabled by an ongoing inc ease in compu ing powe and s o age
capaci y. Due o he ongoing scaling o AI echnologies, hey a e also becoming mo e ele an
and c i ical in many aspec s o mode n socie y and human li e: AI sys ems can ou pe o m
humans in sol ing pa e n ecogni ion asks, help add ess la ge-scale op imiza ion p oblems, and
e en con ol complex sys ems. One speci ic aspec o AI echnology de elopmen is o pi o al
impo ance in his con ex . In ecen yea s, he e has been a push o he abandonmen o la ge-
scale “beau y pagean s,” owa ds ins ead using hyb id human-AI eams o he pe o mance o
complex cogni i e asks, as hese hyb ids a e o en epo ed as mo e e ec i e and e icien . A
he hea o hese hyb id sys ems is a seamless and con inuously econ igu ing in e ac ion
be ween agen ic AI and AI au onomy.
One is hus mo i a ed o s udy possible ways o designing hese hyb id sys ems such ha hey
can balance he symme y o bo h componen s o he sys em o a i e a e ec i e decision-
making. One essen ial equi emen is ha he adop ed o ganiza ion mus e lec and in eg a e he
s eng hs o he indi idual componen s in such a way ha he eme gen beha io is consis en
wi h collec i e goals. The e a e angible bene i s o collabo a ion be ween AI and humans, as AI
In e na ional Jou nal on Cloud Compu ing: Se ices and A chi ec u e (IJCCSA) Vol. 15, No. 1, Feb ua y 2025
2
sys ems can p o ide addi ional expe ise o humans, o e mo e comp ehensi e and as e
esponses o complex e en s, enable he e alua ion o a la ge se o decision op ions in
cons ained ime limi s, and p ocess and analyze mo e in o ma ion han a human ope a o can.
Howe e , o maximize he bene i s, he di e en bu complemen a y capabili ies o AI and
humans need o blend e ec i ely. This is a signi ican challenge o he collabo a ion o human
ope a o s wi h cu en and u u e AI ools, as i equi es a balance be ween cen alized decision-
making in complex mul i-agen sys ems and decen alized con ol, each able o ac independen ly
as needed in hei speci ic en i onmen .
1.1. Backg ound and Signi icance
Backg ound and Signi icance Recen echnological ad ances ha e esul ed in a a ie y o
collabo a i e sys ems ha d aw upon a mix o human and AI capabili ies o equilib a e o he
s eng hs o each cons i uen pa . These sys ems can be ound ac oss a a ie y o domains,
including heal hca e, educa ion, and he de elopmen o new echnologies. Fo example, doc o s
in hospi al se ings o en use AI o aid in clinical assessmen ; simila ly, ama eu and p o essional
a hle es wea senso s cap u ing a a ie y o muscle and body measu emen s, which hey e iew o
enhance hei o m. Technological changes ha e pu he specialis s in a a ied se o disciplines
who a e building hese sys ems in con e sa ion no only wi h humanis s and e hicis s in e es ed in
he psychological, economic, and philosophical impo o hese newly shaped physio-
compu a ional sys ems bu also egula o s and e hicis s con ibu ing o discussions o AI policy
and egula ion. Medical school deans may wonde whe he and how using AI o c ea e eedback
loops o e hei s uden s’ lea ning would shape educa ion in hei ins i u ions and wan o engage
educa ional heo is s in making hose decisions. Technologis s wo king o build hese educa ional
AI sys ems also equi e a a ied se o ools spanning ex book knowledge and con empo a y
esea ch in a eas such as de elopmen al psychology, human-compu e in e ac ion, and machine
lea ning esea ch abou human eedback echniques. Taken oge he , he expe iences in his
complex, highly in e disciplina y space demons a e he need o a de ailed unde s anding o how
enginee ing decisions lead o shi s in he dis ibu ional posi ion o agen ic AI ac oss human and
AI con ibu o s. In addi ion, i hese hyb id sys ems pe sis , he long- e m scalabili y o he
sys em h ough long-li ed powe -sha ing a angemen s be ween humans and AI emains o be
seen.
1.2. Resea ch Objec i es
This esea ch a emp s o in es iga e hyb id human-AI collabo a ions ha cap u e a numbe o
cha ac e is ics o bo h au onomous and agen ic AI ocused sys ems. The goals o his wo k a e
wo old: (1) Unde s anding: Wha de ines an e ec i e and success ul hyb id human-a i icial
in elligence in e ac ion, especially conside ing in e ac ions whe e AI and/o au oma ion may be
eshaping use s' decisions (including ac ions and/o hough p ocesses) as hey occu ? (2)
Op imiza ion: Fo hose conside ing building hyb id human-AI sys ems, wha is he bes p ac ice
o help ensu e hey a e op imized in e ms o hei in e ac ions and impac ? In addi ion o
p o iding heo e ical con ibu ions o cla i y a mo e in-dep h accoun o e ec i e human-AI
in e ac ions, i is sugges ed ha he empi ically de i ed concep ual amewo k will enable
echnological and so wa e enginee s, along wi h s akeholde s, o cul i a e and e ine new ools
o op imizing hyb id human-AI in e ac ions in he u u e.
Resea ch Objec i e 1: Unde s anding. The p ima y esea ch ques ion (RQ1) ha is guiding his
s udy is as ollows: Wha de ines an e ec i e and success ul hyb id human-a i icial in elligence
in e ac ion, especially hose in e ac ions whe e AI and/o au oma ion may be eshaping use s'
decisions (including ac ions and/o hough p ocesses as hey occu )? RQ1 seeks o disen angle
he ac o s ha con ibu e o he success ul in e play be ween human and machine, and especially
In e na ional Jou nal on Cloud Compu ing: Se ices and A chi ec u e (IJCCSA) Vol. 15, No. 1, Feb ua y 2025
3
he p e iously unexplo ed domain whe e AI may o may no be pe o ming asks chosen by he
human designe , bu impac ing he use o poin o sale a he ime o decision making. A hyb id
sys em in AI is likely o be mo e e ec i e i his in e ac ion is esponsi e o use s, da a, o
en i onmen al signals. Unde s anding he dimensions and indica o s o such e icacy can be e
in o m he design o u u e sys ems. To c ea e a mo e in o med and e ec i e design p ac ice, we
mus i s cap u e he exis ing beha io s and mechanisms ha esul in posi i e expe iences and
synch onized ac ions be ween humans and AI. To consolida e he insigh s obse ed, we sugges
s udying a eal-wo ld comme cial se ing o in es iga e which hyb id human-a i icial
in elligence sys ems a e mos success ul. Fo his, RQ2 was designed: How do hyb id sys ems
in e ac wi h people and how does i impac beha io ?
2. FOUNDATIONS OF HUMAN-AI COLLABORATION
Human-AI collabo a ion is based on he in e ac ion and collabo a ion be ween human use s and
AI sys ems. This ype o hyb id in e ac ion can be mo e con e sa ional o in ol e mo e explici
collabo a ion in decision-making, in e ac ion wi h eal-wo ld componen s, and so on, and
encompasses a wide ange o e ms, including mixed-ini ia i e sys ems, collabo a i e au onomy,
and supe iso y con ol o uninhabi ed au onomous sys ems. Au ho s may also discuss hese
concep s as si ua ed wi hin he la ge a ea o human- obo in e ac ion and human- obo
collabo a ion in he indus ial sense.
In his pape , we adop a b oad iew o hyb id in e ac ions. The balance o agen ic AI be ween
he human and machine depends undamen ally on he capabili ies o he wo sys ems and he
dis ibu ion o unc ions ha exis be ween hem. This balance can shi dynamically, as he
capabili ies o ei he he human o he machine o he en i onmen al ci cums ances change.
Human-AI collabo a ions will exis on a spec um o le els o au onomy, which may necessi a e
highe le els o o e sigh , decision au ho i y, and con ol, o may ea u e mo e independen and
unsupe ised con ibu ions by he AI. A he same ime, each hyb id sys em will ha e mul iple
in e ac i e goals, such as e ec i ely supe ising he AI, eaching he AI, co ec ing e o s, and
lea ning om he AI. Le ing he machine pe o m mo e unc ions may lead o inc eased
wo kload o a educed sense o agen ic AI. A he same ime, human-AI collabo a ion has he
po en ial o c ea e sys ems ha a e able o accomplish mo e complica ed goals han he human o
he AI could accomplish in isola ion. By ope a ing wi h in eg a ed easoning and pe cep ion and
ich, nuanced in e ac ion, hyb id sys ems can ake ad an age o he “bes o bo h wo lds.”
Success ul human-AI collabo a ions b ing syne gis ic esul s, adap ing g ace ully o sys ema ic
e o s and unan icipa ed changes in he en i onmen by le e aging he AI in new ways.
2.1. De ini ion and Scope o Human-AI Collabo a ion
Human-AI collabo a ion is a complex concep . A i s co e, i e e s o he in e play be ween
human agen s and AI sys ems ac oss a ious ac i i y con ex s. Laye s o he Human-AI
collabo a ion in e ac ion a e p esen in con ex s like manu ac u ing, heal hca e, banking, and
anspo a ion, whe e he collabo a o s a e expec ed o achie e complemen a y oles o sha ed
objec i es. Alongside he dimension o in e ac ion a e addi ional, ela ed aspec s such as he
manne in which collabo a ion is conc e ized. Fo example, in powe plan con ol ooms, he
in e ac ion wi h AI-complemen ed sys ems is a ma e o classical Human-Machine In e ac ion.
He e, especially well-s udied a e aspec s such as he “uphols e y” o he in e ace: how
in o ma ion is p esen ed in ways op imized o pe cep ion, h oughpu ( o he asks which a e
judged impo an ), and emo ional s ess. A basic a ibu e o he Human-AI collabo a ion,
“upda ing he same implici model o he con ex s o he in e ac ion, hei own oles, and he
means o a success ul in e ac ion” emains, howe e , ele an .
In e na ional Jou nal on Cloud Compu ing: Se ices and A chi ec u e (IJCCSA) Vol. 15, No. 1, Feb ua y 2025
4
The in ol emen o a human pa ne in a collabo a i e se ing is no me ely a deligh ul pai ing
o wo p e iously sepa a e abili ies. A bes , he human in he loop has skills, expe ience, o
knowledge o he si ua ion as well as he modes o in e ac ion which allow hem o o m an
e ec i e o ce mul iplie o he AI (o ice e sa). A wo s , he human will possess impe ec ,
pa ial, o ou -o -da e expe ise, lack ele an in e pe sonal skills (especially in social a eas), no
manage he in e pe sonal ela ionships, o no ha e sui able con e sa ional s a egies. In his
ega d, “collabo a ion” includes mo e han sha ing ep esen a ions o plans. I includes adjus ing
beha io , e o alloca ion, and changing p io i ies in he manne sugges ed by he o he pa ne in
he common goal. In si ua ions whe e AI in e en ions end o be cu ailed by e hical o cul u al
e alua ions, o example, ins uc ions om an AI o he use o o ce, a human ac o ’s
collabo a ion may esul in an ac ion ha is no op imal bu is e hically, mo ally, o legally
accep able. The dis ibu ed expe ise assump ion ha unde pins such collabo a ion uns up
agains cons ained AI eliabili y and use s ess in a numbe o well-documen ed ou lie cases.
Fu he , as we discuss in a la e summa y, collabo a ion can be conside ed, in pa , as a
mechanism o p omo e agen ic AI. In en i onmen s whe e AI au onomy is he no m, occasional
collabo a ion could os e , a he han deg ade, use accep ance o AI decisions by p o iding a
pe cep ion o con ol.
2.2. His o ical Con ex and E olu ion
The his o y o de eloping AI and inco po a ing i in o domains makes a ious o ms o human-
AI collabo a ion possible now. D awing a comp ehensi e his o y o AI is a beyond he scope o
his essay, and we e e he eade o se e al comp ehensi e his o ical accoun s o AI ha a e
inhe en ly connec ed o hese b oade s udies as well. No ably, AI has unde gone mul iple shi s
h oughou i s his o y and can now be seen o ha e gone h ough a ew di e en models o
collabo a ion om bo h concep ual and echnical pe spec i es. Gene al AI would en ail sys ems
ha could pe o m human-le el du ies and can be compa ed o sys ems de eloped oday ha a e
some imes designed o in e ac wi h many di e en capaci ies in a ious spaces. Du ing his ime,
in elligen agen s we e p oposed, which we e sys ems ha embodied AI echniques in so wa e
and ha dwa e sys ems, planning sys ems in less amilia en i onmen s, con ac ne s which we e
sys ems esponsible o con ac ing ou asks o a ange o compu e s, and expe sys ems which
unc ioned wi h a g ea deal o au onomy in sepa a e ields like ensu ing he quali y o pa s in
he ae ospace indus y. Mo eo e , agen s we e en isioned ha we e esponsible o checking
in o ma ion con lic s, while e y specialized, and he e we e also en isioned sys ems ha
pe o med a g ea deal o au onomous asks in modal logics, including many ma hema ical asks.
Ques ions abou wha au onomy could en ail ha e been connec ed o AI e hics and AI as well,
wi h he no ion ha humans may eel h ea ened by au onomous AI. I is a gued ha , while many
may belie e his has been a main ocus in de eloping AI, he ension his pape discusses is
"long- e m coope a ion among en i ies, some wi h e y di e en in e es s, while s ill con inuing
o ecognize each o he ’s au onomy." I is u he a gued ha his au opoie ic legal en i y could
also ex end his ecogni ion o au onomous AI in hei mids . The ques ion a ises, "Wha
algo i hm-based de elope s a e ying o p oduce non-au onomous AI, and o wha deg ee?"
Howe e , gi en he discussion om he his o y o AI, he need o au onomous AI o pe o m
complex decision-making seems ine i able. This poin is also e idenced by he unde s anding
ha au onomy includes he abili y o s op a p ocess. Thus, o ins ance, an au onomous ehicle
should ha e he abili y o b eak con ol o p e en ha m, which also aligns wi h he no ion o
mo al machines. The same ea u e is a necessa y quali y in he biologically inspi ed AI ha
b ings mo al and al uis ic decisions made by o he biologically inspi ed AI, agen s, so wa e
sys ems, o obo s.
In e na ional Jou nal on Cloud Compu ing: Se ices and A chi ec u e (IJCCSA) Vol. 15, No. 1, Feb ua y 2025
5
3. THEORETICAL FRAMEWORKS IN HUMAN-AI COLLABORATION
O e he pas decades, human-compu e in e ac ion (HCI) and a i icial in elligence (AI) esea ch
ha e signi ican ly ad anced how humans can collabo a e wi h in elligen so wa e agen s. Cen al
o his s and o esea ch is a p o ound heo e ical unde s anding o he in e play be ween human
and compu a ional beha io . This issue a ises, among o he s, om he combina ion o human and
AI ac i i y on di e en le els, om in en ion and ac ion all he way o pe cep ion, c ea ion, and
coo dina ion o meaning.
Key heo e ical amewo ks unde pin his s udy. The e is a b oad ange o schools o hough on
which heo e ical insigh s ela e o he in e sec ion o human and a i icial in elligence. Al hough
hese amewo ks a e b ough o bea by HCI and a i icial in elligence esea che s, he objec s o
hei inqui y may also p o ide insigh s ele an o da a-d i en economics. Agen ic AI is an
impo an concep when conside ing heo e ical unde pinnings o aspec s o con ol; ha is, how
humans in e ac in he de elopmen and a angemen o he 'supe imposi ion o con ol' o he
composi e sys em as i in e ac s wi h an AI componen o he same composi ion. The deside a um
in such an ac i e heo y o agen ic AI is o be able o unde s and and con ol he in e ac i e
p ope ies in such socio- echnical sys ems o cons uc a sa e and use ul sys em.
As AI echnologies become ac ualized in sys ems and in e ac wi h humans di ec ly, such
concep s become c i ical in hei ac ualiza ion as design cons ain s in p ac ice. Unde s anding AI
au onomy is impo an in i s implica ions o collabo a i e con ol and in luence, human-AI
ela ions and us , and p ojec ing wo k low, o example. "Wha kind o ac ion pa e ns can be
execu ed au oma ically?" Ideally, we will be able o condi ion mo e g anula ly he kinds o
communica i e exchanges and no ma i e social accoun abili y ha make sense in con ex by
con olling libe ally au onomous AI beha io . Socio echnical sys ems heo y in o ms HCI and AI
hyb id in e ac ion heo y mos ly indi ec ly by es ablishing he dynamics o human- echnology
in e ac ion and p ac ices bo h in o mally and ins i u ionally. Gi en i s s ong g ounding in and
eliance on sociological amewo ks, he heo y o socio echnical sys ems b ings o bea insigh
in o and ope a ionalizes he in es iga i e aspec s o he phenomenon. Humans in such sys ems,
howe e p imi i e o o he wise, a e oday 'wo king' wi h he sys em AI exac ly in his
phenomenological sense when hey a e p o ided wi h cha s o aining da a.
3.1. Agen ic AI and Au onomy in Hyb id Sys ems
I a obo is ega ded as a legal pe son a o ds legal igh s, hen who is o be blamed o he inju y
o loss occasioned by an independen decision by he obo ? Acco ding o au ho s, in hyb id
sys ems people emain ee-willed agen s while a i icial in elligence p o ides oppo uni ies o
machines’ lea ning, deciding, and p o-ac ing o ul ill hei goals. I is clea ha he ope a ional
dynamics o human and au onomy ease ou a dubious dis inc ion, hus equi ing a balance o he
bes ha mony.This can cause a lack o us ; disappoin men and ime was age in pe o mance o
o ganiza ional goals. Main aining his balance in ol es key ac o s such as:
Sys em T anspa ency: Clea AI decision-making p ocesses.
En i onmen al Adap abili y: Handling unce ain y in dynamic con ex s.
Human T us : Building eliabili y and con idence in AI.
Con ol Dynamics: The ha moniza ion be ween decision making s uc u es and
de ini ion o human in e en ion.
Time and Con ex : Rela ionship be ween he le el o decision-making AI au onomy and
decision u gency and complexi y.
In e na ional Jou nal on Cloud Compu ing: Se ices and A chi ec u e (IJCCSA) Vol. 15, No. 1, Feb ua y 2025
6
Concep ual model om disabili y s udies desc ibes he powe and con ol in human- Al
in e ac ions he e o e p o iding di ec ion on how o c ea e e ec i e ela ionships. In his case,
design p inciples o agen ic AI in hyb id sys ems ha e he po en ial o p omo ing us , enhanced
p oduc i i y oge he wi h socie al alignmen , i e hic conce ns a e well emb aced.
Table 1: Key Fac o s o Balancing Human and AI Au onomy in Hyb id Sys ems
Aspec
Desc ip ion
Sys em T anspa ency
The ea u e le s people unde s and exac ly how and why AI sys ems
p oduced hei ecommended decisions.
En i onmen al
Adap abili y
Enables AI o handle unce ain y and ope a e e ec i ely in dynamic and
unp edic able con ex s.
Human T us
This me hod ensu es use s us hei AI sys ems because hese sys ems
wo k well e e y ime.
Con ol Dynamics
The algo i hm le s o ganiza ions con ol decision-making p ocesses
h ough de ined s eps and human pa icipa ion ules.
Time and Con ex
The esea ch explo es how much eedom AI sys ems should ha e when
making essen ial and ad anced decisions.
Design P inciples
The app oach de elops sys ems o build us wi h s a and cus ome s
plus imp o e o ganiza ional pe o mance while add essing e hical
challenges.
3.2. Socio-Technical Sys ems Theo y
Theo y Socio-Technical Sys ems heo y is conce ned wi h he in e play be ween social and
echnical componen s in sys ems. This heo y a gues ha social and echnical dimensions
con inually need o in e ace wi h each o he o assu e e icien sys em ope a ion. Acco ding o
his heo y, sys ems ely on humans, as well as on echnological p ocesses and ools o s ee and
ope a e, and need o be concu en ly designed and delibe a ed acco dingly. This mu ual
adap abili y is desc ibed as s abili y. S abili y is achie ed when human ac o s adap , ha is,
modi y hei in en ions, ac ions, and cogni ion in ways ha a e compa ible wi h he p e-gi en
echnology. F om a socio- echnical sys ems pe spec i e, i is no only necessa y o unde s and
echnology o human-AI in e ac ion bu also embedded AI sys ems ha can deal wi h basic use
inpu and execu e he equi ed ac i i ies.
Resis ance can s em om anxie y owa ds new echnologies o he new asks and o ganiza ional
oles ha come wi h hem, bu h ough commi men , anspa ency, and ea ly in ol emen o
skilled human ope a o s in he design p ocess, he g a i y o his can be s eng hened. Fu he ,
e o s ocusing on wo ka ounds will d i e e o and awa eness on misalignmen s and poin ou
key issues. Ne e heless, sha ing au ho i y o e who is be e placed o make decisions in he
sys em, gi ing o ecei ing suppo , and miscommunica ions can esul om misconnec ions o
in en ions and expec a ions. In iew o his, alignmen e e s o adjus ing machine capabili ies o
be e ep esen human in en ions. Such adjus men s can s em om changes in knowledge and
goals, aining, o p oblem o mula ion. In es iga ions abou alignmen ake place in eal-wo ld
ask se ings. The socio- echnical sys ems pe spec i e an icipa es hese esis ances and shows us
he wide- anging domain o co ec i e s a egies on which o d aw. This is he so o p oblem we
plan o add ess in his pape .
4. CHALLENGES IN BALANCING AGENTIC AI AND AUTONOMY
Managing and alloca ing bo h he agen ic AI and he emaining human au onomy in a hyb id
sys em is a e y di icul hing. The isks such as inhe en au onomy, p i acy, as well as on da a
In e na ional Jou nal on Cloud Compu ing: Se ices and A chi ec u e (IJCCSA) Vol. 15, No. 1, Feb ua y 2025
7
sha ing need o be managed o ensu e ha e e y s ep mee s he s anda d ways o in o med
consen . In ligh o he issues p esen ed in his pape , his issue s ongly sugges s anspa ency in
AI design and he clea dema ca ion o esponsibili ies ele an o AI decision making.
The g ea es conce n is one o managing de eloping, complex a i icial in elligence echnologies.
Cu en laws canno always e ec i ely cope wi h he p oblem; hus, he need o lexible and
uni e sal laws adop ed wi h ega d o cul u al s anda ds and he opinion o expe s om o he
ields. I no well managed, he dispa i y o human decision-making con ol and a i icial
in elligence decision-making powe igge s skep icism and poo coo dina ion.
Sch oe e said ha s iking his balance also equi es knowledge o wha AI can do and canno do
and whe e humans come up sho . By inco po a ing e hical no ms in o he design o echnologies,
de eloping sound p inciples a he highes le el, and engaging in collabo a i e esea ch, hyb id
sys ems can deli e desi able in e ac ion wi h li ing beings a ge ing high au onomous and
agen ial alue.
4.1. E hical Conside a ions
While heo e ical o concep ual ep esen a ions o bo h ac i e agen ic AI and au onomous ac ion
a e alid, hey o en seem mu ually exclusi e in p ac ice. This leads o a numbe o e hical
conside a ions. F om a p i acy pe spec i e, he mo e con ol he AI sys em has o e i s decisions,
he mo e pe sonal in o ma ion and agen ic AI i has access o. Simila ly, he capabili y o make i s
own decisions also makes AI sys ems accoun able o hose decisions. Ensu ing AI sys ems a e
unawa e o people's sensi i e cha ac e is ics is undamen al o elimina ing bias. The mo e
eedom de elope s allow AI sys ems o enac hei own policies and decision-making
app oaches, he less con ol hey can ha e o e his a ibu e. While sys ems wi h dis o ed goals
a e no necessa ily une hical, i a use is ope a ing unde he assump ion o a di e en objec i e,
his has he po en ial o cause mo e undesi able ou comes han a misunde s anding o
unc ionali y alone. Finally, e hical guidelines demand ha AI sys ems mus be con ollable.
Wi hou su ende o a BCI i idly illumina ing an AI sys em's inpu s, he ques ion hen
becomes: o wha ex en should an AI sys em make i s decision-making p ocesses anspa en o
i s human collabo a o s? In a p ac ical sense, i is ex emely di icul o p e-de e mine he
dis ibu ion o skills. The p oblem o con lic ing alues, he e o e, seems o eme ge om
compe i ion a he han coope a ion. The capaci y o e hical ou comes can be impai ed o
imp o ed on wo le els: i s ly, by he ope a ing policy o he sys em, i s 'mo ali y'; and secondly,
by he undamen al goals o di ec i es gi en o he sys em du ing i s de elopmen . F om his
pe spec i e, i is possible o iden i y he po en ial o a spec um o unin en ional e hical
consequences. While sys ems designed o p o-social in en do no , by de aul , igno e he
eedom o human agen s, he e a e po en ial unconscious e hical con lic s a he poin o sys em
de elopmen o be conside ed. P ima ily, de elope s need o es ablish whe he e hics is a ele an
conside a ion in he design o hei sys em, and which e hical amewo k hey migh adhe e o o
adop . They mus also de e mine he ex en o which hey a e willing o comply wi h hese e hical
conside a ions and, i hey do, commi o subsequen e isions o hei sys em o adhe e o and
emain compa ible wi h an e hical s uc u e. Social implica ions and side e ec s mus be
iden i ied, and AI de elope s and o ganiza ions should design hei sys ems in a way ha e lec s
such conside a ions.
4.2. Legal Implica ions
The legal amewo k has o answe he ques ion o who has o bea he esponsibili y in cases
whe e a decision by he AI sys em unde lying he HAI may ha e nega i e consequences,
pa icula ly i hese nega i e consequences a e a esul o a di ision o labo be ween he human
In e na ional Jou nal on Cloud Compu ing: Se ices and A chi ec u e (IJCCSA) Vol. 15, No. 1, Feb ua y 2025
8
ope a o and he HAI. As AI sys ems a e based on p og amming p ocesses ha in mos pa s a e
beyond he indi idual’s con ol, he ques ion a ises whe he an AI sys em can c ea e a si ua ion
whe e nei he he pe son in he loop no he ope a o is esponsible, bu whe e he ou come is
pu ely bad luck. While legal easoning migh objec o he delimi a ion o esponsibili y in cases
whe e he p og amming p ocess is unknown o whe e un o eseeable e en s o in e ac ing sys ems
esul in unin en ionally ha ming somebody, he ex en o which he AI behind he HAI can lead
o ha m ul ou comes by ope a ing as in ended is s ill inadequa e. Dispu abili y o AI decision-
making is agg a a ed by he ac ha he de elopmen s age o AI sys ems is cha ac e ized a
leas by hei sel -lea ning algo i hms and high sensi i i y o inpu da a, meaning ha he ou come
o he decision-making p ocess could a y al hough he subsequen s eps aken by he machine
may be he same.
Fu he implemen a ion o Human-A i icial In elligence in e ac ion app oaches has o co e he
exis ing legal amewo ks. I da a is s o ed and p ocessed, he legal amewo k has o espec da a
p o ec ion egula ions, and i lea ned o in e ed knowledge has o be handled adequa ely wi hin
he Eu opean con ex , which espec s undamen al human igh s, pa icula ly p i acy. Despi e
some a he gene ic legal amewo ks, he le el o da a p o ec ion di e s widely be ween
coun ies. Bu no only do he igh s di e , he ex en as well as he adap a ion shape he exis ing
legal landscape and inally in luence scena io planning. Since AI echnologies and hei en i ies
a e assumed o be global and hei use is no limi ed o a ce ain s a e, he applica ion o laws
should end owa ds he highe s anda d and hei uni e sal p inciples. Ye , he ques ion o o
wha ex en , wi h he exis ing pa chwo k o laws uling AI sys ems on a na ional and in e na ional
le el, one may ely on p o ound, cons uc i e, and coope a i e legisla ion go e ning scena ios
has o be answe ed c i ically. I is necessa y o s eng hen dialogue be ween legal, echnical, and
e hical expe s o p epa e he g ound o co esponding legisla ion o come. Mo eo e ,
unce ain y abou how laws and igh s a e in e p e ed can con ibu e o a comp omised us in
AI-based collabo a i e sys ems. The lowe he us in sys ems, he mo e educed he goal-
o ien ed collabo a ion o humans wi h AI becomes.
5. BEST PRACTICES AND GUIDELINES
When designing a human-AI collabo a ion, i is impo an ha he de elopmen is guided by i s
use s' needs and he specialis asks hey seek o sol e. Gene al design p inciples should e lec
he pu pose o he hyb id sys em, whe he he aim is o inc ease he ole and decision-making o
p o essionals, educe documen a ion loads, wo k mo e e icien ly, o imp o e he pe o mance o
a p ocess o sys em. Mos o hese bes p ac ices skew owa ds open-loop AI sys ems, and
esea ch will be needed o unde s and how o balance hese p inciples when he use is pa o he
con ol loop and he ask is sha ed be ween human and AI.
To achie e in ui i e human-AI collabo a ion, i is impo an ha he use o bo h agen s is in ui i e
enough o he human end-use s, and ha he design o he AI agen acili a es he use o bo h
agen s oge he . Bo h con ol sys ems and humans and laye ed in e p e a ions o how o bes
engage and p oblem-sol e exis in di e en ia ed li e a u es. Bes -p ac ice social esea ch sugges s
ha enginee s ha e clea p inciples and human-cen e ed p ocesses, ac as acili a o s, ea lay
knowledge as a esou ce, and equi e clea go e nance, suppo , and aining. Ongoing educa ion
o de elope s and end-use s o AI sys ems is i al o coo dina ion and coope a ion wi h hyb id
human-AI sys ems. This includes aining ha is pa o embedded educa ion o p o essions and
con inual p o essional de elopmen and aining o p ac i ione s.
1) Use -Cen e ed Design: Sys ems whe e he human is pa o an AI con ol loop equi e speci ic
on -end use -cen e ed design o ensu e use s a e empowe ed o be able o mode a e sys em
decisions. 2) Human-AI Hyb id Sys ems T aining: Ongoing educa ional engagemen wi h g oups
In e na ional Jou nal on Cloud Compu ing: Se ices and A chi ec u e (IJCCSA) Vol. 15, No. 1, Feb ua y 2025
9
o many di e en p o essionals ocuses on he combined AI and human app oach o sol e
challenging eal-wo ld scena ios and dilemmas, and allows aine s o c i ically examine how
new AI echnology is used o in o m and/o make decisions. This mode o aining can be used a
any scale a which people a e using o designing sys ems ha comp ise AI-p ocessed da a o
ad ice. 3) F amewo ks o End-Use and Public Engagemen : An es ablished model exis s o how
o meaning ully engage p o essionals and o he s in codesign o da a-d i en AI ad ice sys ems.
The esul s o his engagemen include expe iences, knowledge, and opinions abou he
pe o mance o AI Face Decision Suppo in hei local ope a ional con ex s. Ini ial phases o
esea ch in ol ed engagemen wi h on -line medical, nu sing, and policy end-use s o cu en
ace decision suppo pa hways, as well as public g oups ha came in o con ac wi h digi al acial
analysis p oduc s. 4) Bes P ac ices o Technological O ganiza ions In ol ing he Use in Bes
P ac ice De elopmen : Bes p ac ices o including use s in he de elopmen o in o ma ion
echnology ha e been de eloped and unded. O ganiza ional pa icipan s in ol ed in codesign
app ecia e being pa o he codesign in e en ion and epo bene i s a all le els o he case
o ganiza ions, including pe sonal ca ee ad ancemen , imp o ed p ojec ou comes, and
o ganiza ional objec i es. 5) Human-Like AI Assis ance Sys ems: Ad ances in compu ing and AI
a e allowing o new o ms o collabo a ions whe e AI-human collabo a ion is inc easingly
blended in sha ed cogni ion models o hyb id sys ems. Human-in- he-Loop AI Sys ems: P o ides
an o e iew o he use o HIL AI sys ems o da e and hei signi icance. Desc ibes he po en ial
socie al, economic, cul u al, en i onmen al, and heal h bene i s o HIL AI sys ems, as well as any
po en ial ad e se o nega i e consequences o hese sys ems.
5.1. Design P inciples o Hyb id Sys ems
Hyb id sys ems blend human and a i icial in elligence. The e a e se e al undamen al design
p inciples c i ical o hyb id sys ems. Fi s , he sys em mus be designed conside ing he use
om he ou se o he p ocess; his is called human-cen e ed design o use -o ien ed design. This
in ol es use esea ch, unde s anding he use ’s capabili ies and limi a ions, i e a i e use es ing,
and he inclusion o he use expe ience and wide e hical issues wi hin he design p ocess.
Collabo a i e app oaches o en ex end human-cen e ed design o human-AI collabo a ion,
speci ically wi h a pa icula emphasis on ensu ing he human e ains con ol. Second, agen ic AI,
o con ol, o e AI sys ems is an impo an aspec o hyb id sys ems whe e human con ol is
sha ed wi h an AI sys em. A ange o design guidelines, ecommenda ions, and me hodologies a e
being de eloped o c ea ing collabo a i e hyb ids and o he kinds o hyb id sys ems. I e a i e
use es ing is equen ly men ioned in he design o hese sys ems, one o he p inciples used in
his wo k oo. Se ing clea objec i es and designing sys ems wi h ea u es ha ma ch he skills
and knowledge o he use ying o achie e hose objec i es is impo an o suppo he
de elopmen o human-AI collabo a ion. I makes use o human-human collabo a ion as an
analogue and a basis o ep esen and speci y he p inciples o human-AI collabo a ion. Al hough
he p inciples o HCI a e well de eloped, he human-AI aspec s a e s ill in de elopmen .
T anspa ency, in e ms o he ope a ion and oles o AI in he sys em, is seen o be an impo an
p inciple o collabo a i e hyb ids.
T anspa ency can also suppo he es ablishmen o empa hy, us , and social accep abili y; all o
which can be impo an in human-AI collabo a ion. In ela ion o u ili y, hyb ids mus sa is y
use s' needs and desi es. The p inciples o inclusi i y and di e si y in design a e impo an . The e
a e p inciples and guidelines wi hin HCI o guide his p ocess o use in e aces and o he
echnologies. The e is po en ial o ad e sa ial a acks, wi h humans s uggling o in e p e AI
ou pu s. Assis i e AI and obo s also need o be accessible o people wi h disabili ies. An
inclusi e design app oach o obo s was p oposed, and much o i may also apply o AI sys ems.
In ela ion o e hics, he p inciples o anspa ency ha e implica ions o p i acy, esponsibili y,
and accoun abili y. T anspa ency has e hical implica ions, as a lack o anspa ency can lead o a
